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
  45 enum task_position {
  46         TASK_EXTENTS,
  47         TASK_FREE_SPACE,
  48         TASK_FS_ROOTS,
  49         TASK_NOTHING, /* have to be the last element */
  50 };
  51
  52 struct task_ctx {
  53         int progress_enabled;
  54         enum task_position tp;
  55
  56         struct task_info *info;
  57 };
  58
  59 static u64 bytes_used = 0;
  60 static u64 total_csum_bytes = 0;
  61 static u64 total_btree_bytes = 0;
  62 static u64 total_fs_tree_bytes = 0;
  63 static u64 total_extent_tree_bytes = 0;
  64 static u64 btree_space_waste = 0;
  65 static u64 data_bytes_allocated = 0;
  66 static u64 data_bytes_referenced = 0;
  67 static int found_old_backref = 0;
  68 static LIST_HEAD(duplicate_extents);
  69 static LIST_HEAD(delete_items);
  70 static int no_holes = 0;
  71 static int init_extent_tree = 0;
  72 static int check_data_csum = 0;
  73 static struct btrfs_fs_info *global_info;
  74 static struct task_ctx ctx = { 0 };
  75 static struct cache_tree *roots_info_cache = NULL;
  76
  77 enum btrfs_check_mode {
  78         CHECK_MODE_ORIGINAL,
  79         CHECK_MODE_LOWMEM,
  80         CHECK_MODE_UNKNOWN,
  81         CHECK_MODE_DEFAULT = CHECK_MODE_ORIGINAL
  82 };
  83
  84 static enum btrfs_check_mode check_mode = CHECK_MODE_DEFAULT;
  85
  86 struct extent_backref {
  87         struct list_head list;
  88         unsigned int is_data:1;
  89         unsigned int found_extent_tree:1;
  90         unsigned int full_backref:1;
  91         unsigned int found_ref:1;
  92         unsigned int broken:1;
  93 };
  94
  95 static inline struct extent_backref* to_extent_backref(struct list_head *entry)
  96 {
  97         return list_entry(entry, struct extent_backref, list);
  98 }
  99
 100 struct data_backref {
 101         struct extent_backref node;
 102         union {
 103                 u64 parent;
 104                 u64 root;
 105         };
 106         u64 owner;
 107         u64 offset;
 108         u64 disk_bytenr;
 109         u64 bytes;
 110         u64 ram_bytes;
 111         u32 num_refs;
 112         u32 found_ref;
 113 };
 114
 115 static inline struct data_backref* to_data_backref(struct extent_backref *back)
 116 {
 117         return container_of(back, struct data_backref, node);
 118 }
 119
 120 /*
 121  * Much like data_backref, just removed the undetermined members
 122  * and change it to use list_head.
 123  * During extent scan, it is stored in root->orphan_data_extent.
 124  * During fs tree scan, it is then moved to inode_rec->orphan_data_extents.
 125  */
 126 struct orphan_data_extent {
 127         struct list_head list;
 128         u64 root;
 129         u64 objectid;
 130         u64 offset;
 131         u64 disk_bytenr;
 132         u64 disk_len;
 133 };
 134
 135 struct tree_backref {
 136         struct extent_backref node;
 137         union {
 138                 u64 parent;
 139                 u64 root;
 140         };
 141 };
 142
 143 static inline struct tree_backref* to_tree_backref(struct extent_backref *back)
 144 {
 145         return container_of(back, struct tree_backref, node);
 146 }
 147
 148 /* Explicit initialization for extent_record::flag_block_full_backref */
 149 enum { FLAG_UNSET = 2 };
 150
 151 struct extent_record {
 152         struct list_head backrefs;
 153         struct list_head dups;
 154         struct list_head list;
 155         struct cache_extent cache;
 156         struct btrfs_disk_key parent_key;
 157         u64 start;
 158         u64 max_size;
 159         u64 nr;
 160         u64 refs;
 161         u64 extent_item_refs;
 162         u64 generation;
 163         u64 parent_generation;
 164         u64 info_objectid;
 165         u32 num_duplicates;
 166         u8 info_level;
 167         unsigned int flag_block_full_backref:2;
 168         unsigned int found_rec:1;
 169         unsigned int content_checked:1;
 170         unsigned int owner_ref_checked:1;
 171         unsigned int is_root:1;
 172         unsigned int metadata:1;
 173         unsigned int bad_full_backref:1;
 174         unsigned int crossing_stripes:1;
 175         unsigned int wrong_chunk_type:1;
 176 };
 177
 178 static inline struct extent_record* to_extent_record(struct list_head *entry)
 179 {
 180         return container_of(entry, struct extent_record, list);
 181 }
 182
 183 struct inode_backref {
 184         struct list_head list;
 185         unsigned int found_dir_item:1;
 186         unsigned int found_dir_index:1;
 187         unsigned int found_inode_ref:1;
 188         u8 filetype;
 189         u8 ref_type;
 190         int errors;
 191         u64 dir;
 192         u64 index;
 193         u16 namelen;
 194         char name[0];
 195 };
 196
 197 static inline struct inode_backref* to_inode_backref(struct list_head *entry)
 198 {
 199         return list_entry(entry, struct inode_backref, list);
 200 }
 201
 202 struct root_item_record {
 203         struct list_head list;
 204         u64 objectid;
 205         u64 bytenr;
 206         u64 last_snapshot;
 207         u8 level;
 208         u8 drop_level;
 209         int level_size;
 210         struct btrfs_key drop_key;
 211 };
 212
 213 #define REF_ERR_NO_DIR_ITEM             (1 << 0)
 214 #define REF_ERR_NO_DIR_INDEX            (1 << 1)
 215 #define REF_ERR_NO_INODE_REF            (1 << 2)
 216 #define REF_ERR_DUP_DIR_ITEM            (1 << 3)
 217 #define REF_ERR_DUP_DIR_INDEX           (1 << 4)
 218 #define REF_ERR_DUP_INODE_REF           (1 << 5)
 219 #define REF_ERR_INDEX_UNMATCH           (1 << 6)
 220 #define REF_ERR_FILETYPE_UNMATCH        (1 << 7)
 221 #define REF_ERR_NAME_TOO_LONG           (1 << 8) // 100
 222 #define REF_ERR_NO_ROOT_REF             (1 << 9)
 223 #define REF_ERR_NO_ROOT_BACKREF         (1 << 10)
 224 #define REF_ERR_DUP_ROOT_REF            (1 << 11)
 225 #define REF_ERR_DUP_ROOT_BACKREF        (1 << 12)
 226
 227 struct file_extent_hole {
 228         struct rb_node node;
 229         u64 start;
 230         u64 len;
 231 };
 232
 233 struct inode_record {
 234         struct list_head backrefs;
 235         unsigned int checked:1;
 236         unsigned int merging:1;
 237         unsigned int found_inode_item:1;
 238         unsigned int found_dir_item:1;
 239         unsigned int found_file_extent:1;
 240         unsigned int found_csum_item:1;
 241         unsigned int some_csum_missing:1;
 242         unsigned int nodatasum:1;
 243         int errors;
 244
 245         u64 ino;
 246         u32 nlink;
 247         u32 imode;
 248         u64 isize;
 249         u64 nbytes;
 250
 251         u32 found_link;
 252         u64 found_size;
 253         u64 extent_start;
 254         u64 extent_end;
 255         struct rb_root holes;
 256         struct list_head orphan_extents;
 257
 258         u32 refs;
 259 };
 260
 261 #define I_ERR_NO_INODE_ITEM             (1 << 0)
 262 #define I_ERR_NO_ORPHAN_ITEM            (1 << 1)
 263 #define I_ERR_DUP_INODE_ITEM            (1 << 2)
 264 #define I_ERR_DUP_DIR_INDEX             (1 << 3)
 265 #define I_ERR_ODD_DIR_ITEM              (1 << 4)
 266 #define I_ERR_ODD_FILE_EXTENT           (1 << 5)
 267 #define I_ERR_BAD_FILE_EXTENT           (1 << 6)
 268 #define I_ERR_FILE_EXTENT_OVERLAP       (1 << 7)
 269 #define I_ERR_FILE_EXTENT_DISCOUNT      (1 << 8) // 100
 270 #define I_ERR_DIR_ISIZE_WRONG           (1 << 9)
 271 #define I_ERR_FILE_NBYTES_WRONG         (1 << 10) // 400
 272 #define I_ERR_ODD_CSUM_ITEM             (1 << 11)
 273 #define I_ERR_SOME_CSUM_MISSING         (1 << 12)
 274 #define I_ERR_LINK_COUNT_WRONG          (1 << 13)
 275 #define I_ERR_FILE_EXTENT_ORPHAN        (1 << 14)
 276
 277 struct root_backref {
 278         struct list_head list;
 279         unsigned int found_dir_item:1;
 280         unsigned int found_dir_index:1;
 281         unsigned int found_back_ref:1;
 282         unsigned int found_forward_ref:1;
 283         unsigned int reachable:1;
 284         int errors;
 285         u64 ref_root;
 286         u64 dir;
 287         u64 index;
 288         u16 namelen;
 289         char name[0];
 290 };
 291
 292 static inline struct root_backref* to_root_backref(struct list_head *entry)
 293 {
 294         return list_entry(entry, struct root_backref, list);
 295 }
 296
 297 struct root_record {
 298         struct list_head backrefs;
 299         struct cache_extent cache;
 300         unsigned int found_root_item:1;
 301         u64 objectid;
 302         u32 found_ref;
 303 };
 304
 305 struct ptr_node {
 306         struct cache_extent cache;
 307         void *data;
 308 };
 309
 310 struct shared_node {
 311         struct cache_extent cache;
 312         struct cache_tree root_cache;
 313         struct cache_tree inode_cache;
 314         struct inode_record *current;
 315         u32 refs;
 316 };
 317
 318 struct block_info {
 319         u64 start;
 320         u32 size;
 321 };
 322
 323 struct walk_control {
 324         struct cache_tree shared;
 325         struct shared_node *nodes[BTRFS_MAX_LEVEL];
 326         int active_node;
 327         int root_level;
 328 };
 329
 330 struct bad_item {
 331         struct btrfs_key key;
 332         u64 root_id;
 333         struct list_head list;
 334 };
 335
 336 struct extent_entry {
 337         u64 bytenr;
 338         u64 bytes;
 339         int count;
 340         int broken;
 341         struct list_head list;
 342 };
 343
 344 struct root_item_info {
 345         /* level of the root */
 346         u8 level;
 347         /* number of nodes at this level, must be 1 for a root */
 348         int node_count;
 349         u64 bytenr;
 350         u64 gen;
 351         struct cache_extent cache_extent;
 352 };
 353
 354 /*
 355  * Error bit for low memory mode check.
 356  *
 357  * Currently no caller cares about it yet.  Just internal use for error
 358  * classification.
 359  */
 360 #define BACKREF_MISSING         (1 << 0) /* Backref missing in extent tree */
 361 #define BACKREF_MISMATCH        (1 << 1) /* Backref exists but does not match */
 362 #define BYTES_UNALIGNED         (1 << 2) /* Some bytes are not aligned */
 363 #define REFERENCER_MISSING      (1 << 3) /* Referencer not found */
 364 #define REFERENCER_MISMATCH     (1 << 4) /* Referenceer found but does not match */
 365 #define CROSSING_STRIPE_BOUNDARY (1 << 4) /* For kernel scrub workaround */
 366 #define ITEM_SIZE_MISMATCH      (1 << 5) /* Bad item size */
 367 #define UNKNOWN_TYPE            (1 << 6) /* Unknown type */
 368 #define ACCOUNTING_MISMATCH     (1 << 7) /* Used space accounting error */
 369 #define CHUNK_TYPE_MISMATCH     (1 << 8)
 370
 371 static void *print_status_check(void *p)
 372 {
 373         struct task_ctx *priv = p;
 374         const char work_indicator[] = { '.', 'o', 'O', 'o' };
 375         uint32_t count = 0;
 376         static char *task_position_string[] = {
 377                 "checking extents",
 378                 "checking free space cache",
 379                 "checking fs roots",
 380         };
 381
 382         task_period_start(priv->info, 1000 /* 1s */);
 383
 384         if (priv->tp == TASK_NOTHING)
 385                 return NULL;
 386
 387         while (1) {
 388                 printf("%s [%c]\r", task_position_string[priv->tp],
 389                                 work_indicator[count % 4]);
 390                 count++;
 391                 fflush(stdout);
 392                 task_period_wait(priv->info);
 393         }
 394         return NULL;
 395 }
 396
 397 static int print_status_return(void *p)
 398 {
 399         printf("\n");
 400         fflush(stdout);
 401
 402         return 0;
 403 }
 404
 405 static enum btrfs_check_mode parse_check_mode(const char *str)
 406 {
 407         if (strcmp(str, "lowmem") == 0)
 408                 return CHECK_MODE_LOWMEM;
 409         if (strcmp(str, "orig") == 0)
 410                 return CHECK_MODE_ORIGINAL;
 411         if (strcmp(str, "original") == 0)
 412                 return CHECK_MODE_ORIGINAL;
 413
 414         return CHECK_MODE_UNKNOWN;
 415 }
 416
 417 /* Compatible function to allow reuse of old codes */
 418 static u64 first_extent_gap(struct rb_root *holes)
 419 {
 420         struct file_extent_hole *hole;
 421
 422         if (RB_EMPTY_ROOT(holes))
 423                 return (u64)-1;
 424
 425         hole = rb_entry(rb_first(holes), struct file_extent_hole, node);
 426         return hole->start;
 427 }
 428
 429 static int compare_hole(struct rb_node *node1, struct rb_node *node2)
 430 {
 431         struct file_extent_hole *hole1;
 432         struct file_extent_hole *hole2;
 433
 434         hole1 = rb_entry(node1, struct file_extent_hole, node);
 435         hole2 = rb_entry(node2, struct file_extent_hole, node);
 436
 437         if (hole1->start > hole2->start)
 438                 return -1;
 439         if (hole1->start < hole2->start)
 440                 return 1;
 441         /* Now hole1->start == hole2->start */
 442         if (hole1->len >= hole2->len)
 443                 /*
 444                  * Hole 1 will be merge center
 445                  * Same hole will be merged later
 446                  */
 447                 return -1;
 448         /* Hole 2 will be merge center */
 449         return 1;
 450 }
 451
 452 /*
 453  * Add a hole to the record
 454  *
 455  * This will do hole merge for copy_file_extent_holes(),
 456  * which will ensure there won't be continuous holes.
 457  */
 458 static int add_file_extent_hole(struct rb_root *holes,
 459                                 u64 start, u64 len)
 460 {
 461         struct file_extent_hole *hole;
 462         struct file_extent_hole *prev = NULL;
 463         struct file_extent_hole *next = NULL;
 464
 465         hole = malloc(sizeof(*hole));
 466         if (!hole)
 467                 return -ENOMEM;
 468         hole->start = start;
 469         hole->len = len;
 470         /* Since compare will not return 0, no -EEXIST will happen */
 471         rb_insert(holes, &hole->node, compare_hole);
 472
 473         /* simple merge with previous hole */
 474         if (rb_prev(&hole->node))
 475                 prev = rb_entry(rb_prev(&hole->node), struct file_extent_hole,
 476                                 node);
 477         if (prev && prev->start + prev->len >= hole->start) {
 478                 hole->len = hole->start + hole->len - prev->start;
 479                 hole->start = prev->start;
 480                 rb_erase(&prev->node, holes);
 481                 free(prev);
 482                 prev = NULL;
 483         }
 484
 485         /* iterate merge with next holes */
 486         while (1) {
 487                 if (!rb_next(&hole->node))
 488                         break;
 489                 next = rb_entry(rb_next(&hole->node), struct file_extent_hole,
 490                                         node);
 491                 if (hole->start + hole->len >= next->start) {
 492                         if (hole->start + hole->len <= next->start + next->len)
 493                                 hole->len = next->start + next->len -
 494                                             hole->start;
 495                         rb_erase(&next->node, holes);
 496                         free(next);
 497                         next = NULL;
 498                 } else
 499                         break;
 500         }
 501         return 0;
 502 }
 503
 504 static int compare_hole_range(struct rb_node *node, void *data)
 505 {
 506         struct file_extent_hole *hole;
 507         u64 start;
 508
 509         hole = (struct file_extent_hole *)data;
 510         start = hole->start;
 511
 512         hole = rb_entry(node, struct file_extent_hole, node);
 513         if (start < hole->start)
 514                 return -1;
 515         if (start >= hole->start && start < hole->start + hole->len)
 516                 return 0;
 517         return 1;
 518 }
 519
 520 /*
 521  * Delete a hole in the record
 522  *
 523  * This will do the hole split and is much restrict than add.
 524  */
 525 static int del_file_extent_hole(struct rb_root *holes,
 526                                 u64 start, u64 len)
 527 {
 528         struct file_extent_hole *hole;
 529         struct file_extent_hole tmp;
 530         u64 prev_start = 0;
 531         u64 prev_len = 0;
 532         u64 next_start = 0;
 533         u64 next_len = 0;
 534         struct rb_node *node;
 535         int have_prev = 0;
 536         int have_next = 0;
 537         int ret = 0;
 538
 539         tmp.start = start;
 540         tmp.len = len;
 541         node = rb_search(holes, &tmp, compare_hole_range, NULL);
 542         if (!node)
 543                 return -EEXIST;
 544         hole = rb_entry(node, struct file_extent_hole, node);
 545         if (start + len > hole->start + hole->len)
 546                 return -EEXIST;
 547
 548         /*
 549          * Now there will be no overlap, delete the hole and re-add the
 550          * split(s) if they exists.
 551          */
 552         if (start > hole->start) {
 553                 prev_start = hole->start;
 554                 prev_len = start - hole->start;
 555                 have_prev = 1;
 556         }
 557         if (hole->start + hole->len > start + len) {
 558                 next_start = start + len;
 559                 next_len = hole->start + hole->len - start - len;
 560                 have_next = 1;
 561         }
 562         rb_erase(node, holes);
 563         free(hole);
 564         if (have_prev) {
 565                 ret = add_file_extent_hole(holes, prev_start, prev_len);
 566                 if (ret < 0)
 567                         return ret;
 568         }
 569         if (have_next) {
 570                 ret = add_file_extent_hole(holes, next_start, next_len);
 571                 if (ret < 0)
 572                         return ret;
 573         }
 574         return 0;
 575 }
 576
 577 static int copy_file_extent_holes(struct rb_root *dst,
 578                                   struct rb_root *src)
 579 {
 580         struct file_extent_hole *hole;
 581         struct rb_node *node;
 582         int ret = 0;
 583
 584         node = rb_first(src);
 585         while (node) {
 586                 hole = rb_entry(node, struct file_extent_hole, node);
 587                 ret = add_file_extent_hole(dst, hole->start, hole->len);
 588                 if (ret)
 589                         break;
 590                 node = rb_next(node);
 591         }
 592         return ret;
 593 }
 594
 595 static void free_file_extent_holes(struct rb_root *holes)
 596 {
 597         struct rb_node *node;
 598         struct file_extent_hole *hole;
 599
 600         node = rb_first(holes);
 601         while (node) {
 602                 hole = rb_entry(node, struct file_extent_hole, node);
 603                 rb_erase(node, holes);
 604                 free(hole);
 605                 node = rb_first(holes);
 606         }
 607 }
 608
 609 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info);
 610
 611 static void record_root_in_trans(struct btrfs_trans_handle *trans,
 612                                  struct btrfs_root *root)
 613 {
 614         if (root->last_trans != trans->transid) {
 615                 root->track_dirty = 1;
 616                 root->last_trans = trans->transid;
 617                 root->commit_root = root->node;
 618                 extent_buffer_get(root->node);
 619         }
 620 }
 621
 622 static u8 imode_to_type(u32 imode)
 623 {
 624 #define S_SHIFT 12
 625         static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
 626                 [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
 627                 [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
 628                 [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
 629                 [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
 630                 [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
 631                 [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
 632                 [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
 633         };
 634
 635         return btrfs_type_by_mode[(imode & S_IFMT) >> S_SHIFT];
 636 #undef S_SHIFT
 637 }
 638
 639 static int device_record_compare(struct rb_node *node1, struct rb_node *node2)
 640 {
 641         struct device_record *rec1;
 642         struct device_record *rec2;
 643
 644         rec1 = rb_entry(node1, struct device_record, node);
 645         rec2 = rb_entry(node2, struct device_record, node);
 646         if (rec1->devid > rec2->devid)
 647                 return -1;
 648         else if (rec1->devid < rec2->devid)
 649                 return 1;
 650         else
 651                 return 0;
 652 }
 653
 654 static struct inode_record *clone_inode_rec(struct inode_record *orig_rec)
 655 {
 656         struct inode_record *rec;
 657         struct inode_backref *backref;
 658         struct inode_backref *orig;
 659         struct inode_backref *tmp;
 660         struct orphan_data_extent *src_orphan;
 661         struct orphan_data_extent *dst_orphan;
 662         struct rb_node *rb;
 663         size_t size;
 664         int ret;
 665
 666         rec = malloc(sizeof(*rec));
 667         if (!rec)
 668                 return ERR_PTR(-ENOMEM);
 669         memcpy(rec, orig_rec, sizeof(*rec));
 670         rec->refs = 1;
 671         INIT_LIST_HEAD(&rec->backrefs);
 672         INIT_LIST_HEAD(&rec->orphan_extents);
 673         rec->holes = RB_ROOT;
 674
 675         list_for_each_entry(orig, &orig_rec->backrefs, list) {
 676                 size = sizeof(*orig) + orig->namelen + 1;
 677                 backref = malloc(size);
 678                 if (!backref) {
 679                         ret = -ENOMEM;
 680                         goto cleanup;
 681                 }
 682                 memcpy(backref, orig, size);
 683                 list_add_tail(&backref->list, &rec->backrefs);
 684         }
 685         list_for_each_entry(src_orphan, &orig_rec->orphan_extents, list) {
 686                 dst_orphan = malloc(sizeof(*dst_orphan));
 687                 if (!dst_orphan) {
 688                         ret = -ENOMEM;
 689                         goto cleanup;
 690                 }
 691                 memcpy(dst_orphan, src_orphan, sizeof(*src_orphan));
 692                 list_add_tail(&dst_orphan->list, &rec->orphan_extents);
 693         }
 694         ret = copy_file_extent_holes(&rec->holes, &orig_rec->holes);
 695         if (ret < 0)
 696                 goto cleanup_rb;
 697
 698         return rec;
 699
 700 cleanup_rb:
 701         rb = rb_first(&rec->holes);
 702         while (rb) {
 703                 struct file_extent_hole *hole;
 704
 705                 hole = rb_entry(rb, struct file_extent_hole, node);
 706                 rb = rb_next(rb);
 707                 free(hole);
 708         }
 709
 710 cleanup:
 711         if (!list_empty(&rec->backrefs))
 712                 list_for_each_entry_safe(orig, tmp, &rec->backrefs, list) {
 713                         list_del(&orig->list);
 714                         free(orig);
 715                 }
 716
 717         if (!list_empty(&rec->orphan_extents))
 718                 list_for_each_entry_safe(orig, tmp, &rec->orphan_extents, list) {
 719                         list_del(&orig->list);
 720                         free(orig);
 721                 }
 722
 723         free(rec);
 724
 725         return ERR_PTR(ret);
 726 }
 727
 728 static void print_orphan_data_extents(struct list_head *orphan_extents,
 729                                       u64 objectid)
 730 {
 731         struct orphan_data_extent *orphan;
 732
 733         if (list_empty(orphan_extents))
 734                 return;
 735         printf("The following data extent is lost in tree %llu:\n",
 736                objectid);
 737         list_for_each_entry(orphan, orphan_extents, list) {
 738                 printf("\tinode: %llu, offset:%llu, disk_bytenr: %llu, disk_len: %llu\n",
 739                        orphan->objectid, orphan->offset, orphan->disk_bytenr,
 740                        orphan->disk_len);
 741         }
 742 }
 743
 744 static void print_inode_error(struct btrfs_root *root, struct inode_record *rec)
 745 {
 746         u64 root_objectid = root->root_key.objectid;
 747         int errors = rec->errors;
 748
 749         if (!errors)
 750                 return;
 751         /* reloc root errors, we print its corresponding fs root objectid*/
 752         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
 753                 root_objectid = root->root_key.offset;
 754                 fprintf(stderr, "reloc");
 755         }
 756         fprintf(stderr, "root %llu inode %llu errors %x",
 757                 (unsigned long long) root_objectid,
 758                 (unsigned long long) rec->ino, rec->errors);
 759
 760         if (errors & I_ERR_NO_INODE_ITEM)
 761                 fprintf(stderr, ", no inode item");
 762         if (errors & I_ERR_NO_ORPHAN_ITEM)
 763                 fprintf(stderr, ", no orphan item");
 764         if (errors & I_ERR_DUP_INODE_ITEM)
 765                 fprintf(stderr, ", dup inode item");
 766         if (errors & I_ERR_DUP_DIR_INDEX)
 767                 fprintf(stderr, ", dup dir index");
 768         if (errors & I_ERR_ODD_DIR_ITEM)
 769                 fprintf(stderr, ", odd dir item");
 770         if (errors & I_ERR_ODD_FILE_EXTENT)
 771                 fprintf(stderr, ", odd file extent");
 772         if (errors & I_ERR_BAD_FILE_EXTENT)
 773                 fprintf(stderr, ", bad file extent");
 774         if (errors & I_ERR_FILE_EXTENT_OVERLAP)
 775                 fprintf(stderr, ", file extent overlap");
 776         if (errors & I_ERR_FILE_EXTENT_DISCOUNT)
 777                 fprintf(stderr, ", file extent discount");
 778         if (errors & I_ERR_DIR_ISIZE_WRONG)
 779                 fprintf(stderr, ", dir isize wrong");
 780         if (errors & I_ERR_FILE_NBYTES_WRONG)
 781                 fprintf(stderr, ", nbytes wrong");
 782         if (errors & I_ERR_ODD_CSUM_ITEM)
 783                 fprintf(stderr, ", odd csum item");
 784         if (errors & I_ERR_SOME_CSUM_MISSING)
 785                 fprintf(stderr, ", some csum missing");
 786         if (errors & I_ERR_LINK_COUNT_WRONG)
 787                 fprintf(stderr, ", link count wrong");
 788         if (errors & I_ERR_FILE_EXTENT_ORPHAN)
 789                 fprintf(stderr, ", orphan file extent");
 790         fprintf(stderr, "\n");
 791         /* Print the orphan extents if needed */
 792         if (errors & I_ERR_FILE_EXTENT_ORPHAN)
 793                 print_orphan_data_extents(&rec->orphan_extents, root->objectid);
 794
 795         /* Print the holes if needed */
 796         if (errors & I_ERR_FILE_EXTENT_DISCOUNT) {
 797                 struct file_extent_hole *hole;
 798                 struct rb_node *node;
 799                 int found = 0;
 800
 801                 node = rb_first(&rec->holes);
 802                 fprintf(stderr, "Found file extent holes:\n");
 803                 while (node) {
 804                         found = 1;
 805                         hole = rb_entry(node, struct file_extent_hole, node);
 806                         fprintf(stderr, "\tstart: %llu, len: %llu\n",
 807                                 hole->start, hole->len);
 808                         node = rb_next(node);
 809                 }
 810                 if (!found)
 811                         fprintf(stderr, "\tstart: 0, len: %llu\n",
 812                                 round_up(rec->isize, root->sectorsize));
 813         }
 814 }
 815
 816 static void print_ref_error(int errors)
 817 {
 818         if (errors & REF_ERR_NO_DIR_ITEM)
 819                 fprintf(stderr, ", no dir item");
 820         if (errors & REF_ERR_NO_DIR_INDEX)
 821                 fprintf(stderr, ", no dir index");
 822         if (errors & REF_ERR_NO_INODE_REF)
 823                 fprintf(stderr, ", no inode ref");
 824         if (errors & REF_ERR_DUP_DIR_ITEM)
 825                 fprintf(stderr, ", dup dir item");
 826         if (errors & REF_ERR_DUP_DIR_INDEX)
 827                 fprintf(stderr, ", dup dir index");
 828         if (errors & REF_ERR_DUP_INODE_REF)
 829                 fprintf(stderr, ", dup inode ref");
 830         if (errors & REF_ERR_INDEX_UNMATCH)
 831                 fprintf(stderr, ", index mismatch");
 832         if (errors & REF_ERR_FILETYPE_UNMATCH)
 833                 fprintf(stderr, ", filetype mismatch");
 834         if (errors & REF_ERR_NAME_TOO_LONG)
 835                 fprintf(stderr, ", name too long");
 836         if (errors & REF_ERR_NO_ROOT_REF)
 837                 fprintf(stderr, ", no root ref");
 838         if (errors & REF_ERR_NO_ROOT_BACKREF)
 839                 fprintf(stderr, ", no root backref");
 840         if (errors & REF_ERR_DUP_ROOT_REF)
 841                 fprintf(stderr, ", dup root ref");
 842         if (errors & REF_ERR_DUP_ROOT_BACKREF)
 843                 fprintf(stderr, ", dup root backref");
 844         fprintf(stderr, "\n");
 845 }
 846
 847 static struct inode_record *get_inode_rec(struct cache_tree *inode_cache,
 848                                           u64 ino, int mod)
 849 {
 850         struct ptr_node *node;
 851         struct cache_extent *cache;
 852         struct inode_record *rec = NULL;
 853         int ret;
 854
 855         cache = lookup_cache_extent(inode_cache, ino, 1);
 856         if (cache) {
 857                 node = container_of(cache, struct ptr_node, cache);
 858                 rec = node->data;
 859                 if (mod && rec->refs > 1) {
 860                         node->data = clone_inode_rec(rec);
 861                         if (IS_ERR(node->data))
 862                                 return node->data;
 863                         rec->refs--;
 864                         rec = node->data;
 865                 }
 866         } else if (mod) {
 867                 rec = calloc(1, sizeof(*rec));
 868                 if (!rec)
 869                         return ERR_PTR(-ENOMEM);
 870                 rec->ino = ino;
 871                 rec->extent_start = (u64)-1;
 872                 rec->refs = 1;
 873                 INIT_LIST_HEAD(&rec->backrefs);
 874                 INIT_LIST_HEAD(&rec->orphan_extents);
 875                 rec->holes = RB_ROOT;
 876
 877                 node = malloc(sizeof(*node));
 878                 if (!node) {
 879                         free(rec);
 880                         return ERR_PTR(-ENOMEM);
 881                 }
 882                 node->cache.start = ino;
 883                 node->cache.size = 1;
 884                 node->data = rec;
 885
 886                 if (ino == BTRFS_FREE_INO_OBJECTID)
 887                         rec->found_link = 1;
 888
 889                 ret = insert_cache_extent(inode_cache, &node->cache);
 890                 if (ret)
 891                         return ERR_PTR(-EEXIST);
 892         }
 893         return rec;
 894 }
 895
 896 static void free_orphan_data_extents(struct list_head *orphan_extents)
 897 {
 898         struct orphan_data_extent *orphan;
 899
 900         while (!list_empty(orphan_extents)) {
 901                 orphan = list_entry(orphan_extents->next,
 902                                     struct orphan_data_extent, list);
 903                 list_del(&orphan->list);
 904                 free(orphan);
 905         }
 906 }
 907
 908 static void free_inode_rec(struct inode_record *rec)
 909 {
 910         struct inode_backref *backref;
 911
 912         if (--rec->refs > 0)
 913                 return;
 914
 915         while (!list_empty(&rec->backrefs)) {
 916                 backref = to_inode_backref(rec->backrefs.next);
 917                 list_del(&backref->list);
 918                 free(backref);
 919         }
 920         free_orphan_data_extents(&rec->orphan_extents);
 921         free_file_extent_holes(&rec->holes);
 922         free(rec);
 923 }
 924
 925 static int can_free_inode_rec(struct inode_record *rec)
 926 {
 927         if (!rec->errors && rec->checked && rec->found_inode_item &&
 928             rec->nlink == rec->found_link && list_empty(&rec->backrefs))
 929                 return 1;
 930         return 0;
 931 }
 932
 933 static void maybe_free_inode_rec(struct cache_tree *inode_cache,
 934                                  struct inode_record *rec)
 935 {
 936         struct cache_extent *cache;
 937         struct inode_backref *tmp, *backref;
 938         struct ptr_node *node;
 939         u8 filetype;
 940
 941         if (!rec->found_inode_item)
 942                 return;
 943
 944         filetype = imode_to_type(rec->imode);
 945         list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
 946                 if (backref->found_dir_item && backref->found_dir_index) {
 947                         if (backref->filetype != filetype)
 948                                 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
 949                         if (!backref->errors && backref->found_inode_ref &&
 950                             rec->nlink == rec->found_link) {
 951                                 list_del(&backref->list);
 952                                 free(backref);
 953                         }
 954                 }
 955         }
 956
 957         if (!rec->checked || rec->merging)
 958                 return;
 959
 960         if (S_ISDIR(rec->imode)) {
 961                 if (rec->found_size != rec->isize)
 962                         rec->errors |= I_ERR_DIR_ISIZE_WRONG;
 963                 if (rec->found_file_extent)
 964                         rec->errors |= I_ERR_ODD_FILE_EXTENT;
 965         } else if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
 966                 if (rec->found_dir_item)
 967                         rec->errors |= I_ERR_ODD_DIR_ITEM;
 968                 if (rec->found_size != rec->nbytes)
 969                         rec->errors |= I_ERR_FILE_NBYTES_WRONG;
 970                 if (rec->nlink > 0 && !no_holes &&
 971                     (rec->extent_end < rec->isize ||
 972                      first_extent_gap(&rec->holes) < rec->isize))
 973                         rec->errors |= I_ERR_FILE_EXTENT_DISCOUNT;
 974         }
 975
 976         if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
 977                 if (rec->found_csum_item && rec->nodatasum)
 978                         rec->errors |= I_ERR_ODD_CSUM_ITEM;
 979                 if (rec->some_csum_missing && !rec->nodatasum)
 980                         rec->errors |= I_ERR_SOME_CSUM_MISSING;
 981         }
 982
 983         BUG_ON(rec->refs != 1);
 984         if (can_free_inode_rec(rec)) {
 985                 cache = lookup_cache_extent(inode_cache, rec->ino, 1);
 986                 node = container_of(cache, struct ptr_node, cache);
 987                 BUG_ON(node->data != rec);
 988                 remove_cache_extent(inode_cache, &node->cache);
 989                 free(node);
 990                 free_inode_rec(rec);
 991         }
 992 }
 993
 994 static int check_orphan_item(struct btrfs_root *root, u64 ino)
 995 {
 996         struct btrfs_path path;
 997         struct btrfs_key key;
 998         int ret;
 999
1000         key.objectid = BTRFS_ORPHAN_OBJECTID;
1001         key.type = BTRFS_ORPHAN_ITEM_KEY;
1002         key.offset = ino;
1003
1004         btrfs_init_path(&path);
1005         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
1006         btrfs_release_path(&path);
1007         if (ret > 0)
1008                 ret = -ENOENT;
1009         return ret;
1010 }
1011
1012 static int process_inode_item(struct extent_buffer *eb,
1013                               int slot, struct btrfs_key *key,
1014                               struct shared_node *active_node)
1015 {
1016         struct inode_record *rec;
1017         struct btrfs_inode_item *item;
1018
1019         rec = active_node->current;
1020         BUG_ON(rec->ino != key->objectid || rec->refs > 1);
1021         if (rec->found_inode_item) {
1022                 rec->errors |= I_ERR_DUP_INODE_ITEM;
1023                 return 1;
1024         }
1025         item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
1026         rec->nlink = btrfs_inode_nlink(eb, item);
1027         rec->isize = btrfs_inode_size(eb, item);
1028         rec->nbytes = btrfs_inode_nbytes(eb, item);
1029         rec->imode = btrfs_inode_mode(eb, item);
1030         if (btrfs_inode_flags(eb, item) & BTRFS_INODE_NODATASUM)
1031                 rec->nodatasum = 1;
1032         rec->found_inode_item = 1;
1033         if (rec->nlink == 0)
1034                 rec->errors |= I_ERR_NO_ORPHAN_ITEM;
1035         maybe_free_inode_rec(&active_node->inode_cache, rec);
1036         return 0;
1037 }
1038
1039 static struct inode_backref *get_inode_backref(struct inode_record *rec,
1040                                                 const char *name,
1041                                                 int namelen, u64 dir)
1042 {
1043         struct inode_backref *backref;
1044
1045         list_for_each_entry(backref, &rec->backrefs, list) {
1046                 if (rec->ino == BTRFS_MULTIPLE_OBJECTIDS)
1047                         break;
1048                 if (backref->dir != dir || backref->namelen != namelen)
1049                         continue;
1050                 if (memcmp(name, backref->name, namelen))
1051                         continue;
1052                 return backref;
1053         }
1054
1055         backref = malloc(sizeof(*backref) + namelen + 1);
1056         if (!backref)
1057                 return NULL;
1058         memset(backref, 0, sizeof(*backref));
1059         backref->dir = dir;
1060         backref->namelen = namelen;
1061         memcpy(backref->name, name, namelen);
1062         backref->name[namelen] = '\0';
1063         list_add_tail(&backref->list, &rec->backrefs);
1064         return backref;
1065 }
1066
1067 static int add_inode_backref(struct cache_tree *inode_cache,
1068                              u64 ino, u64 dir, u64 index,
1069                              const char *name, int namelen,
1070                              u8 filetype, u8 itemtype, int errors)
1071 {
1072         struct inode_record *rec;
1073         struct inode_backref *backref;
1074
1075         rec = get_inode_rec(inode_cache, ino, 1);
1076         BUG_ON(IS_ERR(rec));
1077         backref = get_inode_backref(rec, name, namelen, dir);
1078         BUG_ON(!backref);
1079         if (errors)
1080                 backref->errors |= errors;
1081         if (itemtype == BTRFS_DIR_INDEX_KEY) {
1082                 if (backref->found_dir_index)
1083                         backref->errors |= REF_ERR_DUP_DIR_INDEX;
1084                 if (backref->found_inode_ref && backref->index != index)
1085                         backref->errors |= REF_ERR_INDEX_UNMATCH;
1086                 if (backref->found_dir_item && backref->filetype != filetype)
1087                         backref->errors |= REF_ERR_FILETYPE_UNMATCH;
1088
1089                 backref->index = index;
1090                 backref->filetype = filetype;
1091                 backref->found_dir_index = 1;
1092         } else if (itemtype == BTRFS_DIR_ITEM_KEY) {
1093                 rec->found_link++;
1094                 if (backref->found_dir_item)
1095                         backref->errors |= REF_ERR_DUP_DIR_ITEM;
1096                 if (backref->found_dir_index && backref->filetype != filetype)
1097                         backref->errors |= REF_ERR_FILETYPE_UNMATCH;
1098
1099                 backref->filetype = filetype;
1100                 backref->found_dir_item = 1;
1101         } else if ((itemtype == BTRFS_INODE_REF_KEY) ||
1102                    (itemtype == BTRFS_INODE_EXTREF_KEY)) {
1103                 if (backref->found_inode_ref)
1104                         backref->errors |= REF_ERR_DUP_INODE_REF;
1105                 if (backref->found_dir_index && backref->index != index)
1106                         backref->errors |= REF_ERR_INDEX_UNMATCH;
1107                 else
1108                         backref->index = index;
1109
1110                 backref->ref_type = itemtype;
1111                 backref->found_inode_ref = 1;
1112         } else {
1113                 BUG_ON(1);
1114         }
1115
1116         maybe_free_inode_rec(inode_cache, rec);
1117         return 0;
1118 }
1119
1120 static int merge_inode_recs(struct inode_record *src, struct inode_record *dst,
1121                             struct cache_tree *dst_cache)
1122 {
1123         struct inode_backref *backref;
1124         u32 dir_count = 0;
1125         int ret = 0;
1126
1127         dst->merging = 1;
1128         list_for_each_entry(backref, &src->backrefs, list) {
1129                 if (backref->found_dir_index) {
1130                         add_inode_backref(dst_cache, dst->ino, backref->dir,
1131                                         backref->index, backref->name,
1132                                         backref->namelen, backref->filetype,
1133                                         BTRFS_DIR_INDEX_KEY, backref->errors);
1134                 }
1135                 if (backref->found_dir_item) {
1136                         dir_count++;
1137                         add_inode_backref(dst_cache, dst->ino,
1138                                         backref->dir, 0, backref->name,
1139                                         backref->namelen, backref->filetype,
1140                                         BTRFS_DIR_ITEM_KEY, backref->errors);
1141                 }
1142                 if (backref->found_inode_ref) {
1143                         add_inode_backref(dst_cache, dst->ino,
1144                                         backref->dir, backref->index,
1145                                         backref->name, backref->namelen, 0,
1146                                         backref->ref_type, backref->errors);
1147                 }
1148         }
1149
1150         if (src->found_dir_item)
1151                 dst->found_dir_item = 1;
1152         if (src->found_file_extent)
1153                 dst->found_file_extent = 1;
1154         if (src->found_csum_item)
1155                 dst->found_csum_item = 1;
1156         if (src->some_csum_missing)
1157                 dst->some_csum_missing = 1;
1158         if (first_extent_gap(&dst->holes) > first_extent_gap(&src->holes)) {
1159                 ret = copy_file_extent_holes(&dst->holes, &src->holes);
1160                 if (ret < 0)
1161                         return ret;
1162         }
1163
1164         BUG_ON(src->found_link < dir_count);
1165         dst->found_link += src->found_link - dir_count;
1166         dst->found_size += src->found_size;
1167         if (src->extent_start != (u64)-1) {
1168                 if (dst->extent_start == (u64)-1) {
1169                         dst->extent_start = src->extent_start;
1170                         dst->extent_end = src->extent_end;
1171                 } else {
1172                         if (dst->extent_end > src->extent_start)
1173                                 dst->errors |= I_ERR_FILE_EXTENT_OVERLAP;
1174                         else if (dst->extent_end < src->extent_start) {
1175                                 ret = add_file_extent_hole(&dst->holes,
1176                                         dst->extent_end,
1177                                         src->extent_start - dst->extent_end);
1178                         }
1179                         if (dst->extent_end < src->extent_end)
1180                                 dst->extent_end = src->extent_end;
1181                 }
1182         }
1183
1184         dst->errors |= src->errors;
1185         if (src->found_inode_item) {
1186                 if (!dst->found_inode_item) {
1187                         dst->nlink = src->nlink;
1188                         dst->isize = src->isize;
1189                         dst->nbytes = src->nbytes;
1190                         dst->imode = src->imode;
1191                         dst->nodatasum = src->nodatasum;
1192                         dst->found_inode_item = 1;
1193                 } else {
1194                         dst->errors |= I_ERR_DUP_INODE_ITEM;
1195                 }
1196         }
1197         dst->merging = 0;
1198
1199         return 0;
1200 }
1201
1202 static int splice_shared_node(struct shared_node *src_node,
1203                               struct shared_node *dst_node)
1204 {
1205         struct cache_extent *cache;
1206         struct ptr_node *node, *ins;
1207         struct cache_tree *src, *dst;
1208         struct inode_record *rec, *conflict;
1209         u64 current_ino = 0;
1210         int splice = 0;
1211         int ret;
1212
1213         if (--src_node->refs == 0)
1214                 splice = 1;
1215         if (src_node->current)
1216                 current_ino = src_node->current->ino;
1217
1218         src = &src_node->root_cache;
1219         dst = &dst_node->root_cache;
1220 again:
1221         cache = search_cache_extent(src, 0);
1222         while (cache) {
1223                 node = container_of(cache, struct ptr_node, cache);
1224                 rec = node->data;
1225                 cache = next_cache_extent(cache);
1226
1227                 if (splice) {
1228                         remove_cache_extent(src, &node->cache);
1229                         ins = node;
1230                 } else {
1231                         ins = malloc(sizeof(*ins));
1232                         BUG_ON(!ins);
1233                         ins->cache.start = node->cache.start;
1234                         ins->cache.size = node->cache.size;
1235                         ins->data = rec;
1236                         rec->refs++;
1237                 }
1238                 ret = insert_cache_extent(dst, &ins->cache);
1239                 if (ret == -EEXIST) {
1240                         conflict = get_inode_rec(dst, rec->ino, 1);
1241                         BUG_ON(IS_ERR(conflict));
1242                         merge_inode_recs(rec, conflict, dst);
1243                         if (rec->checked) {
1244                                 conflict->checked = 1;
1245                                 if (dst_node->current == conflict)
1246                                         dst_node->current = NULL;
1247                         }
1248                         maybe_free_inode_rec(dst, conflict);
1249                         free_inode_rec(rec);
1250                         free(ins);
1251                 } else {
1252                         BUG_ON(ret);
1253                 }
1254         }
1255
1256         if (src == &src_node->root_cache) {
1257                 src = &src_node->inode_cache;
1258                 dst = &dst_node->inode_cache;
1259                 goto again;
1260         }
1261
1262         if (current_ino > 0 && (!dst_node->current ||
1263             current_ino > dst_node->current->ino)) {
1264                 if (dst_node->current) {
1265                         dst_node->current->checked = 1;
1266                         maybe_free_inode_rec(dst, dst_node->current);
1267                 }
1268                 dst_node->current = get_inode_rec(dst, current_ino, 1);
1269                 BUG_ON(IS_ERR(dst_node->current));
1270         }
1271         return 0;
1272 }
1273
1274 static void free_inode_ptr(struct cache_extent *cache)
1275 {
1276         struct ptr_node *node;
1277         struct inode_record *rec;
1278
1279         node = container_of(cache, struct ptr_node, cache);
1280         rec = node->data;
1281         free_inode_rec(rec);
1282         free(node);
1283 }
1284
1285 FREE_EXTENT_CACHE_BASED_TREE(inode_recs, free_inode_ptr);
1286
1287 static struct shared_node *find_shared_node(struct cache_tree *shared,
1288                                             u64 bytenr)
1289 {
1290         struct cache_extent *cache;
1291         struct shared_node *node;
1292
1293         cache = lookup_cache_extent(shared, bytenr, 1);
1294         if (cache) {
1295                 node = container_of(cache, struct shared_node, cache);
1296                 return node;
1297         }
1298         return NULL;
1299 }
1300
1301 static int add_shared_node(struct cache_tree *shared, u64 bytenr, u32 refs)
1302 {
1303         int ret;
1304         struct shared_node *node;
1305
1306         node = calloc(1, sizeof(*node));
1307         if (!node)
1308                 return -ENOMEM;
1309         node->cache.start = bytenr;
1310         node->cache.size = 1;
1311         cache_tree_init(&node->root_cache);
1312         cache_tree_init(&node->inode_cache);
1313         node->refs = refs;
1314
1315         ret = insert_cache_extent(shared, &node->cache);
1316
1317         return ret;
1318 }
1319
1320 static int enter_shared_node(struct btrfs_root *root, u64 bytenr, u32 refs,
1321                              struct walk_control *wc, int level)
1322 {
1323         struct shared_node *node;
1324         struct shared_node *dest;
1325         int ret;
1326
1327         if (level == wc->active_node)
1328                 return 0;
1329
1330         BUG_ON(wc->active_node <= level);
1331         node = find_shared_node(&wc->shared, bytenr);
1332         if (!node) {
1333                 ret = add_shared_node(&wc->shared, bytenr, refs);
1334                 BUG_ON(ret);
1335                 node = find_shared_node(&wc->shared, bytenr);
1336                 wc->nodes[level] = node;
1337                 wc->active_node = level;
1338                 return 0;
1339         }
1340
1341         if (wc->root_level == wc->active_node &&
1342             btrfs_root_refs(&root->root_item) == 0) {
1343                 if (--node->refs == 0) {
1344                         free_inode_recs_tree(&node->root_cache);
1345                         free_inode_recs_tree(&node->inode_cache);
1346                         remove_cache_extent(&wc->shared, &node->cache);
1347                         free(node);
1348                 }
1349                 return 1;
1350         }
1351
1352         dest = wc->nodes[wc->active_node];
1353         splice_shared_node(node, dest);
1354         if (node->refs == 0) {
1355                 remove_cache_extent(&wc->shared, &node->cache);
1356                 free(node);
1357         }
1358         return 1;
1359 }
1360
1361 static int leave_shared_node(struct btrfs_root *root,
1362                              struct walk_control *wc, int level)
1363 {
1364         struct shared_node *node;
1365         struct shared_node *dest;
1366         int i;
1367
1368         if (level == wc->root_level)
1369                 return 0;
1370
1371         for (i = level + 1; i < BTRFS_MAX_LEVEL; i++) {
1372                 if (wc->nodes[i])
1373                         break;
1374         }
1375         BUG_ON(i >= BTRFS_MAX_LEVEL);
1376
1377         node = wc->nodes[wc->active_node];
1378         wc->nodes[wc->active_node] = NULL;
1379         wc->active_node = i;
1380
1381         dest = wc->nodes[wc->active_node];
1382         if (wc->active_node < wc->root_level ||
1383             btrfs_root_refs(&root->root_item) > 0) {
1384                 BUG_ON(node->refs <= 1);
1385                 splice_shared_node(node, dest);
1386         } else {
1387                 BUG_ON(node->refs < 2);
1388                 node->refs--;
1389         }
1390         return 0;
1391 }
1392
1393 /*
1394  * Returns:
1395  * < 0 - on error
1396  * 1   - if the root with id child_root_id is a child of root parent_root_id
1397  * 0   - if the root child_root_id isn't a child of the root parent_root_id but
1398  *       has other root(s) as parent(s)
1399  * 2   - if the root child_root_id doesn't have any parent roots
1400  */
1401 static int is_child_root(struct btrfs_root *root, u64 parent_root_id,
1402                          u64 child_root_id)
1403 {
1404         struct btrfs_path path;
1405         struct btrfs_key key;
1406         struct extent_buffer *leaf;
1407         int has_parent = 0;
1408         int ret;
1409
1410         btrfs_init_path(&path);
1411
1412         key.objectid = parent_root_id;
1413         key.type = BTRFS_ROOT_REF_KEY;
1414         key.offset = child_root_id;
1415         ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
1416                                 0, 0);
1417         if (ret < 0)
1418                 return ret;
1419         btrfs_release_path(&path);
1420         if (!ret)
1421                 return 1;
1422
1423         key.objectid = child_root_id;
1424         key.type = BTRFS_ROOT_BACKREF_KEY;
1425         key.offset = 0;
1426         ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
1427                                 0, 0);
1428         if (ret < 0)
1429                 goto out;
1430
1431         while (1) {
1432                 leaf = path.nodes[0];
1433                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1434                         ret = btrfs_next_leaf(root->fs_info->tree_root, &path);
1435                         if (ret)
1436                                 break;
1437                         leaf = path.nodes[0];
1438                 }
1439
1440                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1441                 if (key.objectid != child_root_id ||
1442                     key.type != BTRFS_ROOT_BACKREF_KEY)
1443                         break;
1444
1445                 has_parent = 1;
1446
1447                 if (key.offset == parent_root_id) {
1448                         btrfs_release_path(&path);
1449                         return 1;
1450                 }
1451
1452                 path.slots[0]++;
1453         }
1454 out:
1455         btrfs_release_path(&path);
1456         if (ret < 0)
1457                 return ret;
1458         return has_parent ? 0 : 2;
1459 }
1460
1461 static int process_dir_item(struct btrfs_root *root,
1462                             struct extent_buffer *eb,
1463                             int slot, struct btrfs_key *key,
1464                             struct shared_node *active_node)
1465 {
1466         u32 total;
1467         u32 cur = 0;
1468         u32 len;
1469         u32 name_len;
1470         u32 data_len;
1471         int error;
1472         int nritems = 0;
1473         u8 filetype;
1474         struct btrfs_dir_item *di;
1475         struct inode_record *rec;
1476         struct cache_tree *root_cache;
1477         struct cache_tree *inode_cache;
1478         struct btrfs_key location;
1479         char namebuf[BTRFS_NAME_LEN];
1480
1481         root_cache = &active_node->root_cache;
1482         inode_cache = &active_node->inode_cache;
1483         rec = active_node->current;
1484         rec->found_dir_item = 1;
1485
1486         di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1487         total = btrfs_item_size_nr(eb, slot);
1488         while (cur < total) {
1489                 nritems++;
1490                 btrfs_dir_item_key_to_cpu(eb, di, &location);
1491                 name_len = btrfs_dir_name_len(eb, di);
1492                 data_len = btrfs_dir_data_len(eb, di);
1493                 filetype = btrfs_dir_type(eb, di);
1494
1495                 rec->found_size += name_len;
1496                 if (name_len <= BTRFS_NAME_LEN) {
1497                         len = name_len;
1498                         error = 0;
1499                 } else {
1500                         len = BTRFS_NAME_LEN;
1501                         error = REF_ERR_NAME_TOO_LONG;
1502                 }
1503                 read_extent_buffer(eb, namebuf, (unsigned long)(di + 1), len);
1504
1505                 if (location.type == BTRFS_INODE_ITEM_KEY) {
1506                         add_inode_backref(inode_cache, location.objectid,
1507                                           key->objectid, key->offset, namebuf,
1508                                           len, filetype, key->type, error);
1509                 } else if (location.type == BTRFS_ROOT_ITEM_KEY) {
1510                         add_inode_backref(root_cache, location.objectid,
1511                                           key->objectid, key->offset,
1512                                           namebuf, len, filetype,
1513                                           key->type, error);
1514                 } else {
1515                         fprintf(stderr, "invalid location in dir item %u\n",
1516                                 location.type);
1517                         add_inode_backref(inode_cache, BTRFS_MULTIPLE_OBJECTIDS,
1518                                           key->objectid, key->offset, namebuf,
1519                                           len, filetype, key->type, error);
1520                 }
1521
1522                 len = sizeof(*di) + name_len + data_len;
1523                 di = (struct btrfs_dir_item *)((char *)di + len);
1524                 cur += len;
1525         }
1526         if (key->type == BTRFS_DIR_INDEX_KEY && nritems > 1)
1527                 rec->errors |= I_ERR_DUP_DIR_INDEX;
1528
1529         return 0;
1530 }
1531
1532 static int process_inode_ref(struct extent_buffer *eb,
1533                              int slot, struct btrfs_key *key,
1534                              struct shared_node *active_node)
1535 {
1536         u32 total;
1537         u32 cur = 0;
1538         u32 len;
1539         u32 name_len;
1540         u64 index;
1541         int error;
1542         struct cache_tree *inode_cache;
1543         struct btrfs_inode_ref *ref;
1544         char namebuf[BTRFS_NAME_LEN];
1545
1546         inode_cache = &active_node->inode_cache;
1547
1548         ref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1549         total = btrfs_item_size_nr(eb, slot);
1550         while (cur < total) {
1551                 name_len = btrfs_inode_ref_name_len(eb, ref);
1552                 index = btrfs_inode_ref_index(eb, ref);
1553                 if (name_len <= BTRFS_NAME_LEN) {
1554                         len = name_len;
1555                         error = 0;
1556                 } else {
1557                         len = BTRFS_NAME_LEN;
1558                         error = REF_ERR_NAME_TOO_LONG;
1559                 }
1560                 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
1561                 add_inode_backref(inode_cache, key->objectid, key->offset,
1562                                   index, namebuf, len, 0, key->type, error);
1563
1564                 len = sizeof(*ref) + name_len;
1565                 ref = (struct btrfs_inode_ref *)((char *)ref + len);
1566                 cur += len;
1567         }
1568         return 0;
1569 }
1570
1571 static int process_inode_extref(struct extent_buffer *eb,
1572                                 int slot, struct btrfs_key *key,
1573                                 struct shared_node *active_node)
1574 {
1575         u32 total;
1576         u32 cur = 0;
1577         u32 len;
1578         u32 name_len;
1579         u64 index;
1580         u64 parent;
1581         int error;
1582         struct cache_tree *inode_cache;
1583         struct btrfs_inode_extref *extref;
1584         char namebuf[BTRFS_NAME_LEN];
1585
1586         inode_cache = &active_node->inode_cache;
1587
1588         extref = btrfs_item_ptr(eb, slot, struct btrfs_inode_extref);
1589         total = btrfs_item_size_nr(eb, slot);
1590         while (cur < total) {
1591                 name_len = btrfs_inode_extref_name_len(eb, extref);
1592                 index = btrfs_inode_extref_index(eb, extref);
1593                 parent = btrfs_inode_extref_parent(eb, extref);
1594                 if (name_len <= BTRFS_NAME_LEN) {
1595                         len = name_len;
1596                         error = 0;
1597                 } else {
1598                         len = BTRFS_NAME_LEN;
1599                         error = REF_ERR_NAME_TOO_LONG;
1600                 }
1601                 read_extent_buffer(eb, namebuf,
1602                                    (unsigned long)(extref + 1), len);
1603                 add_inode_backref(inode_cache, key->objectid, parent,
1604                                   index, namebuf, len, 0, key->type, error);
1605
1606                 len = sizeof(*extref) + name_len;
1607                 extref = (struct btrfs_inode_extref *)((char *)extref + len);
1608                 cur += len;
1609         }
1610         return 0;
1611
1612 }
1613
1614 static int count_csum_range(struct btrfs_root *root, u64 start,
1615                             u64 len, u64 *found)
1616 {
1617         struct btrfs_key key;
1618         struct btrfs_path path;
1619         struct extent_buffer *leaf;
1620         int ret;
1621         size_t size;
1622         *found = 0;
1623         u64 csum_end;
1624         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
1625
1626         btrfs_init_path(&path);
1627
1628         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1629         key.offset = start;
1630         key.type = BTRFS_EXTENT_CSUM_KEY;
1631
1632         ret = btrfs_search_slot(NULL, root->fs_info->csum_root,
1633                                 &key, &path, 0, 0);
1634         if (ret < 0)
1635                 goto out;
1636         if (ret > 0 && path.slots[0] > 0) {
1637                 leaf = path.nodes[0];
1638                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0] - 1);
1639                 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
1640                     key.type == BTRFS_EXTENT_CSUM_KEY)
1641                         path.slots[0]--;
1642         }
1643
1644         while (len > 0) {
1645                 leaf = path.nodes[0];
1646                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1647                         ret = btrfs_next_leaf(root->fs_info->csum_root, &path);
1648                         if (ret > 0)
1649                                 break;
1650                         else if (ret < 0)
1651                                 goto out;
1652                         leaf = path.nodes[0];
1653                 }
1654
1655                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1656                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1657                     key.type != BTRFS_EXTENT_CSUM_KEY)
1658                         break;
1659
1660                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1661                 if (key.offset >= start + len)
1662                         break;
1663
1664                 if (key.offset > start)
1665                         start = key.offset;
1666
1667                 size = btrfs_item_size_nr(leaf, path.slots[0]);
1668                 csum_end = key.offset + (size / csum_size) * root->sectorsize;
1669                 if (csum_end > start) {
1670                         size = min(csum_end - start, len);
1671                         len -= size;
1672                         start += size;
1673                         *found += size;
1674                 }
1675
1676                 path.slots[0]++;
1677         }
1678 out:
1679         btrfs_release_path(&path);
1680         if (ret < 0)
1681                 return ret;
1682         return 0;
1683 }
1684
1685 static int process_file_extent(struct btrfs_root *root,
1686                                 struct extent_buffer *eb,
1687                                 int slot, struct btrfs_key *key,
1688                                 struct shared_node *active_node)
1689 {
1690         struct inode_record *rec;
1691         struct btrfs_file_extent_item *fi;
1692         u64 num_bytes = 0;
1693         u64 disk_bytenr = 0;
1694         u64 extent_offset = 0;
1695         u64 mask = root->sectorsize - 1;
1696         int extent_type;
1697         int ret;
1698
1699         rec = active_node->current;
1700         BUG_ON(rec->ino != key->objectid || rec->refs > 1);
1701         rec->found_file_extent = 1;
1702
1703         if (rec->extent_start == (u64)-1) {
1704                 rec->extent_start = key->offset;
1705                 rec->extent_end = key->offset;
1706         }
1707
1708         if (rec->extent_end > key->offset)
1709                 rec->errors |= I_ERR_FILE_EXTENT_OVERLAP;
1710         else if (rec->extent_end < key->offset) {
1711                 ret = add_file_extent_hole(&rec->holes, rec->extent_end,
1712                                            key->offset - rec->extent_end);
1713                 if (ret < 0)
1714                         return ret;
1715         }
1716
1717         fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
1718         extent_type = btrfs_file_extent_type(eb, fi);
1719
1720         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1721                 num_bytes = btrfs_file_extent_inline_len(eb, slot, fi);
1722                 if (num_bytes == 0)
1723                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1724                 rec->found_size += num_bytes;
1725                 num_bytes = (num_bytes + mask) & ~mask;
1726         } else if (extent_type == BTRFS_FILE_EXTENT_REG ||
1727                    extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1728                 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1729                 disk_bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1730                 extent_offset = btrfs_file_extent_offset(eb, fi);
1731                 if (num_bytes == 0 || (num_bytes & mask))
1732                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1733                 if (num_bytes + extent_offset >
1734                     btrfs_file_extent_ram_bytes(eb, fi))
1735                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1736                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC &&
1737                     (btrfs_file_extent_compression(eb, fi) ||
1738                      btrfs_file_extent_encryption(eb, fi) ||
1739                      btrfs_file_extent_other_encoding(eb, fi)))
1740                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1741                 if (disk_bytenr > 0)
1742                         rec->found_size += num_bytes;
1743         } else {
1744                 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1745         }
1746         rec->extent_end = key->offset + num_bytes;
1747
1748         /*
1749          * The data reloc tree will copy full extents into its inode and then
1750          * copy the corresponding csums.  Because the extent it copied could be
1751          * a preallocated extent that hasn't been written to yet there may be no
1752          * csums to copy, ergo we won't have csums for our file extent.  This is
1753          * ok so just don't bother checking csums if the inode belongs to the
1754          * data reloc tree.
1755          */
1756         if (disk_bytenr > 0 &&
1757             btrfs_header_owner(eb) != BTRFS_DATA_RELOC_TREE_OBJECTID) {
1758                 u64 found;
1759                 if (btrfs_file_extent_compression(eb, fi))
1760                         num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1761                 else
1762                         disk_bytenr += extent_offset;
1763
1764                 ret = count_csum_range(root, disk_bytenr, num_bytes, &found);
1765                 if (ret < 0)
1766                         return ret;
1767                 if (extent_type == BTRFS_FILE_EXTENT_REG) {
1768                         if (found > 0)
1769                                 rec->found_csum_item = 1;
1770                         if (found < num_bytes)
1771                                 rec->some_csum_missing = 1;
1772                 } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1773                         if (found > 0)
1774                                 rec->errors |= I_ERR_ODD_CSUM_ITEM;
1775                 }
1776         }
1777         return 0;
1778 }
1779
1780 static int process_one_leaf(struct btrfs_root *root, struct extent_buffer *eb,
1781                             struct walk_control *wc)
1782 {
1783         struct btrfs_key key;
1784         u32 nritems;
1785         int i;
1786         int ret = 0;
1787         struct cache_tree *inode_cache;
1788         struct shared_node *active_node;
1789
1790         if (wc->root_level == wc->active_node &&
1791             btrfs_root_refs(&root->root_item) == 0)
1792                 return 0;
1793
1794         active_node = wc->nodes[wc->active_node];
1795         inode_cache = &active_node->inode_cache;
1796         nritems = btrfs_header_nritems(eb);
1797         for (i = 0; i < nritems; i++) {
1798                 btrfs_item_key_to_cpu(eb, &key, i);
1799
1800                 if (key.objectid == BTRFS_FREE_SPACE_OBJECTID)
1801                         continue;
1802                 if (key.type == BTRFS_ORPHAN_ITEM_KEY)
1803                         continue;
1804
1805                 if (active_node->current == NULL ||
1806                     active_node->current->ino < key.objectid) {
1807                         if (active_node->current) {
1808                                 active_node->current->checked = 1;
1809                                 maybe_free_inode_rec(inode_cache,
1810                                                      active_node->current);
1811                         }
1812                         active_node->current = get_inode_rec(inode_cache,
1813                                                              key.objectid, 1);
1814                         BUG_ON(IS_ERR(active_node->current));
1815                 }
1816                 switch (key.type) {
1817                 case BTRFS_DIR_ITEM_KEY:
1818                 case BTRFS_DIR_INDEX_KEY:
1819                         ret = process_dir_item(root, eb, i, &key, active_node);
1820                         break;
1821                 case BTRFS_INODE_REF_KEY:
1822                         ret = process_inode_ref(eb, i, &key, active_node);
1823                         break;
1824                 case BTRFS_INODE_EXTREF_KEY:
1825                         ret = process_inode_extref(eb, i, &key, active_node);
1826                         break;
1827                 case BTRFS_INODE_ITEM_KEY:
1828                         ret = process_inode_item(eb, i, &key, active_node);
1829                         break;
1830                 case BTRFS_EXTENT_DATA_KEY:
1831                         ret = process_file_extent(root, eb, i, &key,
1832                                                   active_node);
1833                         break;
1834                 default:
1835                         break;
1836                 };
1837         }
1838         return ret;
1839 }
1840
1841 static void reada_walk_down(struct btrfs_root *root,
1842                             struct extent_buffer *node, int slot)
1843 {
1844         u64 bytenr;
1845         u64 ptr_gen;
1846         u32 nritems;
1847         u32 blocksize;
1848         int i;
1849         int level;
1850
1851         level = btrfs_header_level(node);
1852         if (level != 1)
1853                 return;
1854
1855         nritems = btrfs_header_nritems(node);
1856         blocksize = root->nodesize;
1857         for (i = slot; i < nritems; i++) {
1858                 bytenr = btrfs_node_blockptr(node, i);
1859                 ptr_gen = btrfs_node_ptr_generation(node, i);
1860                 readahead_tree_block(root, bytenr, blocksize, ptr_gen);
1861         }
1862 }
1863
1864 /*
1865  * Check the child node/leaf by the following condition:
1866  * 1. the first item key of the node/leaf should be the same with the one
1867  *    in parent.
1868  * 2. block in parent node should match the child node/leaf.
1869  * 3. generation of parent node and child's header should be consistent.
1870  *
1871  * Or the child node/leaf pointed by the key in parent is not valid.
1872  *
1873  * We hope to check leaf owner too, but since subvol may share leaves,
1874  * which makes leaf owner check not so strong, key check should be
1875  * sufficient enough for that case.
1876  */
1877 static int check_child_node(struct btrfs_root *root,
1878                             struct extent_buffer *parent, int slot,
1879                             struct extent_buffer *child)
1880 {
1881         struct btrfs_key parent_key;
1882         struct btrfs_key child_key;
1883         int ret = 0;
1884
1885         btrfs_node_key_to_cpu(parent, &parent_key, slot);
1886         if (btrfs_header_level(child) == 0)
1887                 btrfs_item_key_to_cpu(child, &child_key, 0);
1888         else
1889                 btrfs_node_key_to_cpu(child, &child_key, 0);
1890
1891         if (memcmp(&parent_key, &child_key, sizeof(parent_key))) {
1892                 ret = -EINVAL;
1893                 fprintf(stderr,
1894                         "Wrong key of child node/leaf, wanted: (%llu, %u, %llu), have: (%llu, %u, %llu)\n",
1895                         parent_key.objectid, parent_key.type, parent_key.offset,
1896                         child_key.objectid, child_key.type, child_key.offset);
1897         }
1898         if (btrfs_header_bytenr(child) != btrfs_node_blockptr(parent, slot)) {
1899                 ret = -EINVAL;
1900                 fprintf(stderr, "Wrong block of child node/leaf, wanted: %llu, have: %llu\n",
1901                         btrfs_node_blockptr(parent, slot),
1902                         btrfs_header_bytenr(child));
1903         }
1904         if (btrfs_node_ptr_generation(parent, slot) !=
1905             btrfs_header_generation(child)) {
1906                 ret = -EINVAL;
1907                 fprintf(stderr, "Wrong generation of child node/leaf, wanted: %llu, have: %llu\n",
1908                         btrfs_header_generation(child),
1909                         btrfs_node_ptr_generation(parent, slot));
1910         }
1911         return ret;
1912 }
1913
1914 struct node_refs {
1915         u64 bytenr[BTRFS_MAX_LEVEL];
1916         u64 refs[BTRFS_MAX_LEVEL];
1917 };
1918
1919 static int walk_down_tree(struct btrfs_root *root, struct btrfs_path *path,
1920                           struct walk_control *wc, int *level,
1921                           struct node_refs *nrefs)
1922 {
1923         enum btrfs_tree_block_status status;
1924         u64 bytenr;
1925         u64 ptr_gen;
1926         struct extent_buffer *next;
1927         struct extent_buffer *cur;
1928         u32 blocksize;
1929         int ret, err = 0;
1930         u64 refs;
1931
1932         WARN_ON(*level < 0);
1933         WARN_ON(*level >= BTRFS_MAX_LEVEL);
1934
1935         if (path->nodes[*level]->start == nrefs->bytenr[*level]) {
1936                 refs = nrefs->refs[*level];
1937                 ret = 0;
1938         } else {
1939                 ret = btrfs_lookup_extent_info(NULL, root,
1940                                        path->nodes[*level]->start,
1941                                        *level, 1, &refs, NULL);
1942                 if (ret < 0) {
1943                         err = ret;
1944                         goto out;
1945                 }
1946                 nrefs->bytenr[*level] = path->nodes[*level]->start;
1947                 nrefs->refs[*level] = refs;
1948         }
1949
1950         if (refs > 1) {
1951                 ret = enter_shared_node(root, path->nodes[*level]->start,
1952                                         refs, wc, *level);
1953                 if (ret > 0) {
1954                         err = ret;
1955                         goto out;
1956                 }
1957         }
1958
1959         while (*level >= 0) {
1960                 WARN_ON(*level < 0);
1961                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1962                 cur = path->nodes[*level];
1963
1964                 if (btrfs_header_level(cur) != *level)
1965                         WARN_ON(1);
1966
1967                 if (path->slots[*level] >= btrfs_header_nritems(cur))
1968                         break;
1969                 if (*level == 0) {
1970                         ret = process_one_leaf(root, cur, wc);
1971                         if (ret < 0)
1972                                 err = ret;
1973                         break;
1974                 }
1975                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1976                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1977                 blocksize = root->nodesize;
1978
1979                 if (bytenr == nrefs->bytenr[*level - 1]) {
1980                         refs = nrefs->refs[*level - 1];
1981                 } else {
1982                         ret = btrfs_lookup_extent_info(NULL, root, bytenr,
1983                                         *level - 1, 1, &refs, NULL);
1984                         if (ret < 0) {
1985                                 refs = 0;
1986                         } else {
1987                                 nrefs->bytenr[*level - 1] = bytenr;
1988                                 nrefs->refs[*level - 1] = refs;
1989                         }
1990                 }
1991
1992                 if (refs > 1) {
1993                         ret = enter_shared_node(root, bytenr, refs,
1994                                                 wc, *level - 1);
1995                         if (ret > 0) {
1996                                 path->slots[*level]++;
1997                                 continue;
1998                         }
1999                 }
2000
2001                 next = btrfs_find_tree_block(root, bytenr, blocksize);
2002                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2003                         free_extent_buffer(next);
2004                         reada_walk_down(root, cur, path->slots[*level]);
2005                         next = read_tree_block(root, bytenr, blocksize,
2006                                                ptr_gen);
2007                         if (!extent_buffer_uptodate(next)) {
2008                                 struct btrfs_key node_key;
2009
2010                                 btrfs_node_key_to_cpu(path->nodes[*level],
2011                                                       &node_key,
2012                                                       path->slots[*level]);
2013                                 btrfs_add_corrupt_extent_record(root->fs_info,
2014                                                 &node_key,
2015                                                 path->nodes[*level]->start,
2016                                                 root->nodesize, *level);
2017                                 err = -EIO;
2018                                 goto out;
2019                         }
2020                 }
2021
2022                 ret = check_child_node(root, cur, path->slots[*level], next);
2023                 if (ret) {
2024                         err = ret;
2025                         goto out;
2026                 }
2027
2028                 if (btrfs_is_leaf(next))
2029                         status = btrfs_check_leaf(root, NULL, next);
2030                 else
2031                         status = btrfs_check_node(root, NULL, next);
2032                 if (status != BTRFS_TREE_BLOCK_CLEAN) {
2033                         free_extent_buffer(next);
2034                         err = -EIO;
2035                         goto out;
2036                 }
2037
2038                 *level = *level - 1;
2039                 free_extent_buffer(path->nodes[*level]);
2040                 path->nodes[*level] = next;
2041                 path->slots[*level] = 0;
2042         }
2043 out:
2044         path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2045         return err;
2046 }
2047
2048 static int walk_up_tree(struct btrfs_root *root, struct btrfs_path *path,
2049                         struct walk_control *wc, int *level)
2050 {
2051         int i;
2052         struct extent_buffer *leaf;
2053
2054         for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2055                 leaf = path->nodes[i];
2056                 if (path->slots[i] + 1 < btrfs_header_nritems(leaf)) {
2057                         path->slots[i]++;
2058                         *level = i;
2059                         return 0;
2060                 } else {
2061                         free_extent_buffer(path->nodes[*level]);
2062                         path->nodes[*level] = NULL;
2063                         BUG_ON(*level > wc->active_node);
2064                         if (*level == wc->active_node)
2065                                 leave_shared_node(root, wc, *level);
2066                         *level = i + 1;
2067                 }
2068         }
2069         return 1;
2070 }
2071
2072 static int check_root_dir(struct inode_record *rec)
2073 {
2074         struct inode_backref *backref;
2075         int ret = -1;
2076
2077         if (!rec->found_inode_item || rec->errors)
2078                 goto out;
2079         if (rec->nlink != 1 || rec->found_link != 0)
2080                 goto out;
2081         if (list_empty(&rec->backrefs))
2082                 goto out;
2083         backref = to_inode_backref(rec->backrefs.next);
2084         if (!backref->found_inode_ref)
2085                 goto out;
2086         if (backref->index != 0 || backref->namelen != 2 ||
2087             memcmp(backref->name, "..", 2))
2088                 goto out;
2089         if (backref->found_dir_index || backref->found_dir_item)
2090                 goto out;
2091         ret = 0;
2092 out:
2093         return ret;
2094 }
2095
2096 static int repair_inode_isize(struct btrfs_trans_handle *trans,
2097                               struct btrfs_root *root, struct btrfs_path *path,
2098                               struct inode_record *rec)
2099 {
2100         struct btrfs_inode_item *ei;
2101         struct btrfs_key key;
2102         int ret;
2103
2104         key.objectid = rec->ino;
2105         key.type = BTRFS_INODE_ITEM_KEY;
2106         key.offset = (u64)-1;
2107
2108         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2109         if (ret < 0)
2110                 goto out;
2111         if (ret) {
2112                 if (!path->slots[0]) {
2113                         ret = -ENOENT;
2114                         goto out;
2115                 }
2116                 path->slots[0]--;
2117                 ret = 0;
2118         }
2119         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2120         if (key.objectid != rec->ino) {
2121                 ret = -ENOENT;
2122                 goto out;
2123         }
2124
2125         ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
2126                             struct btrfs_inode_item);
2127         btrfs_set_inode_size(path->nodes[0], ei, rec->found_size);
2128         btrfs_mark_buffer_dirty(path->nodes[0]);
2129         rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
2130         printf("reset isize for dir %Lu root %Lu\n", rec->ino,
2131                root->root_key.objectid);
2132 out:
2133         btrfs_release_path(path);
2134         return ret;
2135 }
2136
2137 static int repair_inode_orphan_item(struct btrfs_trans_handle *trans,
2138                                     struct btrfs_root *root,
2139                                     struct btrfs_path *path,
2140                                     struct inode_record *rec)
2141 {
2142         int ret;
2143
2144         ret = btrfs_add_orphan_item(trans, root, path, rec->ino);
2145         btrfs_release_path(path);
2146         if (!ret)
2147                 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
2148         return ret;
2149 }
2150
2151 static int repair_inode_nbytes(struct btrfs_trans_handle *trans,
2152                                struct btrfs_root *root,
2153                                struct btrfs_path *path,
2154                                struct inode_record *rec)
2155 {
2156         struct btrfs_inode_item *ei;
2157         struct btrfs_key key;
2158         int ret = 0;
2159
2160         key.objectid = rec->ino;
2161         key.type = BTRFS_INODE_ITEM_KEY;
2162         key.offset = 0;
2163
2164         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2165         if (ret) {
2166                 if (ret > 0)
2167                         ret = -ENOENT;
2168                 goto out;
2169         }
2170
2171         /* Since ret == 0, no need to check anything */
2172         ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
2173                             struct btrfs_inode_item);
2174         btrfs_set_inode_nbytes(path->nodes[0], ei, rec->found_size);
2175         btrfs_mark_buffer_dirty(path->nodes[0]);
2176         rec->errors &= ~I_ERR_FILE_NBYTES_WRONG;
2177         printf("reset nbytes for ino %llu root %llu\n",
2178                rec->ino, root->root_key.objectid);
2179 out:
2180         btrfs_release_path(path);
2181         return ret;
2182 }
2183
2184 static int add_missing_dir_index(struct btrfs_root *root,
2185                                  struct cache_tree *inode_cache,
2186                                  struct inode_record *rec,
2187                                  struct inode_backref *backref)
2188 {
2189         struct btrfs_path path;
2190         struct btrfs_trans_handle *trans;
2191         struct btrfs_dir_item *dir_item;
2192         struct extent_buffer *leaf;
2193         struct btrfs_key key;
2194         struct btrfs_disk_key disk_key;
2195         struct inode_record *dir_rec;
2196         unsigned long name_ptr;
2197         u32 data_size = sizeof(*dir_item) + backref->namelen;
2198         int ret;
2199
2200         trans = btrfs_start_transaction(root, 1);
2201         if (IS_ERR(trans))
2202                 return PTR_ERR(trans);
2203
2204         fprintf(stderr, "repairing missing dir index item for inode %llu\n",
2205                 (unsigned long long)rec->ino);
2206
2207         btrfs_init_path(&path);
2208         key.objectid = backref->dir;
2209         key.type = BTRFS_DIR_INDEX_KEY;
2210         key.offset = backref->index;
2211         ret = btrfs_insert_empty_item(trans, root, &path, &key, data_size);
2212         BUG_ON(ret);
2213
2214         leaf = path.nodes[0];
2215         dir_item = btrfs_item_ptr(leaf, path.slots[0], struct btrfs_dir_item);
2216
2217         disk_key.objectid = cpu_to_le64(rec->ino);
2218         disk_key.type = BTRFS_INODE_ITEM_KEY;
2219         disk_key.offset = 0;
2220
2221         btrfs_set_dir_item_key(leaf, dir_item, &disk_key);
2222         btrfs_set_dir_type(leaf, dir_item, imode_to_type(rec->imode));
2223         btrfs_set_dir_data_len(leaf, dir_item, 0);
2224         btrfs_set_dir_name_len(leaf, dir_item, backref->namelen);
2225         name_ptr = (unsigned long)(dir_item + 1);
2226         write_extent_buffer(leaf, backref->name, name_ptr, backref->namelen);
2227         btrfs_mark_buffer_dirty(leaf);
2228         btrfs_release_path(&path);
2229         btrfs_commit_transaction(trans, root);
2230
2231         backref->found_dir_index = 1;
2232         dir_rec = get_inode_rec(inode_cache, backref->dir, 0);
2233         BUG_ON(IS_ERR(dir_rec));
2234         if (!dir_rec)
2235                 return 0;
2236         dir_rec->found_size += backref->namelen;
2237         if (dir_rec->found_size == dir_rec->isize &&
2238             (dir_rec->errors & I_ERR_DIR_ISIZE_WRONG))
2239                 dir_rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
2240         if (dir_rec->found_size != dir_rec->isize)
2241                 dir_rec->errors |= I_ERR_DIR_ISIZE_WRONG;
2242
2243         return 0;
2244 }
2245
2246 static int delete_dir_index(struct btrfs_root *root,
2247                             struct cache_tree *inode_cache,
2248                             struct inode_record *rec,
2249                             struct inode_backref *backref)
2250 {
2251         struct btrfs_trans_handle *trans;
2252         struct btrfs_dir_item *di;
2253         struct btrfs_path path;
2254         int ret = 0;
2255
2256         trans = btrfs_start_transaction(root, 1);
2257         if (IS_ERR(trans))
2258                 return PTR_ERR(trans);
2259
2260         fprintf(stderr, "Deleting bad dir index [%llu,%u,%llu] root %llu\n",
2261                 (unsigned long long)backref->dir,
2262                 BTRFS_DIR_INDEX_KEY, (unsigned long long)backref->index,
2263                 (unsigned long long)root->objectid);
2264
2265         btrfs_init_path(&path);
2266         di = btrfs_lookup_dir_index(trans, root, &path, backref->dir,
2267                                     backref->name, backref->namelen,
2268                                     backref->index, -1);
2269         if (IS_ERR(di)) {
2270                 ret = PTR_ERR(di);
2271                 btrfs_release_path(&path);
2272                 btrfs_commit_transaction(trans, root);
2273                 if (ret == -ENOENT)
2274                         return 0;
2275                 return ret;
2276         }
2277
2278         if (!di)
2279                 ret = btrfs_del_item(trans, root, &path);
2280         else
2281                 ret = btrfs_delete_one_dir_name(trans, root, &path, di);
2282         BUG_ON(ret);
2283         btrfs_release_path(&path);
2284         btrfs_commit_transaction(trans, root);
2285         return ret;
2286 }
2287
2288 static int create_inode_item(struct btrfs_root *root,
2289                              struct inode_record *rec,
2290                              struct inode_backref *backref, int root_dir)
2291 {
2292         struct btrfs_trans_handle *trans;
2293         struct btrfs_inode_item inode_item;
2294         time_t now = time(NULL);
2295         int ret;
2296
2297         trans = btrfs_start_transaction(root, 1);
2298         if (IS_ERR(trans)) {
2299                 ret = PTR_ERR(trans);
2300                 return ret;
2301         }
2302
2303         fprintf(stderr, "root %llu inode %llu recreating inode item, this may "
2304                 "be incomplete, please check permissions and content after "
2305                 "the fsck completes.\n", (unsigned long long)root->objectid,
2306                 (unsigned long long)rec->ino);
2307
2308         memset(&inode_item, 0, sizeof(inode_item));
2309         btrfs_set_stack_inode_generation(&inode_item, trans->transid);
2310         if (root_dir)
2311                 btrfs_set_stack_inode_nlink(&inode_item, 1);
2312         else
2313                 btrfs_set_stack_inode_nlink(&inode_item, rec->found_link);
2314         btrfs_set_stack_inode_nbytes(&inode_item, rec->found_size);
2315         if (rec->found_dir_item) {
2316                 if (rec->found_file_extent)
2317                         fprintf(stderr, "root %llu inode %llu has both a dir "
2318                                 "item and extents, unsure if it is a dir or a "
2319                                 "regular file so setting it as a directory\n",
2320                                 (unsigned long long)root->objectid,
2321                                 (unsigned long long)rec->ino);
2322                 btrfs_set_stack_inode_mode(&inode_item, S_IFDIR | 0755);
2323                 btrfs_set_stack_inode_size(&inode_item, rec->found_size);
2324         } else if (!rec->found_dir_item) {
2325                 btrfs_set_stack_inode_size(&inode_item, rec->extent_end);
2326                 btrfs_set_stack_inode_mode(&inode_item, S_IFREG | 0755);
2327         }
2328         btrfs_set_stack_timespec_sec(&inode_item.atime, now);
2329         btrfs_set_stack_timespec_nsec(&inode_item.atime, 0);
2330         btrfs_set_stack_timespec_sec(&inode_item.ctime, now);
2331         btrfs_set_stack_timespec_nsec(&inode_item.ctime, 0);
2332         btrfs_set_stack_timespec_sec(&inode_item.mtime, now);
2333         btrfs_set_stack_timespec_nsec(&inode_item.mtime, 0);
2334         btrfs_set_stack_timespec_sec(&inode_item.otime, 0);
2335         btrfs_set_stack_timespec_nsec(&inode_item.otime, 0);
2336
2337         ret = btrfs_insert_inode(trans, root, rec->ino, &inode_item);
2338         BUG_ON(ret);
2339         btrfs_commit_transaction(trans, root);
2340         return 0;
2341 }
2342
2343 static int repair_inode_backrefs(struct btrfs_root *root,
2344                                  struct inode_record *rec,
2345                                  struct cache_tree *inode_cache,
2346                                  int delete)
2347 {
2348         struct inode_backref *tmp, *backref;
2349         u64 root_dirid = btrfs_root_dirid(&root->root_item);
2350         int ret = 0;
2351         int repaired = 0;
2352
2353         list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
2354                 if (!delete && rec->ino == root_dirid) {
2355                         if (!rec->found_inode_item) {
2356                                 ret = create_inode_item(root, rec, backref, 1);
2357                                 if (ret)
2358                                         break;
2359                                 repaired++;
2360                         }
2361                 }
2362
2363                 /* Index 0 for root dir's are special, don't mess with it */
2364                 if (rec->ino == root_dirid && backref->index == 0)
2365                         continue;
2366
2367                 if (delete &&
2368                     ((backref->found_dir_index && !backref->found_inode_ref) ||
2369                      (backref->found_dir_index && backref->found_inode_ref &&
2370                       (backref->errors & REF_ERR_INDEX_UNMATCH)))) {
2371                         ret = delete_dir_index(root, inode_cache, rec, backref);
2372                         if (ret)
2373                                 break;
2374                         repaired++;
2375                         list_del(&backref->list);
2376                         free(backref);
2377                 }
2378
2379                 if (!delete && !backref->found_dir_index &&
2380                     backref->found_dir_item && backref->found_inode_ref) {
2381                         ret = add_missing_dir_index(root, inode_cache, rec,
2382                                                     backref);
2383                         if (ret)
2384                                 break;
2385                         repaired++;
2386                         if (backref->found_dir_item &&
2387                             backref->found_dir_index &&
2388                             backref->found_dir_index) {
2389                                 if (!backref->errors &&
2390                                     backref->found_inode_ref) {
2391                                         list_del(&backref->list);
2392                                         free(backref);
2393                                 }
2394                         }
2395                 }
2396
2397                 if (!delete && (!backref->found_dir_index &&
2398                                 !backref->found_dir_item &&
2399                                 backref->found_inode_ref)) {
2400                         struct btrfs_trans_handle *trans;
2401                         struct btrfs_key location;
2402
2403                         ret = check_dir_conflict(root, backref->name,
2404                                                  backref->namelen,
2405                                                  backref->dir,
2406                                                  backref->index);
2407                         if (ret) {
2408                                 /*
2409                                  * let nlink fixing routine to handle it,
2410                                  * which can do it better.
2411                                  */
2412                                 ret = 0;
2413                                 break;
2414                         }
2415                         location.objectid = rec->ino;
2416                         location.type = BTRFS_INODE_ITEM_KEY;
2417                         location.offset = 0;
2418
2419                         trans = btrfs_start_transaction(root, 1);
2420                         if (IS_ERR(trans)) {
2421                                 ret = PTR_ERR(trans);
2422                                 break;
2423                         }
2424                         fprintf(stderr, "adding missing dir index/item pair "
2425                                 "for inode %llu\n",
2426                                 (unsigned long long)rec->ino);
2427                         ret = btrfs_insert_dir_item(trans, root, backref->name,
2428                                                     backref->namelen,
2429                                                     backref->dir, &location,
2430                                                     imode_to_type(rec->imode),
2431                                                     backref->index);
2432                         BUG_ON(ret);
2433                         btrfs_commit_transaction(trans, root);
2434                         repaired++;
2435                 }
2436
2437                 if (!delete && (backref->found_inode_ref &&
2438                                 backref->found_dir_index &&
2439                                 backref->found_dir_item &&
2440                                 !(backref->errors & REF_ERR_INDEX_UNMATCH) &&
2441                                 !rec->found_inode_item)) {
2442                         ret = create_inode_item(root, rec, backref, 0);
2443                         if (ret)
2444                                 break;
2445                         repaired++;
2446                 }
2447
2448         }
2449         return ret ? ret : repaired;
2450 }
2451
2452 /*
2453  * To determine the file type for nlink/inode_item repair
2454  *
2455  * Return 0 if file type is found and BTRFS_FT_* is stored into type.
2456  * Return -ENOENT if file type is not found.
2457  */
2458 static int find_file_type(struct inode_record *rec, u8 *type)
2459 {
2460         struct inode_backref *backref;
2461
2462         /* For inode item recovered case */
2463         if (rec->found_inode_item) {
2464                 *type = imode_to_type(rec->imode);
2465                 return 0;
2466         }
2467
2468         list_for_each_entry(backref, &rec->backrefs, list) {
2469                 if (backref->found_dir_index || backref->found_dir_item) {
2470                         *type = backref->filetype;
2471                         return 0;
2472                 }
2473         }
2474         return -ENOENT;
2475 }
2476
2477 /*
2478  * To determine the file name for nlink repair
2479  *
2480  * Return 0 if file name is found, set name and namelen.
2481  * Return -ENOENT if file name is not found.
2482  */
2483 static int find_file_name(struct inode_record *rec,
2484                           char *name, int *namelen)
2485 {
2486         struct inode_backref *backref;
2487
2488         list_for_each_entry(backref, &rec->backrefs, list) {
2489                 if (backref->found_dir_index || backref->found_dir_item ||
2490                     backref->found_inode_ref) {
2491                         memcpy(name, backref->name, backref->namelen);
2492                         *namelen = backref->namelen;
2493                         return 0;
2494                 }
2495         }
2496         return -ENOENT;
2497 }
2498
2499 /* Reset the nlink of the inode to the correct one */
2500 static int reset_nlink(struct btrfs_trans_handle *trans,
2501                        struct btrfs_root *root,
2502                        struct btrfs_path *path,
2503                        struct inode_record *rec)
2504 {
2505         struct inode_backref *backref;
2506         struct inode_backref *tmp;
2507         struct btrfs_key key;
2508         struct btrfs_inode_item *inode_item;
2509         int ret = 0;
2510
2511         /* We don't believe this either, reset it and iterate backref */
2512         rec->found_link = 0;
2513
2514         /* Remove all backref including the valid ones */
2515         list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
2516                 ret = btrfs_unlink(trans, root, rec->ino, backref->dir,
2517                                    backref->index, backref->name,
2518                                    backref->namelen, 0);
2519                 if (ret < 0)
2520                         goto out;
2521
2522                 /* remove invalid backref, so it won't be added back */
2523                 if (!(backref->found_dir_index &&
2524                       backref->found_dir_item &&
2525                       backref->found_inode_ref)) {
2526                         list_del(&backref->list);
2527                         free(backref);
2528                 } else {
2529                         rec->found_link++;
2530                 }
2531         }
2532
2533         /* Set nlink to 0 */
2534         key.objectid = rec->ino;
2535         key.type = BTRFS_INODE_ITEM_KEY;
2536         key.offset = 0;
2537         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2538         if (ret < 0)
2539                 goto out;
2540         if (ret > 0) {
2541                 ret = -ENOENT;
2542                 goto out;
2543         }
2544         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2545                                     struct btrfs_inode_item);
2546         btrfs_set_inode_nlink(path->nodes[0], inode_item, 0);
2547         btrfs_mark_buffer_dirty(path->nodes[0]);
2548         btrfs_release_path(path);
2549
2550         /*
2551          * Add back valid inode_ref/dir_item/dir_index,
2552          * add_link() will handle the nlink inc, so new nlink must be correct
2553          */
2554         list_for_each_entry(backref, &rec->backrefs, list) {
2555                 ret = btrfs_add_link(trans, root, rec->ino, backref->dir,
2556                                      backref->name, backref->namelen,
2557                                      backref->filetype, &backref->index, 1);
2558                 if (ret < 0)
2559                         goto out;
2560         }
2561 out:
2562         btrfs_release_path(path);
2563         return ret;
2564 }
2565
2566 static int repair_inode_nlinks(struct btrfs_trans_handle *trans,
2567                                struct btrfs_root *root,
2568                                struct btrfs_path *path,
2569                                struct inode_record *rec)
2570 {
2571         char *dir_name = "lost+found";
2572         char namebuf[BTRFS_NAME_LEN] = {0};
2573         u64 lost_found_ino;
2574         u32 mode = 0700;
2575         u8 type = 0;
2576         int namelen = 0;
2577         int name_recovered = 0;
2578         int type_recovered = 0;
2579         int ret = 0;
2580
2581         /*
2582          * Get file name and type first before these invalid inode ref
2583          * are deleted by remove_all_invalid_backref()
2584          */
2585         name_recovered = !find_file_name(rec, namebuf, &namelen);
2586         type_recovered = !find_file_type(rec, &type);
2587
2588         if (!name_recovered) {
2589                 printf("Can't get file name for inode %llu, using '%llu' as fallback\n",
2590                        rec->ino, rec->ino);
2591                 namelen = count_digits(rec->ino);
2592                 sprintf(namebuf, "%llu", rec->ino);
2593                 name_recovered = 1;
2594         }
2595         if (!type_recovered) {
2596                 printf("Can't get file type for inode %llu, using FILE as fallback\n",
2597                        rec->ino);
2598                 type = BTRFS_FT_REG_FILE;
2599                 type_recovered = 1;
2600         }
2601
2602         ret = reset_nlink(trans, root, path, rec);
2603         if (ret < 0) {
2604                 fprintf(stderr,
2605                         "Failed to reset nlink for inode %llu: %s\n",
2606                         rec->ino, strerror(-ret));
2607                 goto out;
2608         }
2609
2610         if (rec->found_link == 0) {
2611                 lost_found_ino = root->highest_inode;
2612                 if (lost_found_ino >= BTRFS_LAST_FREE_OBJECTID) {
2613                         ret = -EOVERFLOW;
2614                         goto out;
2615                 }
2616                 lost_found_ino++;
2617                 ret = btrfs_mkdir(trans, root, dir_name, strlen(dir_name),
2618                                   BTRFS_FIRST_FREE_OBJECTID, &lost_found_ino,
2619                                   mode);
2620                 if (ret < 0) {
2621                         fprintf(stderr, "Failed to create '%s' dir: %s\n",
2622                                 dir_name, strerror(-ret));
2623                         goto out;
2624                 }
2625                 ret = btrfs_add_link(trans, root, rec->ino, lost_found_ino,
2626                                      namebuf, namelen, type, NULL, 1);
2627                 /*
2628                  * Add ".INO" suffix several times to handle case where
2629                  * "FILENAME.INO" is already taken by another file.
2630                  */
2631                 while (ret == -EEXIST) {
2632                         /*
2633                          * Conflicting file name, add ".INO" as suffix * +1 for '.'
2634                          */
2635                         if (namelen + count_digits(rec->ino) + 1 >
2636                             BTRFS_NAME_LEN) {
2637                                 ret = -EFBIG;
2638                                 goto out;
2639                         }
2640                         snprintf(namebuf + namelen, BTRFS_NAME_LEN - namelen,
2641                                  ".%llu", rec->ino);
2642                         namelen += count_digits(rec->ino) + 1;
2643                         ret = btrfs_add_link(trans, root, rec->ino,
2644                                              lost_found_ino, namebuf,
2645                                              namelen, type, NULL, 1);
2646                 }
2647                 if (ret < 0) {
2648                         fprintf(stderr,
2649                                 "Failed to link the inode %llu to %s dir: %s\n",
2650                                 rec->ino, dir_name, strerror(-ret));
2651                         goto out;
2652                 }
2653                 /*
2654                  * Just increase the found_link, don't actually add the
2655                  * backref. This will make things easier and this inode
2656                  * record will be freed after the repair is done.
2657                  * So fsck will not report problem about this inode.
2658                  */
2659                 rec->found_link++;
2660                 printf("Moving file '%.*s' to '%s' dir since it has no valid backref\n",
2661                        namelen, namebuf, dir_name);
2662         }
2663         printf("Fixed the nlink of inode %llu\n", rec->ino);
2664 out:
2665         /*
2666          * Clear the flag anyway, or we will loop forever for the same inode
2667          * as it will not be removed from the bad inode list and the dead loop
2668          * happens.
2669          */
2670         rec->errors &= ~I_ERR_LINK_COUNT_WRONG;
2671         btrfs_release_path(path);
2672         return ret;
2673 }
2674
2675 /*
2676  * Check if there is any normal(reg or prealloc) file extent for given
2677  * ino.
2678  * This is used to determine the file type when neither its dir_index/item or
2679  * inode_item exists.
2680  *
2681  * This will *NOT* report error, if any error happens, just consider it does
2682  * not have any normal file extent.
2683  */
2684 static int find_normal_file_extent(struct btrfs_root *root, u64 ino)
2685 {
2686         struct btrfs_path path;
2687         struct btrfs_key key;
2688         struct btrfs_key found_key;
2689         struct btrfs_file_extent_item *fi;
2690         u8 type;
2691         int ret = 0;
2692
2693         btrfs_init_path(&path);
2694         key.objectid = ino;
2695         key.type = BTRFS_EXTENT_DATA_KEY;
2696         key.offset = 0;
2697
2698         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
2699         if (ret < 0) {
2700                 ret = 0;
2701                 goto out;
2702         }
2703         if (ret && path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
2704                 ret = btrfs_next_leaf(root, &path);
2705                 if (ret) {
2706                         ret = 0;
2707                         goto out;
2708                 }
2709         }
2710         while (1) {
2711                 btrfs_item_key_to_cpu(path.nodes[0], &found_key,
2712                                       path.slots[0]);
2713                 if (found_key.objectid != ino ||
2714                     found_key.type != BTRFS_EXTENT_DATA_KEY)
2715                         break;
2716                 fi = btrfs_item_ptr(path.nodes[0], path.slots[0],
2717                                     struct btrfs_file_extent_item);
2718                 type = btrfs_file_extent_type(path.nodes[0], fi);
2719                 if (type != BTRFS_FILE_EXTENT_INLINE) {
2720                         ret = 1;
2721                         goto out;
2722                 }
2723         }
2724 out:
2725         btrfs_release_path(&path);
2726         return ret;
2727 }
2728
2729 static u32 btrfs_type_to_imode(u8 type)
2730 {
2731         static u32 imode_by_btrfs_type[] = {
2732                 [BTRFS_FT_REG_FILE]     = S_IFREG,
2733                 [BTRFS_FT_DIR]          = S_IFDIR,
2734                 [BTRFS_FT_CHRDEV]       = S_IFCHR,
2735                 [BTRFS_FT_BLKDEV]       = S_IFBLK,
2736                 [BTRFS_FT_FIFO]         = S_IFIFO,
2737                 [BTRFS_FT_SOCK]         = S_IFSOCK,
2738                 [BTRFS_FT_SYMLINK]      = S_IFLNK,
2739         };
2740
2741         return imode_by_btrfs_type[(type)];
2742 }
2743
2744 static int repair_inode_no_item(struct btrfs_trans_handle *trans,
2745                                 struct btrfs_root *root,
2746                                 struct btrfs_path *path,
2747                                 struct inode_record *rec)
2748 {
2749         u8 filetype;
2750         u32 mode = 0700;
2751         int type_recovered = 0;
2752         int ret = 0;
2753
2754         printf("Trying to rebuild inode:%llu\n", rec->ino);
2755
2756         type_recovered = !find_file_type(rec, &filetype);
2757
2758         /*
2759          * Try to determine inode type if type not found.
2760          *
2761          * For found regular file extent, it must be FILE.
2762          * For found dir_item/index, it must be DIR.
2763          *
2764          * For undetermined one, use FILE as fallback.
2765          *
2766          * TODO:
2767          * 1. If found backref(inode_index/item is already handled) to it,
2768          *    it must be DIR.
2769          *    Need new inode-inode ref structure to allow search for that.
2770          */
2771         if (!type_recovered) {
2772                 if (rec->found_file_extent &&
2773                     find_normal_file_extent(root, rec->ino)) {
2774                         type_recovered = 1;
2775                         filetype = BTRFS_FT_REG_FILE;
2776                 } else if (rec->found_dir_item) {
2777                         type_recovered = 1;
2778                         filetype = BTRFS_FT_DIR;
2779                 } else if (!list_empty(&rec->orphan_extents)) {
2780                         type_recovered = 1;
2781                         filetype = BTRFS_FT_REG_FILE;
2782                 } else{
2783                         printf("Can't determine the filetype for inode %llu, assume it is a normal file\n",
2784                                rec->ino);
2785                         type_recovered = 1;
2786                         filetype = BTRFS_FT_REG_FILE;
2787                 }
2788         }
2789
2790         ret = btrfs_new_inode(trans, root, rec->ino,
2791                               mode | btrfs_type_to_imode(filetype));
2792         if (ret < 0)
2793                 goto out;
2794
2795         /*
2796          * Here inode rebuild is done, we only rebuild the inode item,
2797          * don't repair the nlink(like move to lost+found).
2798          * That is the job of nlink repair.
2799          *
2800          * We just fill the record and return
2801          */
2802         rec->found_dir_item = 1;
2803         rec->imode = mode | btrfs_type_to_imode(filetype);
2804         rec->nlink = 0;
2805         rec->errors &= ~I_ERR_NO_INODE_ITEM;
2806         /* Ensure the inode_nlinks repair function will be called */
2807         rec->errors |= I_ERR_LINK_COUNT_WRONG;
2808 out:
2809         return ret;
2810 }
2811
2812 static int repair_inode_orphan_extent(struct btrfs_trans_handle *trans,
2813                                       struct btrfs_root *root,
2814                                       struct btrfs_path *path,
2815                                       struct inode_record *rec)
2816 {
2817         struct orphan_data_extent *orphan;
2818         struct orphan_data_extent *tmp;
2819         int ret = 0;
2820
2821         list_for_each_entry_safe(orphan, tmp, &rec->orphan_extents, list) {
2822                 /*
2823                  * Check for conflicting file extents
2824                  *
2825                  * Here we don't know whether the extents is compressed or not,
2826                  * so we can only assume it not compressed nor data offset,
2827                  * and use its disk_len as extent length.
2828                  */
2829                 ret = btrfs_get_extent(NULL, root, path, orphan->objectid,
2830                                        orphan->offset, orphan->disk_len, 0);
2831                 btrfs_release_path(path);
2832                 if (ret < 0)
2833                         goto out;
2834                 if (!ret) {
2835                         fprintf(stderr,
2836                                 "orphan extent (%llu, %llu) conflicts, delete the orphan\n",
2837                                 orphan->disk_bytenr, orphan->disk_len);
2838                         ret = btrfs_free_extent(trans,
2839                                         root->fs_info->extent_root,
2840                                         orphan->disk_bytenr, orphan->disk_len,
2841                                         0, root->objectid, orphan->objectid,
2842                                         orphan->offset);
2843                         if (ret < 0)
2844                                 goto out;
2845                 }
2846                 ret = btrfs_insert_file_extent(trans, root, orphan->objectid,
2847                                 orphan->offset, orphan->disk_bytenr,
2848                                 orphan->disk_len, orphan->disk_len);
2849                 if (ret < 0)
2850                         goto out;
2851
2852                 /* Update file size info */
2853                 rec->found_size += orphan->disk_len;
2854                 if (rec->found_size == rec->nbytes)
2855                         rec->errors &= ~I_ERR_FILE_NBYTES_WRONG;
2856
2857                 /* Update the file extent hole info too */
2858                 ret = del_file_extent_hole(&rec->holes, orphan->offset,
2859                                            orphan->disk_len);
2860                 if (ret < 0)
2861                         goto out;
2862                 if (RB_EMPTY_ROOT(&rec->holes))
2863                         rec->errors &= ~I_ERR_FILE_EXTENT_DISCOUNT;
2864
2865                 list_del(&orphan->list);
2866                 free(orphan);
2867         }
2868         rec->errors &= ~I_ERR_FILE_EXTENT_ORPHAN;
2869 out:
2870         return ret;
2871 }
2872
2873 static int repair_inode_discount_extent(struct btrfs_trans_handle *trans,
2874                                         struct btrfs_root *root,
2875                                         struct btrfs_path *path,
2876                                         struct inode_record *rec)
2877 {
2878         struct rb_node *node;
2879         struct file_extent_hole *hole;
2880         int found = 0;
2881         int ret = 0;
2882
2883         node = rb_first(&rec->holes);
2884
2885         while (node) {
2886                 found = 1;
2887                 hole = rb_entry(node, struct file_extent_hole, node);
2888                 ret = btrfs_punch_hole(trans, root, rec->ino,
2889                                        hole->start, hole->len);
2890                 if (ret < 0)
2891                         goto out;
2892                 ret = del_file_extent_hole(&rec->holes, hole->start,
2893                                            hole->len);
2894                 if (ret < 0)
2895                         goto out;
2896                 if (RB_EMPTY_ROOT(&rec->holes))
2897                         rec->errors &= ~I_ERR_FILE_EXTENT_DISCOUNT;
2898                 node = rb_first(&rec->holes);
2899         }
2900         /* special case for a file losing all its file extent */
2901         if (!found) {
2902                 ret = btrfs_punch_hole(trans, root, rec->ino, 0,
2903                                        round_up(rec->isize, root->sectorsize));
2904                 if (ret < 0)
2905                         goto out;
2906         }
2907         printf("Fixed discount file extents for inode: %llu in root: %llu\n",
2908                rec->ino, root->objectid);
2909 out:
2910         return ret;
2911 }
2912
2913 static int try_repair_inode(struct btrfs_root *root, struct inode_record *rec)
2914 {
2915         struct btrfs_trans_handle *trans;
2916         struct btrfs_path path;
2917         int ret = 0;
2918
2919         if (!(rec->errors & (I_ERR_DIR_ISIZE_WRONG |
2920                              I_ERR_NO_ORPHAN_ITEM |
2921                              I_ERR_LINK_COUNT_WRONG |
2922                              I_ERR_NO_INODE_ITEM |
2923                              I_ERR_FILE_EXTENT_ORPHAN |
2924                              I_ERR_FILE_EXTENT_DISCOUNT|
2925                              I_ERR_FILE_NBYTES_WRONG)))
2926                 return rec->errors;
2927
2928         /*
2929          * For nlink repair, it may create a dir and add link, so
2930          * 2 for parent(256)'s dir_index and dir_item
2931          * 2 for lost+found dir's inode_item and inode_ref
2932          * 1 for the new inode_ref of the file
2933          * 2 for lost+found dir's dir_index and dir_item for the file
2934          */
2935         trans = btrfs_start_transaction(root, 7);
2936         if (IS_ERR(trans))
2937                 return PTR_ERR(trans);
2938
2939         btrfs_init_path(&path);
2940         if (rec->errors & I_ERR_NO_INODE_ITEM)
2941                 ret = repair_inode_no_item(trans, root, &path, rec);
2942         if (!ret && rec->errors & I_ERR_FILE_EXTENT_ORPHAN)
2943                 ret = repair_inode_orphan_extent(trans, root, &path, rec);
2944         if (!ret && rec->errors & I_ERR_FILE_EXTENT_DISCOUNT)
2945                 ret = repair_inode_discount_extent(trans, root, &path, rec);
2946         if (!ret && rec->errors & I_ERR_DIR_ISIZE_WRONG)
2947                 ret = repair_inode_isize(trans, root, &path, rec);
2948         if (!ret && rec->errors & I_ERR_NO_ORPHAN_ITEM)
2949                 ret = repair_inode_orphan_item(trans, root, &path, rec);
2950         if (!ret && rec->errors & I_ERR_LINK_COUNT_WRONG)
2951                 ret = repair_inode_nlinks(trans, root, &path, rec);
2952         if (!ret && rec->errors & I_ERR_FILE_NBYTES_WRONG)
2953                 ret = repair_inode_nbytes(trans, root, &path, rec);
2954         btrfs_commit_transaction(trans, root);
2955         btrfs_release_path(&path);
2956         return ret;
2957 }
2958
2959 static int check_inode_recs(struct btrfs_root *root,
2960                             struct cache_tree *inode_cache)
2961 {
2962         struct cache_extent *cache;
2963         struct ptr_node *node;
2964         struct inode_record *rec;
2965         struct inode_backref *backref;
2966         int stage = 0;
2967         int ret = 0;
2968         int err = 0;
2969         u64 error = 0;
2970         u64 root_dirid = btrfs_root_dirid(&root->root_item);
2971
2972         if (btrfs_root_refs(&root->root_item) == 0) {
2973                 if (!cache_tree_empty(inode_cache))
2974                         fprintf(stderr, "warning line %d\n", __LINE__);
2975                 return 0;
2976         }
2977
2978         /*
2979          * We need to record the highest inode number for later 'lost+found'
2980          * dir creation.
2981          * We must select an ino not used/referred by any existing inode, or
2982          * 'lost+found' ino may be a missing ino in a corrupted leaf,
2983          * this may cause 'lost+found' dir has wrong nlinks.
2984          */
2985         cache = last_cache_extent(inode_cache);
2986         if (cache) {
2987                 node = container_of(cache, struct ptr_node, cache);
2988                 rec = node->data;
2989                 if (rec->ino > root->highest_inode)
2990                         root->highest_inode = rec->ino;
2991         }
2992
2993         /*
2994          * We need to repair backrefs first because we could change some of the
2995          * errors in the inode recs.
2996          *
2997          * We also need to go through and delete invalid backrefs first and then
2998          * add the correct ones second.  We do this because we may get EEXIST
2999          * when adding back the correct index because we hadn't yet deleted the
3000          * invalid index.
3001          *
3002          * For example, if we were missing a dir index then the directories
3003          * isize would be wrong, so if we fixed the isize to what we thought it
3004          * would be and then fixed the backref we'd still have a invalid fs, so
3005          * we need to add back the dir index and then check to see if the isize
3006          * is still wrong.
3007          */
3008         while (stage < 3) {
3009                 stage++;
3010                 if (stage == 3 && !err)
3011                         break;
3012
3013                 cache = search_cache_extent(inode_cache, 0);
3014                 while (repair && cache) {
3015                         node = container_of(cache, struct ptr_node, cache);
3016                         rec = node->data;
3017                         cache = next_cache_extent(cache);
3018
3019                         /* Need to free everything up and rescan */
3020                         if (stage == 3) {
3021                                 remove_cache_extent(inode_cache, &node->cache);
3022                                 free(node);
3023                                 free_inode_rec(rec);
3024                                 continue;
3025                         }
3026
3027                         if (list_empty(&rec->backrefs))
3028                                 continue;
3029
3030                         ret = repair_inode_backrefs(root, rec, inode_cache,
3031                                                     stage == 1);
3032                         if (ret < 0) {
3033                                 err = ret;
3034                                 stage = 2;
3035                                 break;
3036                         } if (ret > 0) {
3037                                 err = -EAGAIN;
3038                         }
3039                 }
3040         }
3041         if (err)
3042                 return err;
3043
3044         rec = get_inode_rec(inode_cache, root_dirid, 0);
3045         BUG_ON(IS_ERR(rec));
3046         if (rec) {
3047                 ret = check_root_dir(rec);
3048                 if (ret) {
3049                         fprintf(stderr, "root %llu root dir %llu error\n",
3050                                 (unsigned long long)root->root_key.objectid,
3051                                 (unsigned long long)root_dirid);
3052                         print_inode_error(root, rec);
3053                         error++;
3054                 }
3055         } else {
3056                 if (repair) {
3057                         struct btrfs_trans_handle *trans;
3058
3059                         trans = btrfs_start_transaction(root, 1);
3060                         if (IS_ERR(trans)) {
3061                                 err = PTR_ERR(trans);
3062                                 return err;
3063                         }
3064
3065                         fprintf(stderr,
3066                                 "root %llu missing its root dir, recreating\n",
3067                                 (unsigned long long)root->objectid);
3068
3069                         ret = btrfs_make_root_dir(trans, root, root_dirid);
3070                         BUG_ON(ret);
3071
3072                         btrfs_commit_transaction(trans, root);
3073                         return -EAGAIN;
3074                 }
3075
3076                 fprintf(stderr, "root %llu root dir %llu not found\n",
3077                         (unsigned long long)root->root_key.objectid,
3078                         (unsigned long long)root_dirid);
3079         }
3080
3081         while (1) {
3082                 cache = search_cache_extent(inode_cache, 0);
3083                 if (!cache)
3084                         break;
3085                 node = container_of(cache, struct ptr_node, cache);
3086                 rec = node->data;
3087                 remove_cache_extent(inode_cache, &node->cache);
3088                 free(node);
3089                 if (rec->ino == root_dirid ||
3090                     rec->ino == BTRFS_ORPHAN_OBJECTID) {
3091                         free_inode_rec(rec);
3092                         continue;
3093                 }
3094
3095                 if (rec->errors & I_ERR_NO_ORPHAN_ITEM) {
3096                         ret = check_orphan_item(root, rec->ino);
3097                         if (ret == 0)
3098                                 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
3099                         if (can_free_inode_rec(rec)) {
3100                                 free_inode_rec(rec);
3101                                 continue;
3102                         }
3103                 }
3104
3105                 if (!rec->found_inode_item)
3106                         rec->errors |= I_ERR_NO_INODE_ITEM;
3107                 if (rec->found_link != rec->nlink)
3108                         rec->errors |= I_ERR_LINK_COUNT_WRONG;
3109                 if (repair) {
3110                         ret = try_repair_inode(root, rec);
3111                         if (ret == 0 && can_free_inode_rec(rec)) {
3112                                 free_inode_rec(rec);
3113                                 continue;
3114                         }
3115                         ret = 0;
3116                 }
3117
3118                 if (!(repair && ret == 0))
3119                         error++;
3120                 print_inode_error(root, rec);
3121                 list_for_each_entry(backref, &rec->backrefs, list) {
3122                         if (!backref->found_dir_item)
3123                                 backref->errors |= REF_ERR_NO_DIR_ITEM;
3124                         if (!backref->found_dir_index)
3125                                 backref->errors |= REF_ERR_NO_DIR_INDEX;
3126                         if (!backref->found_inode_ref)
3127                                 backref->errors |= REF_ERR_NO_INODE_REF;
3128                         fprintf(stderr, "\tunresolved ref dir %llu index %llu"
3129                                 " namelen %u name %s filetype %d errors %x",
3130                                 (unsigned long long)backref->dir,
3131                                 (unsigned long long)backref->index,
3132                                 backref->namelen, backref->name,
3133                                 backref->filetype, backref->errors);
3134                         print_ref_error(backref->errors);
3135                 }
3136                 free_inode_rec(rec);
3137         }
3138         return (error > 0) ? -1 : 0;
3139 }
3140
3141 static struct root_record *get_root_rec(struct cache_tree *root_cache,
3142                                         u64 objectid)
3143 {
3144         struct cache_extent *cache;
3145         struct root_record *rec = NULL;
3146         int ret;
3147
3148         cache = lookup_cache_extent(root_cache, objectid, 1);
3149         if (cache) {
3150                 rec = container_of(cache, struct root_record, cache);
3151         } else {
3152                 rec = calloc(1, sizeof(*rec));
3153                 if (!rec)
3154                         return ERR_PTR(-ENOMEM);
3155                 rec->objectid = objectid;
3156                 INIT_LIST_HEAD(&rec->backrefs);
3157                 rec->cache.start = objectid;
3158                 rec->cache.size = 1;
3159
3160                 ret = insert_cache_extent(root_cache, &rec->cache);
3161                 if (ret)
3162                         return ERR_PTR(-EEXIST);
3163         }
3164         return rec;
3165 }
3166
3167 static struct root_backref *get_root_backref(struct root_record *rec,
3168                                              u64 ref_root, u64 dir, u64 index,
3169                                              const char *name, int namelen)
3170 {
3171         struct root_backref *backref;
3172
3173         list_for_each_entry(backref, &rec->backrefs, list) {
3174                 if (backref->ref_root != ref_root || backref->dir != dir ||
3175                     backref->namelen != namelen)
3176                         continue;
3177                 if (memcmp(name, backref->name, namelen))
3178                         continue;
3179                 return backref;
3180         }
3181
3182         backref = calloc(1, sizeof(*backref) + namelen + 1);
3183         if (!backref)
3184                 return NULL;
3185         backref->ref_root = ref_root;
3186         backref->dir = dir;
3187         backref->index = index;
3188         backref->namelen = namelen;
3189         memcpy(backref->name, name, namelen);
3190         backref->name[namelen] = '\0';
3191         list_add_tail(&backref->list, &rec->backrefs);
3192         return backref;
3193 }
3194
3195 static void free_root_record(struct cache_extent *cache)
3196 {
3197         struct root_record *rec;
3198         struct root_backref *backref;
3199
3200         rec = container_of(cache, struct root_record, cache);
3201         while (!list_empty(&rec->backrefs)) {
3202                 backref = to_root_backref(rec->backrefs.next);
3203                 list_del(&backref->list);
3204                 free(backref);
3205         }
3206
3207         free(rec);
3208 }
3209
3210 FREE_EXTENT_CACHE_BASED_TREE(root_recs, free_root_record);
3211
3212 static int add_root_backref(struct cache_tree *root_cache,
3213                             u64 root_id, u64 ref_root, u64 dir, u64 index,
3214                             const char *name, int namelen,
3215                             int item_type, int errors)
3216 {
3217         struct root_record *rec;
3218         struct root_backref *backref;
3219
3220         rec = get_root_rec(root_cache, root_id);
3221         BUG_ON(IS_ERR(rec));
3222         backref = get_root_backref(rec, ref_root, dir, index, name, namelen);
3223         BUG_ON(!backref);
3224
3225         backref->errors |= errors;
3226
3227         if (item_type != BTRFS_DIR_ITEM_KEY) {
3228                 if (backref->found_dir_index || backref->found_back_ref ||
3229                     backref->found_forward_ref) {
3230                         if (backref->index != index)
3231                                 backref->errors |= REF_ERR_INDEX_UNMATCH;
3232                 } else {
3233                         backref->index = index;
3234                 }
3235         }
3236
3237         if (item_type == BTRFS_DIR_ITEM_KEY) {
3238                 if (backref->found_forward_ref)
3239                         rec->found_ref++;
3240                 backref->found_dir_item = 1;
3241         } else if (item_type == BTRFS_DIR_INDEX_KEY) {
3242                 backref->found_dir_index = 1;
3243         } else if (item_type == BTRFS_ROOT_REF_KEY) {
3244                 if (backref->found_forward_ref)
3245                         backref->errors |= REF_ERR_DUP_ROOT_REF;
3246                 else if (backref->found_dir_item)
3247                         rec->found_ref++;
3248                 backref->found_forward_ref = 1;
3249         } else if (item_type == BTRFS_ROOT_BACKREF_KEY) {
3250                 if (backref->found_back_ref)
3251                         backref->errors |= REF_ERR_DUP_ROOT_BACKREF;
3252                 backref->found_back_ref = 1;
3253         } else {
3254                 BUG_ON(1);
3255         }
3256
3257         if (backref->found_forward_ref && backref->found_dir_item)
3258                 backref->reachable = 1;
3259         return 0;
3260 }
3261
3262 static int merge_root_recs(struct btrfs_root *root,
3263                            struct cache_tree *src_cache,
3264                            struct cache_tree *dst_cache)
3265 {
3266         struct cache_extent *cache;
3267         struct ptr_node *node;
3268         struct inode_record *rec;
3269         struct inode_backref *backref;
3270         int ret = 0;
3271
3272         if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
3273                 free_inode_recs_tree(src_cache);
3274                 return 0;
3275         }
3276
3277         while (1) {
3278                 cache = search_cache_extent(src_cache, 0);
3279                 if (!cache)
3280                         break;
3281                 node = container_of(cache, struct ptr_node, cache);
3282                 rec = node->data;
3283                 remove_cache_extent(src_cache, &node->cache);
3284                 free(node);
3285
3286                 ret = is_child_root(root, root->objectid, rec->ino);
3287                 if (ret < 0)
3288                         break;
3289                 else if (ret == 0)
3290                         goto skip;
3291
3292                 list_for_each_entry(backref, &rec->backrefs, list) {
3293                         BUG_ON(backref->found_inode_ref);
3294                         if (backref->found_dir_item)
3295                                 add_root_backref(dst_cache, rec->ino,
3296                                         root->root_key.objectid, backref->dir,
3297                                         backref->index, backref->name,
3298                                         backref->namelen, BTRFS_DIR_ITEM_KEY,
3299                                         backref->errors);
3300                         if (backref->found_dir_index)
3301                                 add_root_backref(dst_cache, rec->ino,
3302                                         root->root_key.objectid, backref->dir,
3303                                         backref->index, backref->name,
3304                                         backref->namelen, BTRFS_DIR_INDEX_KEY,
3305                                         backref->errors);
3306                 }
3307 skip:
3308                 free_inode_rec(rec);
3309         }
3310         if (ret < 0)
3311                 return ret;
3312         return 0;
3313 }
3314
3315 static int check_root_refs(struct btrfs_root *root,
3316                            struct cache_tree *root_cache)
3317 {
3318         struct root_record *rec;
3319         struct root_record *ref_root;
3320         struct root_backref *backref;
3321         struct cache_extent *cache;
3322         int loop = 1;
3323         int ret;
3324         int error;
3325         int errors = 0;
3326
3327         rec = get_root_rec(root_cache, BTRFS_FS_TREE_OBJECTID);
3328         BUG_ON(IS_ERR(rec));
3329         rec->found_ref = 1;
3330
3331         /* fixme: this can not detect circular references */
3332         while (loop) {
3333                 loop = 0;
3334                 cache = search_cache_extent(root_cache, 0);
3335                 while (1) {
3336                         if (!cache)
3337                                 break;
3338                         rec = container_of(cache, struct root_record, cache);
3339                         cache = next_cache_extent(cache);
3340
3341                         if (rec->found_ref == 0)
3342                                 continue;
3343
3344                         list_for_each_entry(backref, &rec->backrefs, list) {
3345                                 if (!backref->reachable)
3346                                         continue;
3347
3348                                 ref_root = get_root_rec(root_cache,
3349                                                         backref->ref_root);
3350                                 BUG_ON(IS_ERR(ref_root));
3351                                 if (ref_root->found_ref > 0)
3352                                         continue;
3353
3354                                 backref->reachable = 0;
3355                                 rec->found_ref--;
3356                                 if (rec->found_ref == 0)
3357                                         loop = 1;
3358                         }
3359                 }
3360         }
3361
3362         cache = search_cache_extent(root_cache, 0);
3363         while (1) {
3364                 if (!cache)
3365                         break;
3366                 rec = container_of(cache, struct root_record, cache);
3367                 cache = next_cache_extent(cache);
3368
3369                 if (rec->found_ref == 0 &&
3370                     rec->objectid >= BTRFS_FIRST_FREE_OBJECTID &&
3371                     rec->objectid <= BTRFS_LAST_FREE_OBJECTID) {
3372                         ret = check_orphan_item(root->fs_info->tree_root,
3373                                                 rec->objectid);
3374                         if (ret == 0)
3375                                 continue;
3376
3377                         /*
3378                          * If we don't have a root item then we likely just have
3379                          * a dir item in a snapshot for this root but no actual
3380                          * ref key or anything so it's meaningless.
3381                          */
3382                         if (!rec->found_root_item)
3383                                 continue;
3384                         errors++;
3385                         fprintf(stderr, "fs tree %llu not referenced\n",
3386                                 (unsigned long long)rec->objectid);
3387                 }
3388
3389                 error = 0;
3390                 if (rec->found_ref > 0 && !rec->found_root_item)
3391                         error = 1;
3392                 list_for_each_entry(backref, &rec->backrefs, list) {
3393                         if (!backref->found_dir_item)
3394                                 backref->errors |= REF_ERR_NO_DIR_ITEM;
3395                         if (!backref->found_dir_index)
3396                                 backref->errors |= REF_ERR_NO_DIR_INDEX;
3397                         if (!backref->found_back_ref)
3398                                 backref->errors |= REF_ERR_NO_ROOT_BACKREF;
3399                         if (!backref->found_forward_ref)
3400                                 backref->errors |= REF_ERR_NO_ROOT_REF;
3401                         if (backref->reachable && backref->errors)
3402                                 error = 1;
3403                 }
3404                 if (!error)
3405                         continue;
3406
3407                 errors++;
3408                 fprintf(stderr, "fs tree %llu refs %u %s\n",
3409                         (unsigned long long)rec->objectid, rec->found_ref,
3410                          rec->found_root_item ? "" : "not found");
3411
3412                 list_for_each_entry(backref, &rec->backrefs, list) {
3413                         if (!backref->reachable)
3414                                 continue;
3415                         if (!backref->errors && rec->found_root_item)
3416                                 continue;
3417                         fprintf(stderr, "\tunresolved ref root %llu dir %llu"
3418                                 " index %llu namelen %u name %s errors %x\n",
3419                                 (unsigned long long)backref->ref_root,
3420                                 (unsigned long long)backref->dir,
3421                                 (unsigned long long)backref->index,
3422                                 backref->namelen, backref->name,
3423                                 backref->errors);
3424                         print_ref_error(backref->errors);
3425                 }
3426         }
3427         return errors > 0 ? 1 : 0;
3428 }
3429
3430 static int process_root_ref(struct extent_buffer *eb, int slot,
3431                             struct btrfs_key *key,
3432                             struct cache_tree *root_cache)
3433 {
3434         u64 dirid;
3435         u64 index;
3436         u32 len;
3437         u32 name_len;
3438         struct btrfs_root_ref *ref;
3439         char namebuf[BTRFS_NAME_LEN];
3440         int error;
3441
3442         ref = btrfs_item_ptr(eb, slot, struct btrfs_root_ref);
3443
3444         dirid = btrfs_root_ref_dirid(eb, ref);
3445         index = btrfs_root_ref_sequence(eb, ref);
3446         name_len = btrfs_root_ref_name_len(eb, ref);
3447
3448         if (name_len <= BTRFS_NAME_LEN) {
3449                 len = name_len;
3450                 error = 0;
3451         } else {
3452                 len = BTRFS_NAME_LEN;
3453                 error = REF_ERR_NAME_TOO_LONG;
3454         }
3455         read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
3456
3457         if (key->type == BTRFS_ROOT_REF_KEY) {
3458                 add_root_backref(root_cache, key->offset, key->objectid, dirid,
3459                                  index, namebuf, len, key->type, error);
3460         } else {
3461                 add_root_backref(root_cache, key->objectid, key->offset, dirid,
3462                                  index, namebuf, len, key->type, error);
3463         }
3464         return 0;
3465 }
3466
3467 static void free_corrupt_block(struct cache_extent *cache)
3468 {
3469         struct btrfs_corrupt_block *corrupt;
3470
3471         corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
3472         free(corrupt);
3473 }
3474
3475 FREE_EXTENT_CACHE_BASED_TREE(corrupt_blocks, free_corrupt_block);
3476
3477 /*
3478  * Repair the btree of the given root.
3479  *
3480  * The fix is to remove the node key in corrupt_blocks cache_tree.
3481  * and rebalance the tree.
3482  * After the fix, the btree should be writeable.
3483  */
3484 static int repair_btree(struct btrfs_root *root,
3485                         struct cache_tree *corrupt_blocks)
3486 {
3487         struct btrfs_trans_handle *trans;
3488         struct btrfs_path path;
3489         struct btrfs_corrupt_block *corrupt;
3490         struct cache_extent *cache;
3491         struct btrfs_key key;
3492         u64 offset;
3493         int level;
3494         int ret = 0;
3495
3496         if (cache_tree_empty(corrupt_blocks))
3497                 return 0;
3498
3499         trans = btrfs_start_transaction(root, 1);
3500         if (IS_ERR(trans)) {
3501                 ret = PTR_ERR(trans);
3502                 fprintf(stderr, "Error starting transaction: %s\n",
3503                         strerror(-ret));
3504                 return ret;
3505         }
3506         btrfs_init_path(&path);
3507         cache = first_cache_extent(corrupt_blocks);
3508         while (cache) {
3509                 corrupt = container_of(cache, struct btrfs_corrupt_block,
3510                                        cache);
3511                 level = corrupt->level;
3512                 path.lowest_level = level;
3513                 key.objectid = corrupt->key.objectid;
3514                 key.type = corrupt->key.type;
3515                 key.offset = corrupt->key.offset;
3516
3517                 /*
3518                  * Here we don't want to do any tree balance, since it may
3519                  * cause a balance with corrupted brother leaf/node,
3520                  * so ins_len set to 0 here.
3521                  * Balance will be done after all corrupt node/leaf is deleted.
3522                  */
3523                 ret = btrfs_search_slot(trans, root, &key, &path, 0, 1);
3524                 if (ret < 0)
3525                         goto out;
3526                 offset = btrfs_node_blockptr(path.nodes[level],
3527                                              path.slots[level]);
3528
3529                 /* Remove the ptr */
3530                 ret = btrfs_del_ptr(trans, root, &path, level,
3531                                     path.slots[level]);
3532                 if (ret < 0)
3533                         goto out;
3534                 /*
3535                  * Remove the corresponding extent
3536                  * return value is not concerned.
3537                  */
3538                 btrfs_release_path(&path);
3539                 ret = btrfs_free_extent(trans, root, offset, root->nodesize,
3540                                         0, root->root_key.objectid,
3541                                         level - 1, 0);
3542                 cache = next_cache_extent(cache);
3543         }
3544
3545         /* Balance the btree using btrfs_search_slot() */
3546         cache = first_cache_extent(corrupt_blocks);
3547         while (cache) {
3548                 corrupt = container_of(cache, struct btrfs_corrupt_block,
3549                                        cache);
3550                 memcpy(&key, &corrupt->key, sizeof(key));
3551                 ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
3552                 if (ret < 0)
3553                         goto out;
3554                 /* return will always >0 since it won't find the item */
3555                 ret = 0;
3556                 btrfs_release_path(&path);
3557                 cache = next_cache_extent(cache);
3558         }
3559 out:
3560         btrfs_commit_transaction(trans, root);
3561         btrfs_release_path(&path);
3562         return ret;
3563 }
3564
3565 static int check_fs_root(struct btrfs_root *root,
3566                          struct cache_tree *root_cache,
3567                          struct walk_control *wc)
3568 {
3569         int ret = 0;
3570         int err = 0;
3571         int wret;
3572         int level;
3573         struct btrfs_path path;
3574         struct shared_node root_node;
3575         struct root_record *rec;
3576         struct btrfs_root_item *root_item = &root->root_item;
3577         struct cache_tree corrupt_blocks;
3578         struct orphan_data_extent *orphan;
3579         struct orphan_data_extent *tmp;
3580         enum btrfs_tree_block_status status;
3581         struct node_refs nrefs;
3582
3583         /*
3584          * Reuse the corrupt_block cache tree to record corrupted tree block
3585          *
3586          * Unlike the usage in extent tree check, here we do it in a per
3587          * fs/subvol tree base.
3588          */
3589         cache_tree_init(&corrupt_blocks);
3590         root->fs_info->corrupt_blocks = &corrupt_blocks;
3591
3592         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
3593                 rec = get_root_rec(root_cache, root->root_key.objectid);
3594                 BUG_ON(IS_ERR(rec));
3595                 if (btrfs_root_refs(root_item) > 0)
3596                         rec->found_root_item = 1;
3597         }
3598
3599         btrfs_init_path(&path);
3600         memset(&root_node, 0, sizeof(root_node));
3601         cache_tree_init(&root_node.root_cache);
3602         cache_tree_init(&root_node.inode_cache);
3603         memset(&nrefs, 0, sizeof(nrefs));
3604
3605         /* Move the orphan extent record to corresponding inode_record */
3606         list_for_each_entry_safe(orphan, tmp,
3607                                  &root->orphan_data_extents, list) {
3608                 struct inode_record *inode;
3609
3610                 inode = get_inode_rec(&root_node.inode_cache, orphan->objectid,
3611                                       1);
3612                 BUG_ON(IS_ERR(inode));
3613                 inode->errors |= I_ERR_FILE_EXTENT_ORPHAN;
3614                 list_move(&orphan->list, &inode->orphan_extents);
3615         }
3616
3617         level = btrfs_header_level(root->node);
3618         memset(wc->nodes, 0, sizeof(wc->nodes));
3619         wc->nodes[level] = &root_node;
3620         wc->active_node = level;
3621         wc->root_level = level;
3622
3623         /* We may not have checked the root block, lets do that now */
3624         if (btrfs_is_leaf(root->node))
3625                 status = btrfs_check_leaf(root, NULL, root->node);
3626         else
3627                 status = btrfs_check_node(root, NULL, root->node);
3628         if (status != BTRFS_TREE_BLOCK_CLEAN)
3629                 return -EIO;
3630
3631         if (btrfs_root_refs(root_item) > 0 ||
3632             btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3633                 path.nodes[level] = root->node;
3634                 extent_buffer_get(root->node);
3635                 path.slots[level] = 0;
3636         } else {
3637                 struct btrfs_key key;
3638                 struct btrfs_disk_key found_key;
3639
3640                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3641                 level = root_item->drop_level;
3642                 path.lowest_level = level;
3643                 if (level > btrfs_header_level(root->node) ||
3644                     level >= BTRFS_MAX_LEVEL) {
3645                         error("ignoring invalid drop level: %u", level);
3646                         goto skip_walking;
3647                 }
3648                 wret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
3649                 if (wret < 0)
3650                         goto skip_walking;
3651                 btrfs_node_key(path.nodes[level], &found_key,
3652                                 path.slots[level]);
3653                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3654                                         sizeof(found_key)));
3655         }
3656
3657         while (1) {
3658                 wret = walk_down_tree(root, &path, wc, &level, &nrefs);
3659                 if (wret < 0)
3660                         ret = wret;
3661                 if (wret != 0)
3662                         break;
3663
3664                 wret = walk_up_tree(root, &path, wc, &level);
3665                 if (wret < 0)
3666                         ret = wret;
3667                 if (wret != 0)
3668                         break;
3669         }
3670 skip_walking:
3671         btrfs_release_path(&path);
3672
3673         if (!cache_tree_empty(&corrupt_blocks)) {
3674                 struct cache_extent *cache;
3675                 struct btrfs_corrupt_block *corrupt;
3676
3677                 printf("The following tree block(s) is corrupted in tree %llu:\n",
3678                        root->root_key.objectid);
3679                 cache = first_cache_extent(&corrupt_blocks);
3680                 while (cache) {
3681                         corrupt = container_of(cache,
3682                                                struct btrfs_corrupt_block,
3683                                                cache);
3684                         printf("\ttree block bytenr: %llu, level: %d, node key: (%llu, %u, %llu)\n",
3685                                cache->start, corrupt->level,
3686                                corrupt->key.objectid, corrupt->key.type,
3687                                corrupt->key.offset);
3688                         cache = next_cache_extent(cache);
3689                 }
3690                 if (repair) {
3691                         printf("Try to repair the btree for root %llu\n",
3692                                root->root_key.objectid);
3693                         ret = repair_btree(root, &corrupt_blocks);
3694                         if (ret < 0)
3695                                 fprintf(stderr, "Failed to repair btree: %s\n",
3696                                         strerror(-ret));
3697                         if (!ret)
3698                                 printf("Btree for root %llu is fixed\n",
3699                                        root->root_key.objectid);
3700                 }
3701         }
3702
3703         err = merge_root_recs(root, &root_node.root_cache, root_cache);
3704         if (err < 0)
3705                 ret = err;
3706
3707         if (root_node.current) {
3708                 root_node.current->checked = 1;
3709                 maybe_free_inode_rec(&root_node.inode_cache,
3710                                 root_node.current);
3711         }
3712
3713         err = check_inode_recs(root, &root_node.inode_cache);
3714         if (!ret)
3715                 ret = err;
3716
3717         free_corrupt_blocks_tree(&corrupt_blocks);
3718         root->fs_info->corrupt_blocks = NULL;
3719         free_orphan_data_extents(&root->orphan_data_extents);
3720         return ret;
3721 }
3722
3723 static int fs_root_objectid(u64 objectid)
3724 {
3725         if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
3726             objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3727                 return 1;
3728         return is_fstree(objectid);
3729 }
3730
3731 static int check_fs_roots(struct btrfs_root *root,
3732                           struct cache_tree *root_cache)
3733 {
3734         struct btrfs_path path;
3735         struct btrfs_key key;
3736         struct walk_control wc;
3737         struct extent_buffer *leaf, *tree_node;
3738         struct btrfs_root *tmp_root;
3739         struct btrfs_root *tree_root = root->fs_info->tree_root;
3740         int ret;
3741         int err = 0;
3742
3743         if (ctx.progress_enabled) {
3744                 ctx.tp = TASK_FS_ROOTS;
3745                 task_start(ctx.info);
3746         }
3747
3748         /*
3749          * Just in case we made any changes to the extent tree that weren't
3750          * reflected into the free space cache yet.
3751          */
3752         if (repair)
3753                 reset_cached_block_groups(root->fs_info);
3754         memset(&wc, 0, sizeof(wc));
3755         cache_tree_init(&wc.shared);
3756         btrfs_init_path(&path);
3757
3758 again:
3759         key.offset = 0;
3760         key.objectid = 0;
3761         key.type = BTRFS_ROOT_ITEM_KEY;
3762         ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
3763         if (ret < 0) {
3764                 err = 1;
3765                 goto out;
3766         }
3767         tree_node = tree_root->node;
3768         while (1) {
3769                 if (tree_node != tree_root->node) {
3770                         free_root_recs_tree(root_cache);
3771                         btrfs_release_path(&path);
3772                         goto again;
3773                 }
3774                 leaf = path.nodes[0];
3775                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
3776                         ret = btrfs_next_leaf(tree_root, &path);
3777                         if (ret) {
3778                                 if (ret < 0)
3779                                         err = 1;
3780                                 break;
3781                         }
3782                         leaf = path.nodes[0];
3783                 }
3784                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
3785                 if (key.type == BTRFS_ROOT_ITEM_KEY &&
3786                     fs_root_objectid(key.objectid)) {
3787                         if (key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
3788                                 tmp_root = btrfs_read_fs_root_no_cache(
3789                                                 root->fs_info, &key);
3790                         } else {
3791                                 key.offset = (u64)-1;
3792                                 tmp_root = btrfs_read_fs_root(
3793                                                 root->fs_info, &key);
3794                         }
3795                         if (IS_ERR(tmp_root)) {
3796                                 err = 1;
3797                                 goto next;
3798                         }
3799                         ret = check_fs_root(tmp_root, root_cache, &wc);
3800                         if (ret == -EAGAIN) {
3801                                 free_root_recs_tree(root_cache);
3802                                 btrfs_release_path(&path);
3803                                 goto again;
3804                         }
3805                         if (ret)
3806                                 err = 1;
3807                         if (key.objectid == BTRFS_TREE_RELOC_OBJECTID)
3808                                 btrfs_free_fs_root(tmp_root);
3809                 } else if (key.type == BTRFS_ROOT_REF_KEY ||
3810                            key.type == BTRFS_ROOT_BACKREF_KEY) {
3811                         process_root_ref(leaf, path.slots[0], &key,
3812                                          root_cache);
3813                 }
3814 next:
3815                 path.slots[0]++;
3816         }
3817 out:
3818         btrfs_release_path(&path);
3819         if (err)
3820                 free_extent_cache_tree(&wc.shared);
3821         if (!cache_tree_empty(&wc.shared))
3822                 fprintf(stderr, "warning line %d\n", __LINE__);
3823
3824         task_stop(ctx.info);
3825
3826         return err;
3827 }
3828
3829 static int all_backpointers_checked(struct extent_record *rec, int print_errs)
3830 {
3831         struct list_head *cur = rec->backrefs.next;
3832         struct extent_backref *back;
3833         struct tree_backref *tback;
3834         struct data_backref *dback;
3835         u64 found = 0;
3836         int err = 0;
3837
3838         while(cur != &rec->backrefs) {
3839                 back = to_extent_backref(cur);
3840                 cur = cur->next;
3841                 if (!back->found_extent_tree) {
3842                         err = 1;
3843                         if (!print_errs)
3844                                 goto out;
3845                         if (back->is_data) {
3846                                 dback = to_data_backref(back);
3847                                 fprintf(stderr, "Backref %llu %s %llu"
3848                                         " owner %llu offset %llu num_refs %lu"
3849                                         " not found in extent tree\n",
3850                                         (unsigned long long)rec->start,
3851                                         back->full_backref ?
3852                                         "parent" : "root",
3853                                         back->full_backref ?
3854                                         (unsigned long long)dback->parent:
3855                                         (unsigned long long)dback->root,
3856                                         (unsigned long long)dback->owner,
3857                                         (unsigned long long)dback->offset,
3858                                         (unsigned long)dback->num_refs);
3859                         } else {
3860                                 tback = to_tree_backref(back);
3861                                 fprintf(stderr, "Backref %llu parent %llu"
3862                                         " root %llu not found in extent tree\n",
3863                                         (unsigned long long)rec->start,
3864                                         (unsigned long long)tback->parent,
3865                                         (unsigned long long)tback->root);
3866                         }
3867                 }
3868                 if (!back->is_data && !back->found_ref) {
3869                         err = 1;
3870                         if (!print_errs)
3871                                 goto out;
3872                         tback = to_tree_backref(back);
3873                         fprintf(stderr, "Backref %llu %s %llu not referenced back %p\n",
3874                                 (unsigned long long)rec->start,
3875                                 back->full_backref ? "parent" : "root",
3876                                 back->full_backref ?
3877                                 (unsigned long long)tback->parent :
3878                                 (unsigned long long)tback->root, back);
3879                 }
3880                 if (back->is_data) {
3881                         dback = to_data_backref(back);
3882                         if (dback->found_ref != dback->num_refs) {
3883                                 err = 1;
3884                                 if (!print_errs)
3885                                         goto out;
3886                                 fprintf(stderr, "Incorrect local backref count"
3887                                         " on %llu %s %llu owner %llu"
3888                                         " offset %llu found %u wanted %u back %p\n",
3889                                         (unsigned long long)rec->start,
3890                                         back->full_backref ?
3891                                         "parent" : "root",
3892                                         back->full_backref ?
3893                                         (unsigned long long)dback->parent:
3894                                         (unsigned long long)dback->root,
3895                                         (unsigned long long)dback->owner,
3896                                         (unsigned long long)dback->offset,
3897                                         dback->found_ref, dback->num_refs, back);
3898                         }
3899                         if (dback->disk_bytenr != rec->start) {
3900                                 err = 1;
3901                                 if (!print_errs)
3902                                         goto out;
3903                                 fprintf(stderr, "Backref disk bytenr does not"
3904                                         " match extent record, bytenr=%llu, "
3905                                         "ref bytenr=%llu\n",
3906                                         (unsigned long long)rec->start,
3907                                         (unsigned long long)dback->disk_bytenr);
3908                         }
3909
3910                         if (dback->bytes != rec->nr) {
3911                                 err = 1;
3912                                 if (!print_errs)
3913                                         goto out;
3914                                 fprintf(stderr, "Backref bytes do not match "
3915                                         "extent backref, bytenr=%llu, ref "
3916                                         "bytes=%llu, backref bytes=%llu\n",
3917                                         (unsigned long long)rec->start,
3918                                         (unsigned long long)rec->nr,
3919                                         (unsigned long long)dback->bytes);
3920                         }
3921                 }
3922                 if (!back->is_data) {
3923                         found += 1;
3924                 } else {
3925                         dback = to_data_backref(back);
3926                         found += dback->found_ref;
3927                 }
3928         }
3929         if (found != rec->refs) {
3930                 err = 1;
3931                 if (!print_errs)
3932                         goto out;
3933                 fprintf(stderr, "Incorrect global backref count "
3934                         "on %llu found %llu wanted %llu\n",
3935                         (unsigned long long)rec->start,
3936                         (unsigned long long)found,
3937                         (unsigned long long)rec->refs);
3938         }
3939 out:
3940         return err;
3941 }
3942
3943 static int free_all_extent_backrefs(struct extent_record *rec)
3944 {
3945         struct extent_backref *back;
3946         struct list_head *cur;
3947         while (!list_empty(&rec->backrefs)) {
3948                 cur = rec->backrefs.next;
3949                 back = to_extent_backref(cur);
3950                 list_del(cur);
3951                 free(back);
3952         }
3953         return 0;
3954 }
3955
3956 static void free_extent_record_cache(struct btrfs_fs_info *fs_info,
3957                                      struct cache_tree *extent_cache)
3958 {
3959         struct cache_extent *cache;
3960         struct extent_record *rec;
3961
3962         while (1) {
3963                 cache = first_cache_extent(extent_cache);
3964                 if (!cache)
3965                         break;
3966                 rec = container_of(cache, struct extent_record, cache);
3967                 remove_cache_extent(extent_cache, cache);
3968                 free_all_extent_backrefs(rec);
3969                 free(rec);
3970         }
3971 }
3972
3973 static int maybe_free_extent_rec(struct cache_tree *extent_cache,
3974                                  struct extent_record *rec)
3975 {
3976         if (rec->content_checked && rec->owner_ref_checked &&
3977             rec->extent_item_refs == rec->refs && rec->refs > 0 &&
3978             rec->num_duplicates == 0 && !all_backpointers_checked(rec, 0) &&
3979             !rec->bad_full_backref && !rec->crossing_stripes &&
3980             !rec->wrong_chunk_type) {
3981                 remove_cache_extent(extent_cache, &rec->cache);
3982                 free_all_extent_backrefs(rec);
3983                 list_del_init(&rec->list);
3984                 free(rec);
3985         }
3986         return 0;
3987 }
3988
3989 static int check_owner_ref(struct btrfs_root *root,
3990                             struct extent_record *rec,
3991                             struct extent_buffer *buf)
3992 {
3993         struct extent_backref *node;
3994         struct tree_backref *back;
3995         struct btrfs_root *ref_root;
3996         struct btrfs_key key;
3997         struct btrfs_path path;
3998         struct extent_buffer *parent;
3999         int level;
4000         int found = 0;
4001         int ret;
4002
4003         list_for_each_entry(node, &rec->backrefs, list) {
4004                 if (node->is_data)
4005                         continue;
4006                 if (!node->found_ref)
4007                         continue;
4008                 if (node->full_backref)
4009                         continue;
4010                 back = to_tree_backref(node);
4011                 if (btrfs_header_owner(buf) == back->root)
4012                         return 0;
4013         }
4014         BUG_ON(rec->is_root);
4015
4016         /* try to find the block by search corresponding fs tree */
4017         key.objectid = btrfs_header_owner(buf);
4018         key.type = BTRFS_ROOT_ITEM_KEY;
4019         key.offset = (u64)-1;
4020
4021         ref_root = btrfs_read_fs_root(root->fs_info, &key);
4022         if (IS_ERR(ref_root))
4023                 return 1;
4024
4025         level = btrfs_header_level(buf);
4026         if (level == 0)
4027                 btrfs_item_key_to_cpu(buf, &key, 0);
4028         else
4029                 btrfs_node_key_to_cpu(buf, &key, 0);
4030
4031         btrfs_init_path(&path);
4032         path.lowest_level = level + 1;
4033         ret = btrfs_search_slot(NULL, ref_root, &key, &path, 0, 0);
4034         if (ret < 0)
4035                 return 0;
4036
4037         parent = path.nodes[level + 1];
4038         if (parent && buf->start == btrfs_node_blockptr(parent,
4039                                                         path.slots[level + 1]))
4040                 found = 1;
4041
4042         btrfs_release_path(&path);
4043         return found ? 0 : 1;
4044 }
4045
4046 static int is_extent_tree_record(struct extent_record *rec)
4047 {
4048         struct list_head *cur = rec->backrefs.next;
4049         struct extent_backref *node;
4050         struct tree_backref *back;
4051         int is_extent = 0;
4052
4053         while(cur != &rec->backrefs) {
4054                 node = to_extent_backref(cur);
4055                 cur = cur->next;
4056                 if (node->is_data)
4057                         return 0;
4058                 back = to_tree_backref(node);
4059                 if (node->full_backref)
4060                         return 0;
4061                 if (back->root == BTRFS_EXTENT_TREE_OBJECTID)
4062                         is_extent = 1;
4063         }
4064         return is_extent;
4065 }
4066
4067
4068 static int record_bad_block_io(struct btrfs_fs_info *info,
4069                                struct cache_tree *extent_cache,
4070                                u64 start, u64 len)
4071 {
4072         struct extent_record *rec;
4073         struct cache_extent *cache;
4074         struct btrfs_key key;
4075
4076         cache = lookup_cache_extent(extent_cache, start, len);
4077         if (!cache)
4078                 return 0;
4079
4080         rec = container_of(cache, struct extent_record, cache);
4081         if (!is_extent_tree_record(rec))
4082                 return 0;
4083
4084         btrfs_disk_key_to_cpu(&key, &rec->parent_key);
4085         return btrfs_add_corrupt_extent_record(info, &key, start, len, 0);
4086 }
4087
4088 static int swap_values(struct btrfs_root *root, struct btrfs_path *path,
4089                        struct extent_buffer *buf, int slot)
4090 {
4091         if (btrfs_header_level(buf)) {
4092                 struct btrfs_key_ptr ptr1, ptr2;
4093
4094                 read_extent_buffer(buf, &ptr1, btrfs_node_key_ptr_offset(slot),
4095                                    sizeof(struct btrfs_key_ptr));
4096                 read_extent_buffer(buf, &ptr2,
4097                                    btrfs_node_key_ptr_offset(slot + 1),
4098                                    sizeof(struct btrfs_key_ptr));
4099                 write_extent_buffer(buf, &ptr1,
4100                                     btrfs_node_key_ptr_offset(slot + 1),
4101                                     sizeof(struct btrfs_key_ptr));
4102                 write_extent_buffer(buf, &ptr2,
4103                                     btrfs_node_key_ptr_offset(slot),
4104                                     sizeof(struct btrfs_key_ptr));
4105                 if (slot == 0) {
4106                         struct btrfs_disk_key key;
4107                         btrfs_node_key(buf, &key, 0);
4108                         btrfs_fixup_low_keys(root, path, &key,
4109                                              btrfs_header_level(buf) + 1);
4110                 }
4111         } else {
4112                 struct btrfs_item *item1, *item2;
4113                 struct btrfs_key k1, k2;
4114                 char *item1_data, *item2_data;
4115                 u32 item1_offset, item2_offset, item1_size, item2_size;
4116
4117                 item1 = btrfs_item_nr(slot);
4118                 item2 = btrfs_item_nr(slot + 1);
4119                 btrfs_item_key_to_cpu(buf, &k1, slot);
4120                 btrfs_item_key_to_cpu(buf, &k2, slot + 1);
4121                 item1_offset = btrfs_item_offset(buf, item1);
4122                 item2_offset = btrfs_item_offset(buf, item2);
4123                 item1_size = btrfs_item_size(buf, item1);
4124                 item2_size = btrfs_item_size(buf, item2);
4125
4126                 item1_data = malloc(item1_size);
4127                 if (!item1_data)
4128                         return -ENOMEM;
4129                 item2_data = malloc(item2_size);
4130                 if (!item2_data) {
4131                         free(item1_data);
4132                         return -ENOMEM;
4133                 }
4134
4135                 read_extent_buffer(buf, item1_data, item1_offset, item1_size);
4136                 read_extent_buffer(buf, item2_data, item2_offset, item2_size);
4137
4138                 write_extent_buffer(buf, item1_data, item2_offset, item2_size);
4139                 write_extent_buffer(buf, item2_data, item1_offset, item1_size);
4140                 free(item1_data);
4141                 free(item2_data);
4142
4143                 btrfs_set_item_offset(buf, item1, item2_offset);
4144                 btrfs_set_item_offset(buf, item2, item1_offset);
4145                 btrfs_set_item_size(buf, item1, item2_size);
4146                 btrfs_set_item_size(buf, item2, item1_size);
4147
4148                 path->slots[0] = slot;
4149                 btrfs_set_item_key_unsafe(root, path, &k2);
4150                 path->slots[0] = slot + 1;
4151                 btrfs_set_item_key_unsafe(root, path, &k1);
4152         }
4153         return 0;
4154 }
4155
4156 static int fix_key_order(struct btrfs_trans_handle *trans,
4157                          struct btrfs_root *root,
4158                          struct btrfs_path *path)
4159 {
4160         struct extent_buffer *buf;
4161         struct btrfs_key k1, k2;
4162         int i;
4163         int level = path->lowest_level;
4164         int ret = -EIO;
4165
4166         buf = path->nodes[level];
4167         for (i = 0; i < btrfs_header_nritems(buf) - 1; i++) {
4168                 if (level) {
4169                         btrfs_node_key_to_cpu(buf, &k1, i);
4170                         btrfs_node_key_to_cpu(buf, &k2, i + 1);
4171                 } else {
4172                         btrfs_item_key_to_cpu(buf, &k1, i);
4173                         btrfs_item_key_to_cpu(buf, &k2, i + 1);
4174                 }
4175                 if (btrfs_comp_cpu_keys(&k1, &k2) < 0)
4176                         continue;
4177                 ret = swap_values(root, path, buf, i);
4178                 if (ret)
4179                         break;
4180                 btrfs_mark_buffer_dirty(buf);
4181                 i = 0;
4182         }
4183         return ret;
4184 }
4185
4186 static int delete_bogus_item(struct btrfs_trans_handle *trans,
4187                              struct btrfs_root *root,
4188                              struct btrfs_path *path,
4189                              struct extent_buffer *buf, int slot)
4190 {
4191         struct btrfs_key key;
4192         int nritems = btrfs_header_nritems(buf);
4193
4194         btrfs_item_key_to_cpu(buf, &key, slot);
4195
4196         /* These are all the keys we can deal with missing. */
4197         if (key.type != BTRFS_DIR_INDEX_KEY &&
4198             key.type != BTRFS_EXTENT_ITEM_KEY &&
4199             key.type != BTRFS_METADATA_ITEM_KEY &&
4200             key.type != BTRFS_TREE_BLOCK_REF_KEY &&
4201             key.type != BTRFS_EXTENT_DATA_REF_KEY)
4202                 return -1;
4203
4204         printf("Deleting bogus item [%llu,%u,%llu] at slot %d on block %llu\n",
4205                (unsigned long long)key.objectid, key.type,
4206                (unsigned long long)key.offset, slot, buf->start);
4207         memmove_extent_buffer(buf, btrfs_item_nr_offset(slot),
4208                               btrfs_item_nr_offset(slot + 1),
4209                               sizeof(struct btrfs_item) *
4210                               (nritems - slot - 1));
4211         btrfs_set_header_nritems(buf, nritems - 1);
4212         if (slot == 0) {
4213                 struct btrfs_disk_key disk_key;
4214
4215                 btrfs_item_key(buf, &disk_key, 0);
4216                 btrfs_fixup_low_keys(root, path, &disk_key, 1);
4217         }
4218         btrfs_mark_buffer_dirty(buf);
4219         return 0;
4220 }
4221
4222 static int fix_item_offset(struct btrfs_trans_handle *trans,
4223                            struct btrfs_root *root,
4224                            struct btrfs_path *path)
4225 {
4226         struct extent_buffer *buf;
4227         int i;
4228         int ret = 0;
4229
4230         /* We should only get this for leaves */
4231         BUG_ON(path->lowest_level);
4232         buf = path->nodes[0];
4233 again:
4234         for (i = 0; i < btrfs_header_nritems(buf); i++) {
4235                 unsigned int shift = 0, offset;
4236
4237                 if (i == 0 && btrfs_item_end_nr(buf, i) !=
4238                     BTRFS_LEAF_DATA_SIZE(root)) {
4239                         if (btrfs_item_end_nr(buf, i) >
4240                             BTRFS_LEAF_DATA_SIZE(root)) {
4241                                 ret = delete_bogus_item(trans, root, path,
4242                                                         buf, i);
4243                                 if (!ret)
4244                                         goto again;
4245                                 fprintf(stderr, "item is off the end of the "
4246                                         "leaf, can't fix\n");
4247                                 ret = -EIO;
4248                                 break;
4249                         }
4250                         shift = BTRFS_LEAF_DATA_SIZE(root) -
4251                                 btrfs_item_end_nr(buf, i);
4252                 } else if (i > 0 && btrfs_item_end_nr(buf, i) !=
4253                            btrfs_item_offset_nr(buf, i - 1)) {
4254                         if (btrfs_item_end_nr(buf, i) >
4255                             btrfs_item_offset_nr(buf, i - 1)) {
4256                                 ret = delete_bogus_item(trans, root, path,
4257                                                         buf, i);
4258                                 if (!ret)
4259                                         goto again;
4260                                 fprintf(stderr, "items overlap, can't fix\n");
4261                                 ret = -EIO;
4262                                 break;
4263                         }
4264                         shift = btrfs_item_offset_nr(buf, i - 1) -
4265                                 btrfs_item_end_nr(buf, i);
4266                 }
4267                 if (!shift)
4268                         continue;
4269
4270                 printf("Shifting item nr %d by %u bytes in block %llu\n",
4271                        i, shift, (unsigned long long)buf->start);
4272                 offset = btrfs_item_offset_nr(buf, i);
4273                 memmove_extent_buffer(buf,
4274                                       btrfs_leaf_data(buf) + offset + shift,
4275                                       btrfs_leaf_data(buf) + offset,
4276                                       btrfs_item_size_nr(buf, i));
4277                 btrfs_set_item_offset(buf, btrfs_item_nr(i),
4278                                       offset + shift);
4279                 btrfs_mark_buffer_dirty(buf);
4280         }
4281
4282         /*
4283          * We may have moved things, in which case we want to exit so we don't
4284          * write those changes out.  Once we have proper abort functionality in
4285          * progs this can be changed to something nicer.
4286          */
4287         BUG_ON(ret);
4288         return ret;
4289 }
4290
4291 /*
4292  * Attempt to fix basic block failures.  If we can't fix it for whatever reason
4293  * then just return -EIO.
4294  */
4295 static int try_to_fix_bad_block(struct btrfs_root *root,
4296                                 struct extent_buffer *buf,
4297                                 enum btrfs_tree_block_status status)
4298 {
4299         struct btrfs_trans_handle *trans;
4300         struct ulist *roots;
4301         struct ulist_node *node;
4302         struct btrfs_root *search_root;
4303         struct btrfs_path path;
4304         struct ulist_iterator iter;
4305         struct btrfs_key root_key, key;
4306         int ret;
4307
4308         if (status != BTRFS_TREE_BLOCK_BAD_KEY_ORDER &&
4309             status != BTRFS_TREE_BLOCK_INVALID_OFFSETS)
4310                 return -EIO;
4311
4312         ret = btrfs_find_all_roots(NULL, root->fs_info, buf->start, 0, &roots);
4313         if (ret)
4314                 return -EIO;
4315
4316         btrfs_init_path(&path);
4317         ULIST_ITER_INIT(&iter);
4318         while ((node = ulist_next(roots, &iter))) {
4319                 root_key.objectid = node->val;
4320                 root_key.type = BTRFS_ROOT_ITEM_KEY;
4321                 root_key.offset = (u64)-1;
4322
4323                 search_root = btrfs_read_fs_root(root->fs_info, &root_key);
4324                 if (IS_ERR(root)) {
4325                         ret = -EIO;
4326                         break;
4327                 }
4328
4329
4330                 trans = btrfs_start_transaction(search_root, 0);
4331                 if (IS_ERR(trans)) {
4332                         ret = PTR_ERR(trans);
4333                         break;
4334                 }
4335
4336                 path.lowest_level = btrfs_header_level(buf);
4337                 path.skip_check_block = 1;
4338                 if (path.lowest_level)
4339                         btrfs_node_key_to_cpu(buf, &key, 0);
4340                 else
4341                         btrfs_item_key_to_cpu(buf, &key, 0);
4342                 ret = btrfs_search_slot(trans, search_root, &key, &path, 0, 1);
4343                 if (ret) {
4344                         ret = -EIO;
4345                         btrfs_commit_transaction(trans, search_root);
4346                         break;
4347                 }
4348                 if (status == BTRFS_TREE_BLOCK_BAD_KEY_ORDER)
4349                         ret = fix_key_order(trans, search_root, &path);
4350                 else if (status == BTRFS_TREE_BLOCK_INVALID_OFFSETS)
4351                         ret = fix_item_offset(trans, search_root, &path);
4352                 if (ret) {
4353                         btrfs_commit_transaction(trans, search_root);
4354                         break;
4355                 }
4356                 btrfs_release_path(&path);
4357                 btrfs_commit_transaction(trans, search_root);
4358         }
4359         ulist_free(roots);
4360         btrfs_release_path(&path);
4361         return ret;
4362 }
4363
4364 static int check_block(struct btrfs_root *root,
4365                        struct cache_tree *extent_cache,
4366                        struct extent_buffer *buf, u64 flags)
4367 {
4368         struct extent_record *rec;
4369         struct cache_extent *cache;
4370         struct btrfs_key key;
4371         enum btrfs_tree_block_status status;
4372         int ret = 0;
4373         int level;
4374
4375         cache = lookup_cache_extent(extent_cache, buf->start, buf->len);
4376         if (!cache)
4377                 return 1;
4378         rec = container_of(cache, struct extent_record, cache);
4379         rec->generation = btrfs_header_generation(buf);
4380
4381         level = btrfs_header_level(buf);
4382         if (btrfs_header_nritems(buf) > 0) {
4383
4384                 if (level == 0)
4385                         btrfs_item_key_to_cpu(buf, &key, 0);
4386                 else
4387                         btrfs_node_key_to_cpu(buf, &key, 0);
4388
4389                 rec->info_objectid = key.objectid;
4390         }
4391         rec->info_level = level;
4392
4393         if (btrfs_is_leaf(buf))
4394                 status = btrfs_check_leaf(root, &rec->parent_key, buf);
4395         else
4396                 status = btrfs_check_node(root, &rec->parent_key, buf);
4397
4398         if (status != BTRFS_TREE_BLOCK_CLEAN) {
4399                 if (repair)
4400                         status = try_to_fix_bad_block(root, buf, status);
4401                 if (status != BTRFS_TREE_BLOCK_CLEAN) {
4402                         ret = -EIO;
4403                         fprintf(stderr, "bad block %llu\n",
4404                                 (unsigned long long)buf->start);
4405                 } else {
4406                         /*
4407                          * Signal to callers we need to start the scan over
4408                          * again since we'll have cowed blocks.
4409                          */
4410                         ret = -EAGAIN;
4411                 }
4412         } else {
4413                 rec->content_checked = 1;
4414                 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
4415                         rec->owner_ref_checked = 1;
4416                 else {
4417                         ret = check_owner_ref(root, rec, buf);
4418                         if (!ret)
4419                                 rec->owner_ref_checked = 1;
4420                 }
4421         }
4422         if (!ret)
4423                 maybe_free_extent_rec(extent_cache, rec);
4424         return ret;
4425 }
4426
4427 static struct tree_backref *find_tree_backref(struct extent_record *rec,
4428                                                 u64 parent, u64 root)
4429 {
4430         struct list_head *cur = rec->backrefs.next;
4431         struct extent_backref *node;
4432         struct tree_backref *back;
4433
4434         while(cur != &rec->backrefs) {
4435                 node = to_extent_backref(cur);
4436                 cur = cur->next;
4437                 if (node->is_data)
4438                         continue;
4439                 back = to_tree_backref(node);
4440                 if (parent > 0) {
4441                         if (!node->full_backref)
4442                                 continue;
4443                         if (parent == back->parent)
4444                                 return back;
4445                 } else {
4446                         if (node->full_backref)
4447                                 continue;
4448                         if (back->root == root)
4449                                 return back;
4450                 }
4451         }
4452         return NULL;
4453 }
4454
4455 static struct tree_backref *alloc_tree_backref(struct extent_record *rec,
4456                                                 u64 parent, u64 root)
4457 {
4458         struct tree_backref *ref = malloc(sizeof(*ref));
4459
4460         if (!ref)
4461                 return NULL;
4462         memset(&ref->node, 0, sizeof(ref->node));
4463         if (parent > 0) {
4464                 ref->parent = parent;
4465                 ref->node.full_backref = 1;
4466         } else {
4467                 ref->root = root;
4468                 ref->node.full_backref = 0;
4469         }
4470         list_add_tail(&ref->node.list, &rec->backrefs);
4471
4472         return ref;
4473 }
4474
4475 static struct data_backref *find_data_backref(struct extent_record *rec,
4476                                                 u64 parent, u64 root,
4477                                                 u64 owner, u64 offset,
4478                                                 int found_ref,
4479                                                 u64 disk_bytenr, u64 bytes)
4480 {
4481         struct list_head *cur = rec->backrefs.next;
4482         struct extent_backref *node;
4483         struct data_backref *back;
4484
4485         while(cur != &rec->backrefs) {
4486                 node = to_extent_backref(cur);
4487                 cur = cur->next;
4488                 if (!node->is_data)
4489                         continue;
4490                 back = to_data_backref(node);
4491                 if (parent > 0) {
4492                         if (!node->full_backref)
4493                                 continue;
4494                         if (parent == back->parent)
4495                                 return back;
4496                 } else {
4497                         if (node->full_backref)
4498                                 continue;
4499                         if (back->root == root && back->owner == owner &&
4500                             back->offset == offset) {
4501                                 if (found_ref && node->found_ref &&
4502                                     (back->bytes != bytes ||
4503                                     back->disk_bytenr != disk_bytenr))
4504                                         continue;
4505                                 return back;
4506                         }
4507                 }
4508         }
4509         return NULL;
4510 }
4511
4512 static struct data_backref *alloc_data_backref(struct extent_record *rec,
4513                                                 u64 parent, u64 root,
4514                                                 u64 owner, u64 offset,
4515                                                 u64 max_size)
4516 {
4517         struct data_backref *ref = malloc(sizeof(*ref));
4518
4519         if (!ref)
4520                 return NULL;
4521         memset(&ref->node, 0, sizeof(ref->node));
4522         ref->node.is_data = 1;
4523
4524         if (parent > 0) {
4525                 ref->parent = parent;
4526                 ref->owner = 0;
4527                 ref->offset = 0;
4528                 ref->node.full_backref = 1;
4529         } else {
4530                 ref->root = root;
4531                 ref->owner = owner;
4532                 ref->offset = offset;
4533                 ref->node.full_backref = 0;
4534         }
4535         ref->bytes = max_size;
4536         ref->found_ref = 0;
4537         ref->num_refs = 0;
4538         list_add_tail(&ref->node.list, &rec->backrefs);
4539         if (max_size > rec->max_size)
4540                 rec->max_size = max_size;
4541         return ref;
4542 }
4543
4544 /* Check if the type of extent matches with its chunk */
4545 static void check_extent_type(struct extent_record *rec)
4546 {
4547         struct btrfs_block_group_cache *bg_cache;
4548
4549         bg_cache = btrfs_lookup_first_block_group(global_info, rec->start);
4550         if (!bg_cache)
4551                 return;
4552
4553         /* data extent, check chunk directly*/
4554         if (!rec->metadata) {
4555                 if (!(bg_cache->flags & BTRFS_BLOCK_GROUP_DATA))
4556                         rec->wrong_chunk_type = 1;
4557                 return;
4558         }
4559
4560         /* metadata extent, check the obvious case first */
4561         if (!(bg_cache->flags & (BTRFS_BLOCK_GROUP_SYSTEM |
4562                                  BTRFS_BLOCK_GROUP_METADATA))) {
4563                 rec->wrong_chunk_type = 1;
4564                 return;
4565         }
4566
4567         /*
4568          * Check SYSTEM extent, as it's also marked as metadata, we can only
4569          * make sure it's a SYSTEM extent by its backref
4570          */
4571         if (!list_empty(&rec->backrefs)) {
4572                 struct extent_backref *node;
4573                 struct tree_backref *tback;
4574                 u64 bg_type;
4575
4576                 node = to_extent_backref(rec->backrefs.next);
4577                 if (node->is_data) {
4578                         /* tree block shouldn't have data backref */
4579                         rec->wrong_chunk_type = 1;
4580                         return;
4581                 }
4582                 tback = container_of(node, struct tree_backref, node);
4583
4584                 if (tback->root == BTRFS_CHUNK_TREE_OBJECTID)
4585                         bg_type = BTRFS_BLOCK_GROUP_SYSTEM;
4586                 else
4587                         bg_type = BTRFS_BLOCK_GROUP_METADATA;
4588                 if (!(bg_cache->flags & bg_type))
4589                         rec->wrong_chunk_type = 1;
4590         }
4591 }
4592
4593 /*
4594  * Allocate a new extent record, fill default values from @tmpl and insert int
4595  * @extent_cache. Caller is supposed to make sure the [start,nr) is not in
4596  * the cache, otherwise it fails.
4597  */
4598 static int add_extent_rec_nolookup(struct cache_tree *extent_cache,
4599                 struct extent_record *tmpl)
4600 {
4601         struct extent_record *rec;
4602         int ret = 0;
4603
4604         rec = malloc(sizeof(*rec));
4605         if (!rec)
4606                 return -ENOMEM;
4607         rec->start = tmpl->start;
4608         rec->max_size = tmpl->max_size;
4609         rec->nr = max(tmpl->nr, tmpl->max_size);
4610         rec->found_rec = tmpl->found_rec;
4611         rec->content_checked = tmpl->content_checked;
4612         rec->owner_ref_checked = tmpl->owner_ref_checked;
4613         rec->num_duplicates = 0;
4614         rec->metadata = tmpl->metadata;
4615         rec->flag_block_full_backref = FLAG_UNSET;
4616         rec->bad_full_backref = 0;
4617         rec->crossing_stripes = 0;
4618         rec->wrong_chunk_type = 0;
4619         rec->is_root = tmpl->is_root;
4620         rec->refs = tmpl->refs;
4621         rec->extent_item_refs = tmpl->extent_item_refs;
4622         rec->parent_generation = tmpl->parent_generation;
4623         INIT_LIST_HEAD(&rec->backrefs);
4624         INIT_LIST_HEAD(&rec->dups);
4625         INIT_LIST_HEAD(&rec->list);
4626         memcpy(&rec->parent_key, &tmpl->parent_key, sizeof(tmpl->parent_key));
4627         rec->cache.start = tmpl->start;
4628         rec->cache.size = tmpl->nr;
4629         ret = insert_cache_extent(extent_cache, &rec->cache);
4630         if (ret) {
4631                 free(rec);
4632                 return ret;
4633         }
4634         bytes_used += rec->nr;
4635
4636         if (tmpl->metadata)
4637                 rec->crossing_stripes = check_crossing_stripes(global_info,
4638                                 rec->start, global_info->tree_root->nodesize);
4639         check_extent_type(rec);
4640         return ret;
4641 }
4642
4643 /*
4644  * Lookup and modify an extent, some values of @tmpl are interpreted verbatim,
4645  * some are hints:
4646  * - refs              - if found, increase refs
4647  * - is_root           - if found, set
4648  * - content_checked   - if found, set
4649  * - owner_ref_checked - if found, set
4650  *
4651  * If not found, create a new one, initialize and insert.
4652  */
4653 static int add_extent_rec(struct cache_tree *extent_cache,
4654                 struct extent_record *tmpl)
4655 {
4656         struct extent_record *rec;
4657         struct cache_extent *cache;
4658         int ret = 0;
4659         int dup = 0;
4660
4661         cache = lookup_cache_extent(extent_cache, tmpl->start, tmpl->nr);
4662         if (cache) {
4663                 rec = container_of(cache, struct extent_record, cache);
4664                 if (tmpl->refs)
4665                         rec->refs++;
4666                 if (rec->nr == 1)
4667                         rec->nr = max(tmpl->nr, tmpl->max_size);
4668
4669                 /*
4670                  * We need to make sure to reset nr to whatever the extent
4671                  * record says was the real size, this way we can compare it to
4672                  * the backrefs.
4673                  */
4674                 if (tmpl->found_rec) {
4675                         if (tmpl->start != rec->start || rec->found_rec) {
4676                                 struct extent_record *tmp;
4677
4678                                 dup = 1;
4679                                 if (list_empty(&rec->list))
4680                                         list_add_tail(&rec->list,
4681                                                       &duplicate_extents);
4682
4683                                 /*
4684                                  * We have to do this song and dance in case we
4685                                  * find an extent record that falls inside of
4686                                  * our current extent record but does not have
4687                                  * the same objectid.
4688                                  */
4689                                 tmp = malloc(sizeof(*tmp));
4690                                 if (!tmp)
4691                                         return -ENOMEM;
4692                                 tmp->start = tmpl->start;
4693                                 tmp->max_size = tmpl->max_size;
4694                                 tmp->nr = tmpl->nr;
4695                                 tmp->found_rec = 1;
4696                                 tmp->metadata = tmpl->metadata;
4697                                 tmp->extent_item_refs = tmpl->extent_item_refs;
4698                                 INIT_LIST_HEAD(&tmp->list);
4699                                 list_add_tail(&tmp->list, &rec->dups);
4700                                 rec->num_duplicates++;
4701                         } else {
4702                                 rec->nr = tmpl->nr;
4703                                 rec->found_rec = 1;
4704                         }
4705                 }
4706
4707                 if (tmpl->extent_item_refs && !dup) {
4708                         if (rec->extent_item_refs) {
4709                                 fprintf(stderr, "block %llu rec "
4710                                         "extent_item_refs %llu, passed %llu\n",
4711                                         (unsigned long long)tmpl->start,
4712                                         (unsigned long long)
4713                                                         rec->extent_item_refs,
4714                                         (unsigned long long)tmpl->extent_item_refs);
4715                         }
4716                         rec->extent_item_refs = tmpl->extent_item_refs;
4717                 }
4718                 if (tmpl->is_root)
4719                         rec->is_root = 1;
4720                 if (tmpl->content_checked)
4721                         rec->content_checked = 1;
4722                 if (tmpl->owner_ref_checked)
4723                         rec->owner_ref_checked = 1;
4724                 memcpy(&rec->parent_key, &tmpl->parent_key,
4725                                 sizeof(tmpl->parent_key));
4726                 if (tmpl->parent_generation)
4727                         rec->parent_generation = tmpl->parent_generation;
4728                 if (rec->max_size < tmpl->max_size)
4729                         rec->max_size = tmpl->max_size;
4730
4731                 /*
4732                  * A metadata extent can't cross stripe_len boundary, otherwise
4733                  * kernel scrub won't be able to handle it.
4734                  * As now stripe_len is fixed to BTRFS_STRIPE_LEN, just check
4735                  * it.
4736                  */
4737                 if (tmpl->metadata)
4738                         rec->crossing_stripes = check_crossing_stripes(
4739                                         global_info, rec->start,
4740                                         global_info->tree_root->nodesize);
4741                 check_extent_type(rec);
4742                 maybe_free_extent_rec(extent_cache, rec);
4743                 return ret;
4744         }
4745
4746         ret = add_extent_rec_nolookup(extent_cache, tmpl);
4747
4748         return ret;
4749 }
4750
4751 static int add_tree_backref(struct cache_tree *extent_cache, u64 bytenr,
4752                             u64 parent, u64 root, int found_ref)
4753 {
4754         struct extent_record *rec;
4755         struct tree_backref *back;
4756         struct cache_extent *cache;
4757         int ret;
4758
4759         cache = lookup_cache_extent(extent_cache, bytenr, 1);
4760         if (!cache) {
4761                 struct extent_record tmpl;
4762
4763                 memset(&tmpl, 0, sizeof(tmpl));
4764                 tmpl.start = bytenr;
4765                 tmpl.nr = 1;
4766                 tmpl.metadata = 1;
4767
4768                 ret = add_extent_rec_nolookup(extent_cache, &tmpl);
4769                 if (ret)
4770                         return ret;
4771
4772                 /* really a bug in cache_extent implement now */
4773                 cache = lookup_cache_extent(extent_cache, bytenr, 1);
4774                 if (!cache)
4775                         return -ENOENT;
4776         }
4777
4778         rec = container_of(cache, struct extent_record, cache);
4779         if (rec->start != bytenr) {
4780                 /*
4781                  * Several cause, from unaligned bytenr to over lapping extents
4782                  */
4783                 return -EEXIST;
4784         }
4785
4786         back = find_tree_backref(rec, parent, root);
4787         if (!back) {
4788                 back = alloc_tree_backref(rec, parent, root);
4789                 if (!back)
4790                         return -ENOMEM;
4791         }
4792
4793         if (found_ref) {
4794                 if (back->node.found_ref) {
4795                         fprintf(stderr, "Extent back ref already exists "
4796                                 "for %llu parent %llu root %llu \n",
4797                                 (unsigned long long)bytenr,
4798                                 (unsigned long long)parent,
4799                                 (unsigned long long)root);
4800                 }
4801                 back->node.found_ref = 1;
4802         } else {
4803                 if (back->node.found_extent_tree) {
4804                         fprintf(stderr, "Extent back ref already exists "
4805                                 "for %llu parent %llu root %llu \n",
4806                                 (unsigned long long)bytenr,
4807                                 (unsigned long long)parent,
4808                                 (unsigned long long)root);
4809                 }
4810                 back->node.found_extent_tree = 1;
4811         }
4812         check_extent_type(rec);
4813         maybe_free_extent_rec(extent_cache, rec);
4814         return 0;
4815 }
4816
4817 static int add_data_backref(struct cache_tree *extent_cache, u64 bytenr,
4818                             u64 parent, u64 root, u64 owner, u64 offset,
4819                             u32 num_refs, int found_ref, u64 max_size)
4820 {
4821         struct extent_record *rec;
4822         struct data_backref *back;
4823         struct cache_extent *cache;
4824         int ret;
4825
4826         cache = lookup_cache_extent(extent_cache, bytenr, 1);
4827         if (!cache) {
4828                 struct extent_record tmpl;
4829
4830                 memset(&tmpl, 0, sizeof(tmpl));
4831                 tmpl.start = bytenr;
4832                 tmpl.nr = 1;
4833                 tmpl.max_size = max_size;
4834
4835                 ret = add_extent_rec_nolookup(extent_cache, &tmpl);
4836                 if (ret)
4837                         return ret;
4838
4839                 cache = lookup_cache_extent(extent_cache, bytenr, 1);
4840                 if (!cache)
4841                         abort();
4842         }
4843
4844         rec = container_of(cache, struct extent_record, cache);
4845         if (rec->max_size < max_size)
4846                 rec->max_size = max_size;
4847
4848         /*
4849          * If found_ref is set then max_size is the real size and must match the
4850          * existing refs.  So if we have already found a ref then we need to
4851          * make sure that this ref matches the existing one, otherwise we need
4852          * to add a new backref so we can notice that the backrefs don't match
4853          * and we need to figure out who is telling the truth.  This is to
4854          * account for that awful fsync bug I introduced where we'd end up with
4855          * a btrfs_file_extent_item that would have its length include multiple
4856          * prealloc extents or point inside of a prealloc extent.
4857          */
4858         back = find_data_backref(rec, parent, root, owner, offset, found_ref,
4859                                  bytenr, max_size);
4860         if (!back) {
4861                 back = alloc_data_backref(rec, parent, root, owner, offset,
4862                                           max_size);
4863                 BUG_ON(!back);
4864         }
4865
4866         if (found_ref) {
4867                 BUG_ON(num_refs != 1);
4868                 if (back->node.found_ref)
4869                         BUG_ON(back->bytes != max_size);
4870                 back->node.found_ref = 1;
4871                 back->found_ref += 1;
4872                 back->bytes = max_size;
4873                 back->disk_bytenr = bytenr;
4874                 rec->refs += 1;
4875                 rec->content_checked = 1;
4876                 rec->owner_ref_checked = 1;
4877         } else {
4878                 if (back->node.found_extent_tree) {
4879                         fprintf(stderr, "Extent back ref already exists "
4880                                 "for %llu parent %llu root %llu "
4881                                 "owner %llu offset %llu num_refs %lu\n",
4882                                 (unsigned long long)bytenr,
4883                                 (unsigned long long)parent,
4884                                 (unsigned long long)root,
4885                                 (unsigned long long)owner,
4886                                 (unsigned long long)offset,
4887                                 (unsigned long)num_refs);
4888                 }
4889                 back->num_refs = num_refs;
4890                 back->node.found_extent_tree = 1;
4891         }
4892         maybe_free_extent_rec(extent_cache, rec);
4893         return 0;
4894 }
4895
4896 static int add_pending(struct cache_tree *pending,
4897                        struct cache_tree *seen, u64 bytenr, u32 size)
4898 {
4899         int ret;
4900         ret = add_cache_extent(seen, bytenr, size);
4901         if (ret)
4902                 return ret;
4903         add_cache_extent(pending, bytenr, size);
4904         return 0;
4905 }
4906
4907 static int pick_next_pending(struct cache_tree *pending,
4908                         struct cache_tree *reada,
4909                         struct cache_tree *nodes,
4910                         u64 last, struct block_info *bits, int bits_nr,
4911                         int *reada_bits)
4912 {
4913         unsigned long node_start = last;
4914         struct cache_extent *cache;
4915         int ret;
4916
4917         cache = search_cache_extent(reada, 0);
4918         if (cache) {
4919                 bits[0].start = cache->start;
4920                 bits[0].size = cache->size;
4921                 *reada_bits = 1;
4922                 return 1;
4923         }
4924         *reada_bits = 0;
4925         if (node_start > 32768)
4926                 node_start -= 32768;
4927
4928         cache = search_cache_extent(nodes, node_start);
4929         if (!cache)
4930                 cache = search_cache_extent(nodes, 0);
4931
4932         if (!cache) {
4933                  cache = search_cache_extent(pending, 0);
4934                  if (!cache)
4935                          return 0;
4936                  ret = 0;
4937                  do {
4938                          bits[ret].start = cache->start;
4939                          bits[ret].size = cache->size;
4940                          cache = next_cache_extent(cache);
4941                          ret++;
4942                  } while (cache && ret < bits_nr);
4943                  return ret;
4944         }
4945
4946         ret = 0;
4947         do {
4948                 bits[ret].start = cache->start;
4949                 bits[ret].size = cache->size;
4950                 cache = next_cache_extent(cache);
4951                 ret++;
4952         } while (cache && ret < bits_nr);
4953
4954         if (bits_nr - ret > 8) {
4955                 u64 lookup = bits[0].start + bits[0].size;
4956                 struct cache_extent *next;
4957                 next = search_cache_extent(pending, lookup);
4958                 while(next) {
4959                         if (next->start - lookup > 32768)
4960                                 break;
4961                         bits[ret].start = next->start;
4962                         bits[ret].size = next->size;
4963                         lookup = next->start + next->size;
4964                         ret++;
4965                         if (ret == bits_nr)
4966                                 break;
4967                         next = next_cache_extent(next);
4968                         if (!next)
4969                                 break;
4970                 }
4971         }
4972         return ret;
4973 }
4974
4975 static void free_chunk_record(struct cache_extent *cache)
4976 {
4977         struct chunk_record *rec;
4978
4979         rec = container_of(cache, struct chunk_record, cache);
4980         list_del_init(&rec->list);
4981         list_del_init(&rec->dextents);
4982         free(rec);
4983 }
4984
4985 void free_chunk_cache_tree(struct cache_tree *chunk_cache)
4986 {
4987         cache_tree_free_extents(chunk_cache, free_chunk_record);
4988 }
4989
4990 static void free_device_record(struct rb_node *node)
4991 {
4992         struct device_record *rec;
4993
4994         rec = container_of(node, struct device_record, node);
4995         free(rec);
4996 }
4997
4998 FREE_RB_BASED_TREE(device_cache, free_device_record);
4999
5000 int insert_block_group_record(struct block_group_tree *tree,
5001                               struct block_group_record *bg_rec)
5002 {
5003         int ret;
5004
5005         ret = insert_cache_extent(&tree->tree, &bg_rec->cache);
5006         if (ret)
5007                 return ret;
5008
5009         list_add_tail(&bg_rec->list, &tree->block_groups);
5010         return 0;
5011 }
5012
5013 static void free_block_group_record(struct cache_extent *cache)
5014 {
5015         struct block_group_record *rec;
5016
5017         rec = container_of(cache, struct block_group_record, cache);
5018         list_del_init(&rec->list);
5019         free(rec);
5020 }
5021
5022 void free_block_group_tree(struct block_group_tree *tree)
5023 {
5024         cache_tree_free_extents(&tree->tree, free_block_group_record);
5025 }
5026
5027 int insert_device_extent_record(struct device_extent_tree *tree,
5028                                 struct device_extent_record *de_rec)
5029 {
5030         int ret;
5031
5032         /*
5033          * Device extent is a bit different from the other extents, because
5034          * the extents which belong to the different devices may have the
5035          * same start and size, so we need use the special extent cache
5036          * search/insert functions.
5037          */
5038         ret = insert_cache_extent2(&tree->tree, &de_rec->cache);
5039         if (ret)
5040                 return ret;
5041
5042         list_add_tail(&de_rec->chunk_list, &tree->no_chunk_orphans);
5043         list_add_tail(&de_rec->device_list, &tree->no_device_orphans);
5044         return 0;
5045 }
5046
5047 static void free_device_extent_record(struct cache_extent *cache)
5048 {
5049         struct device_extent_record *rec;
5050
5051         rec = container_of(cache, struct device_extent_record, cache);
5052         if (!list_empty(&rec->chunk_list))
5053                 list_del_init(&rec->chunk_list);
5054         if (!list_empty(&rec->device_list))
5055                 list_del_init(&rec->device_list);
5056         free(rec);
5057 }
5058
5059 void free_device_extent_tree(struct device_extent_tree *tree)
5060 {
5061         cache_tree_free_extents(&tree->tree, free_device_extent_record);
5062 }
5063
5064 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5065 static int process_extent_ref_v0(struct cache_tree *extent_cache,
5066                                  struct extent_buffer *leaf, int slot)
5067 {
5068         struct btrfs_extent_ref_v0 *ref0;
5069         struct btrfs_key key;
5070         int ret;
5071
5072         btrfs_item_key_to_cpu(leaf, &key, slot);
5073         ref0 = btrfs_item_ptr(leaf, slot, struct btrfs_extent_ref_v0);
5074         if (btrfs_ref_objectid_v0(leaf, ref0) < BTRFS_FIRST_FREE_OBJECTID) {
5075                 ret = add_tree_backref(extent_cache, key.objectid, key.offset,
5076                                 0, 0);
5077         } else {
5078                 ret = add_data_backref(extent_cache, key.objectid, key.offset,
5079                                 0, 0, 0, btrfs_ref_count_v0(leaf, ref0), 0, 0);
5080         }
5081         return ret;
5082 }
5083 #endif
5084
5085 struct chunk_record *btrfs_new_chunk_record(struct extent_buffer *leaf,
5086                                             struct btrfs_key *key,
5087                                             int slot)
5088 {
5089         struct btrfs_chunk *ptr;
5090         struct chunk_record *rec;
5091         int num_stripes, i;
5092
5093         ptr = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
5094         num_stripes = btrfs_chunk_num_stripes(leaf, ptr);
5095
5096         rec = calloc(1, btrfs_chunk_record_size(num_stripes));
5097         if (!rec) {
5098                 fprintf(stderr, "memory allocation failed\n");
5099                 exit(-1);
5100         }
5101
5102         INIT_LIST_HEAD(&rec->list);
5103         INIT_LIST_HEAD(&rec->dextents);
5104         rec->bg_rec = NULL;
5105
5106         rec->cache.start = key->offset;
5107         rec->cache.size = btrfs_chunk_length(leaf, ptr);
5108
5109         rec->generation = btrfs_header_generation(leaf);
5110
5111         rec->objectid = key->objectid;
5112         rec->type = key->type;
5113         rec->offset = key->offset;
5114
5115         rec->length = rec->cache.size;
5116         rec->owner = btrfs_chunk_owner(leaf, ptr);
5117         rec->stripe_len = btrfs_chunk_stripe_len(leaf, ptr);
5118         rec->type_flags = btrfs_chunk_type(leaf, ptr);
5119         rec->io_width = btrfs_chunk_io_width(leaf, ptr);
5120         rec->io_align = btrfs_chunk_io_align(leaf, ptr);
5121         rec->sector_size = btrfs_chunk_sector_size(leaf, ptr);
5122         rec->num_stripes = num_stripes;
5123         rec->sub_stripes = btrfs_chunk_sub_stripes(leaf, ptr);
5124
5125         for (i = 0; i < rec->num_stripes; ++i) {
5126                 rec->stripes[i].devid =
5127                         btrfs_stripe_devid_nr(leaf, ptr, i);
5128                 rec->stripes[i].offset =
5129                         btrfs_stripe_offset_nr(leaf, ptr, i);
5130                 read_extent_buffer(leaf, rec->stripes[i].dev_uuid,
5131                                 (unsigned long)btrfs_stripe_dev_uuid_nr(ptr, i),
5132                                 BTRFS_UUID_SIZE);
5133         }
5134
5135         return rec;
5136 }
5137
5138 static int process_chunk_item(struct cache_tree *chunk_cache,
5139                               struct btrfs_key *key, struct extent_buffer *eb,
5140                               int slot)
5141 {
5142         struct chunk_record *rec;
5143         struct btrfs_chunk *chunk;
5144         int ret = 0;
5145
5146         chunk = btrfs_item_ptr(eb, slot, struct btrfs_chunk);
5147         /*
5148          * Do extra check for this chunk item,
5149          *
5150          * It's still possible one can craft a leaf with CHUNK_ITEM, with
5151          * wrong onwer(3) out of chunk tree, to pass both chunk tree check
5152          * and owner<->key_type check.
5153          */
5154         ret = btrfs_check_chunk_valid(global_info->tree_root, eb, chunk, slot,
5155                                       key->offset);
5156         if (ret < 0) {
5157                 error("chunk(%llu, %llu) is not valid, ignore it",
5158                       key->offset, btrfs_chunk_length(eb, chunk));
5159                 return 0;
5160         }
5161         rec = btrfs_new_chunk_record(eb, key, slot);
5162         ret = insert_cache_extent(chunk_cache, &rec->cache);
5163         if (ret) {
5164                 fprintf(stderr, "Chunk[%llu, %llu] existed.\n",
5165                         rec->offset, rec->length);
5166                 free(rec);
5167         }
5168
5169         return ret;
5170 }
5171
5172 static int process_device_item(struct rb_root *dev_cache,
5173                 struct btrfs_key *key, struct extent_buffer *eb, int slot)
5174 {
5175         struct btrfs_dev_item *ptr;
5176         struct device_record *rec;
5177         int ret = 0;
5178
5179         ptr = btrfs_item_ptr(eb,
5180                 slot, struct btrfs_dev_item);
5181
5182         rec = malloc(sizeof(*rec));
5183         if (!rec) {
5184                 fprintf(stderr, "memory allocation failed\n");
5185                 return -ENOMEM;
5186         }
5187
5188         rec->devid = key->offset;
5189         rec->generation = btrfs_header_generation(eb);
5190
5191         rec->objectid = key->objectid;
5192         rec->type = key->type;
5193         rec->offset = key->offset;
5194
5195         rec->devid = btrfs_device_id(eb, ptr);
5196         rec->total_byte = btrfs_device_total_bytes(eb, ptr);
5197         rec->byte_used = btrfs_device_bytes_used(eb, ptr);
5198
5199         ret = rb_insert(dev_cache, &rec->node, device_record_compare);
5200         if (ret) {
5201                 fprintf(stderr, "Device[%llu] existed.\n", rec->devid);
5202                 free(rec);
5203         }
5204
5205         return ret;
5206 }
5207
5208 struct block_group_record *
5209 btrfs_new_block_group_record(struct extent_buffer *leaf, struct btrfs_key *key,
5210                              int slot)
5211 {
5212         struct btrfs_block_group_item *ptr;
5213         struct block_group_record *rec;
5214
5215         rec = calloc(1, sizeof(*rec));
5216         if (!rec) {
5217                 fprintf(stderr, "memory allocation failed\n");
5218                 exit(-1);
5219         }
5220
5221         rec->cache.start = key->objectid;
5222         rec->cache.size = key->offset;
5223
5224         rec->generation = btrfs_header_generation(leaf);
5225
5226         rec->objectid = key->objectid;
5227         rec->type = key->type;
5228         rec->offset = key->offset;
5229
5230         ptr = btrfs_item_ptr(leaf, slot, struct btrfs_block_group_item);
5231         rec->flags = btrfs_disk_block_group_flags(leaf, ptr);
5232
5233         INIT_LIST_HEAD(&rec->list);
5234
5235         return rec;
5236 }
5237
5238 static int process_block_group_item(struct block_group_tree *block_group_cache,
5239                                     struct btrfs_key *key,
5240                                     struct extent_buffer *eb, int slot)
5241 {
5242         struct block_group_record *rec;
5243         int ret = 0;
5244
5245         rec = btrfs_new_block_group_record(eb, key, slot);
5246         ret = insert_block_group_record(block_group_cache, rec);
5247         if (ret) {
5248                 fprintf(stderr, "Block Group[%llu, %llu] existed.\n",
5249                         rec->objectid, rec->offset);
5250                 free(rec);
5251         }
5252
5253         return ret;
5254 }
5255
5256 struct device_extent_record *
5257 btrfs_new_device_extent_record(struct extent_buffer *leaf,
5258                                struct btrfs_key *key, int slot)
5259 {
5260         struct device_extent_record *rec;
5261         struct btrfs_dev_extent *ptr;
5262
5263         rec = calloc(1, sizeof(*rec));
5264         if (!rec) {
5265                 fprintf(stderr, "memory allocation failed\n");
5266                 exit(-1);
5267         }
5268
5269         rec->cache.objectid = key->objectid;
5270         rec->cache.start = key->offset;
5271
5272         rec->generation = btrfs_header_generation(leaf);
5273
5274         rec->objectid = key->objectid;
5275         rec->type = key->type;
5276         rec->offset = key->offset;
5277
5278         ptr = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
5279         rec->chunk_objecteid =
5280                 btrfs_dev_extent_chunk_objectid(leaf, ptr);
5281         rec->chunk_offset =
5282                 btrfs_dev_extent_chunk_offset(leaf, ptr);
5283         rec->length = btrfs_dev_extent_length(leaf, ptr);
5284         rec->cache.size = rec->length;
5285
5286         INIT_LIST_HEAD(&rec->chunk_list);
5287         INIT_LIST_HEAD(&rec->device_list);
5288
5289         return rec;
5290 }
5291
5292 static int
5293 process_device_extent_item(struct device_extent_tree *dev_extent_cache,
5294                            struct btrfs_key *key, struct extent_buffer *eb,
5295                            int slot)
5296 {
5297         struct device_extent_record *rec;
5298         int ret;
5299
5300         rec = btrfs_new_device_extent_record(eb, key, slot);
5301         ret = insert_device_extent_record(dev_extent_cache, rec);
5302         if (ret) {
5303                 fprintf(stderr,
5304                         "Device extent[%llu, %llu, %llu] existed.\n",
5305                         rec->objectid, rec->offset, rec->length);
5306                 free(rec);
5307         }
5308
5309         return ret;
5310 }
5311
5312 static int process_extent_item(struct btrfs_root *root,
5313                                struct cache_tree *extent_cache,
5314                                struct extent_buffer *eb, int slot)
5315 {
5316         struct btrfs_extent_item *ei;
5317         struct btrfs_extent_inline_ref *iref;
5318         struct btrfs_extent_data_ref *dref;
5319         struct btrfs_shared_data_ref *sref;
5320         struct btrfs_key key;
5321         struct extent_record tmpl;
5322         unsigned long end;
5323         unsigned long ptr;
5324         int ret;
5325         int type;
5326         u32 item_size = btrfs_item_size_nr(eb, slot);
5327         u64 refs = 0;
5328         u64 offset;
5329         u64 num_bytes;
5330         int metadata = 0;
5331
5332         btrfs_item_key_to_cpu(eb, &key, slot);
5333
5334         if (key.type == BTRFS_METADATA_ITEM_KEY) {
5335                 metadata = 1;
5336                 num_bytes = root->nodesize;
5337         } else {
5338                 num_bytes = key.offset;
5339         }
5340
5341         if (!IS_ALIGNED(key.objectid, root->sectorsize)) {
5342                 error("ignoring invalid extent, bytenr %llu is not aligned to %u",
5343                       key.objectid, root->sectorsize);
5344                 return -EIO;
5345         }
5346         if (item_size < sizeof(*ei)) {
5347 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5348                 struct btrfs_extent_item_v0 *ei0;
5349                 BUG_ON(item_size != sizeof(*ei0));
5350                 ei0 = btrfs_item_ptr(eb, slot, struct btrfs_extent_item_v0);
5351                 refs = btrfs_extent_refs_v0(eb, ei0);
5352 #else
5353                 BUG();
5354 #endif
5355                 memset(&tmpl, 0, sizeof(tmpl));
5356                 tmpl.start = key.objectid;
5357                 tmpl.nr = num_bytes;
5358                 tmpl.extent_item_refs = refs;
5359                 tmpl.metadata = metadata;
5360                 tmpl.found_rec = 1;
5361                 tmpl.max_size = num_bytes;
5362
5363                 return add_extent_rec(extent_cache, &tmpl);
5364         }
5365
5366         ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
5367         refs = btrfs_extent_refs(eb, ei);
5368         if (btrfs_extent_flags(eb, ei) & BTRFS_EXTENT_FLAG_TREE_BLOCK)
5369                 metadata = 1;
5370         else
5371                 metadata = 0;
5372         if (metadata && num_bytes != root->nodesize) {
5373                 error("ignore invalid metadata extent, length %llu does not equal to %u",
5374                       num_bytes, root->nodesize);
5375                 return -EIO;
5376         }
5377         if (!metadata && !IS_ALIGNED(num_bytes, root->sectorsize)) {
5378                 error("ignore invalid data extent, length %llu is not aligned to %u",
5379                       num_bytes, root->sectorsize);
5380                 return -EIO;
5381         }
5382
5383         memset(&tmpl, 0, sizeof(tmpl));
5384         tmpl.start = key.objectid;
5385         tmpl.nr = num_bytes;
5386         tmpl.extent_item_refs = refs;
5387         tmpl.metadata = metadata;
5388         tmpl.found_rec = 1;
5389         tmpl.max_size = num_bytes;
5390         add_extent_rec(extent_cache, &tmpl);
5391
5392         ptr = (unsigned long)(ei + 1);
5393         if (btrfs_extent_flags(eb, ei) & BTRFS_EXTENT_FLAG_TREE_BLOCK &&
5394             key.type == BTRFS_EXTENT_ITEM_KEY)
5395                 ptr += sizeof(struct btrfs_tree_block_info);
5396
5397         end = (unsigned long)ei + item_size;
5398         while (ptr < end) {
5399                 iref = (struct btrfs_extent_inline_ref *)ptr;
5400                 type = btrfs_extent_inline_ref_type(eb, iref);
5401                 offset = btrfs_extent_inline_ref_offset(eb, iref);
5402                 switch (type) {
5403                 case BTRFS_TREE_BLOCK_REF_KEY:
5404                         ret = add_tree_backref(extent_cache, key.objectid,
5405                                         0, offset, 0);
5406                         if (ret < 0)
5407                                 error("add_tree_backref failed: %s",
5408                                       strerror(-ret));
5409                         break;
5410                 case BTRFS_SHARED_BLOCK_REF_KEY:
5411                         ret = add_tree_backref(extent_cache, key.objectid,
5412                                         offset, 0, 0);
5413                         if (ret < 0)
5414                                 error("add_tree_backref failed: %s",
5415                                       strerror(-ret));
5416                         break;
5417                 case BTRFS_EXTENT_DATA_REF_KEY:
5418                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
5419                         add_data_backref(extent_cache, key.objectid, 0,
5420                                         btrfs_extent_data_ref_root(eb, dref),
5421                                         btrfs_extent_data_ref_objectid(eb,
5422                                                                        dref),
5423                                         btrfs_extent_data_ref_offset(eb, dref),
5424                                         btrfs_extent_data_ref_count(eb, dref),
5425                                         0, num_bytes);
5426                         break;
5427                 case BTRFS_SHARED_DATA_REF_KEY:
5428                         sref = (struct btrfs_shared_data_ref *)(iref + 1);
5429                         add_data_backref(extent_cache, key.objectid, offset,
5430                                         0, 0, 0,
5431                                         btrfs_shared_data_ref_count(eb, sref),
5432                                         0, num_bytes);
5433                         break;
5434                 default:
5435                         fprintf(stderr, "corrupt extent record: key %Lu %u %Lu\n",
5436                                 key.objectid, key.type, num_bytes);
5437                         goto out;
5438                 }
5439                 ptr += btrfs_extent_inline_ref_size(type);
5440         }
5441         WARN_ON(ptr > end);
5442 out:
5443         return 0;
5444 }
5445
5446 static int check_cache_range(struct btrfs_root *root,
5447                              struct btrfs_block_group_cache *cache,
5448                              u64 offset, u64 bytes)
5449 {
5450         struct btrfs_free_space *entry;
5451         u64 *logical;
5452         u64 bytenr;
5453         int stripe_len;
5454         int i, nr, ret;
5455
5456         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
5457                 bytenr = btrfs_sb_offset(i);
5458                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
5459                                        cache->key.objectid, bytenr, 0,
5460                                        &logical, &nr, &stripe_len);
5461                 if (ret)
5462                         return ret;
5463
5464                 while (nr--) {
5465                         if (logical[nr] + stripe_len <= offset)
5466                                 continue;
5467                         if (offset + bytes <= logical[nr])
5468                                 continue;
5469                         if (logical[nr] == offset) {
5470                                 if (stripe_len >= bytes) {
5471                                         free(logical);
5472                                         return 0;
5473                                 }
5474                                 bytes -= stripe_len;
5475                                 offset += stripe_len;
5476                         } else if (logical[nr] < offset) {
5477                                 if (logical[nr] + stripe_len >=
5478                                     offset + bytes) {
5479                                         free(logical);
5480                                         return 0;
5481                                 }
5482                                 bytes = (offset + bytes) -
5483                                         (logical[nr] + stripe_len);
5484                                 offset = logical[nr] + stripe_len;
5485                         } else {
5486                                 /*
5487                                  * Could be tricky, the super may land in the
5488                                  * middle of the area we're checking.  First
5489                                  * check the easiest case, it's at the end.
5490                                  */
5491                                 if (logical[nr] + stripe_len >=
5492                                     bytes + offset) {
5493                                         bytes = logical[nr] - offset;
5494                                         continue;
5495                                 }
5496
5497                                 /* Check the left side */
5498                                 ret = check_cache_range(root, cache,
5499                                                         offset,
5500                                                         logical[nr] - offset);
5501                                 if (ret) {
5502                                         free(logical);
5503                                         return ret;
5504                                 }
5505
5506                                 /* Now we continue with the right side */
5507                                 bytes = (offset + bytes) -
5508                                         (logical[nr] + stripe_len);
5509                                 offset = logical[nr] + stripe_len;
5510                         }
5511                 }
5512
5513                 free(logical);
5514         }
5515
5516         entry = btrfs_find_free_space(cache->free_space_ctl, offset, bytes);
5517         if (!entry) {
5518                 fprintf(stderr, "There is no free space entry for %Lu-%Lu\n",
5519                         offset, offset+bytes);
5520                 return -EINVAL;
5521         }
5522
5523         if (entry->offset != offset) {
5524                 fprintf(stderr, "Wanted offset %Lu, found %Lu\n", offset,
5525                         entry->offset);
5526                 return -EINVAL;
5527         }
5528
5529         if (entry->bytes != bytes) {
5530                 fprintf(stderr, "Wanted bytes %Lu, found %Lu for off %Lu\n",
5531                         bytes, entry->bytes, offset);
5532                 return -EINVAL;
5533         }
5534
5535         unlink_free_space(cache->free_space_ctl, entry);
5536         free(entry);
5537         return 0;
5538 }
5539
5540 static int verify_space_cache(struct btrfs_root *root,
5541                               struct btrfs_block_group_cache *cache)
5542 {
5543         struct btrfs_path path;
5544         struct extent_buffer *leaf;
5545         struct btrfs_key key;
5546         u64 last;
5547         int ret = 0;
5548
5549         root = root->fs_info->extent_root;
5550
5551         last = max_t(u64, cache->key.objectid, BTRFS_SUPER_INFO_OFFSET);
5552
5553         btrfs_init_path(&path);
5554         key.objectid = last;
5555         key.offset = 0;
5556         key.type = BTRFS_EXTENT_ITEM_KEY;
5557         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
5558         if (ret < 0)
5559                 goto out;
5560         ret = 0;
5561         while (1) {
5562                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
5563                         ret = btrfs_next_leaf(root, &path);
5564                         if (ret < 0)
5565                                 goto out;
5566                         if (ret > 0) {
5567                                 ret = 0;
5568                                 break;
5569                         }
5570                 }
5571                 leaf = path.nodes[0];
5572                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
5573                 if (key.objectid >= cache->key.offset + cache->key.objectid)
5574                         break;
5575                 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
5576                     key.type != BTRFS_METADATA_ITEM_KEY) {
5577                         path.slots[0]++;
5578                         continue;
5579                 }
5580
5581                 if (last == key.objectid) {
5582                         if (key.type == BTRFS_EXTENT_ITEM_KEY)
5583                                 last = key.objectid + key.offset;
5584                         else
5585                                 last = key.objectid + root->nodesize;
5586                         path.slots[0]++;
5587                         continue;
5588                 }
5589
5590                 ret = check_cache_range(root, cache, last,
5591                                         key.objectid - last);
5592                 if (ret)
5593                         break;
5594                 if (key.type == BTRFS_EXTENT_ITEM_KEY)
5595                         last = key.objectid + key.offset;
5596                 else
5597                         last = key.objectid + root->nodesize;
5598                 path.slots[0]++;
5599         }
5600
5601         if (last < cache->key.objectid + cache->key.offset)
5602                 ret = check_cache_range(root, cache, last,
5603                                         cache->key.objectid +
5604                                         cache->key.offset - last);
5605
5606 out:
5607         btrfs_release_path(&path);
5608
5609         if (!ret &&
5610             !RB_EMPTY_ROOT(&cache->free_space_ctl->free_space_offset)) {
5611                 fprintf(stderr, "There are still entries left in the space "
5612                         "cache\n");
5613                 ret = -EINVAL;
5614         }
5615
5616         return ret;
5617 }
5618
5619 static int check_space_cache(struct btrfs_root *root)
5620 {
5621         struct btrfs_block_group_cache *cache;
5622         u64 start = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
5623         int ret;
5624         int error = 0;
5625
5626         if (btrfs_super_cache_generation(root->fs_info->super_copy) != -1ULL &&
5627             btrfs_super_generation(root->fs_info->super_copy) !=
5628             btrfs_super_cache_generation(root->fs_info->super_copy)) {
5629                 printf("cache and super generation don't match, space cache "
5630                        "will be invalidated\n");
5631                 return 0;
5632         }
5633
5634         if (ctx.progress_enabled) {
5635                 ctx.tp = TASK_FREE_SPACE;
5636                 task_start(ctx.info);
5637         }
5638
5639         while (1) {
5640                 cache = btrfs_lookup_first_block_group(root->fs_info, start);
5641                 if (!cache)
5642                         break;
5643
5644                 start = cache->key.objectid + cache->key.offset;
5645                 if (!cache->free_space_ctl) {
5646                         if (btrfs_init_free_space_ctl(cache,
5647                                                       root->sectorsize)) {
5648                                 ret = -ENOMEM;
5649                                 break;
5650                         }
5651                 } else {
5652                         btrfs_remove_free_space_cache(cache);
5653                 }
5654
5655                 if (btrfs_fs_compat_ro(root->fs_info,
5656                                        BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE)) {
5657                         ret = exclude_super_stripes(root, cache);
5658                         if (ret) {
5659                                 fprintf(stderr, "could not exclude super stripes: %s\n",
5660                                         strerror(-ret));
5661                                 error++;
5662                                 continue;
5663                         }
5664                         ret = load_free_space_tree(root->fs_info, cache);
5665                         free_excluded_extents(root, cache);
5666                         if (ret < 0) {
5667                                 fprintf(stderr, "could not load free space tree: %s\n",
5668                                         strerror(-ret));
5669                                 error++;
5670                                 continue;
5671                         }
5672                         error += ret;
5673                 } else {
5674                         ret = load_free_space_cache(root->fs_info, cache);
5675                         if (!ret)
5676                                 continue;
5677                 }
5678
5679                 ret = verify_space_cache(root, cache);
5680                 if (ret) {
5681                         fprintf(stderr, "cache appears valid but isn't %Lu\n",
5682                                 cache->key.objectid);
5683                         error++;
5684                 }
5685         }
5686
5687         task_stop(ctx.info);
5688
5689         return error ? -EINVAL : 0;
5690 }
5691
5692 static int check_extent_csums(struct btrfs_root *root, u64 bytenr,
5693                         u64 num_bytes, unsigned long leaf_offset,
5694                         struct extent_buffer *eb) {
5695
5696         u64 offset = 0;
5697         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
5698         char *data;
5699         unsigned long csum_offset;
5700         u32 csum;
5701         u32 csum_expected;
5702         u64 read_len;
5703         u64 data_checked = 0;
5704         u64 tmp;
5705         int ret = 0;
5706         int mirror;
5707         int num_copies;
5708
5709         if (num_bytes % root->sectorsize)
5710                 return -EINVAL;
5711
5712         data = malloc(num_bytes);
5713         if (!data)
5714                 return -ENOMEM;
5715
5716         while (offset < num_bytes) {
5717                 mirror = 0;
5718 again:
5719                 read_len = num_bytes - offset;
5720                 /* read as much space once a time */
5721                 ret = read_extent_data(root, data + offset,
5722                                 bytenr + offset, &read_len, mirror);
5723                 if (ret)
5724                         goto out;
5725                 data_checked = 0;
5726                 /* verify every 4k data's checksum */
5727                 while (data_checked < read_len) {
5728                         csum = ~(u32)0;
5729                         tmp = offset + data_checked;
5730
5731                         csum = btrfs_csum_data(NULL, (char *)data + tmp,
5732                                                csum, root->sectorsize);
5733                         btrfs_csum_final(csum, (u8 *)&csum);
5734
5735                         csum_offset = leaf_offset +
5736                                  tmp / root->sectorsize * csum_size;
5737                         read_extent_buffer(eb, (char *)&csum_expected,
5738                                            csum_offset, csum_size);
5739                         /* try another mirror */
5740                         if (csum != csum_expected) {
5741                                 fprintf(stderr, "mirror %d bytenr %llu csum %u expected csum %u\n",
5742                                                 mirror, bytenr + tmp,
5743                                                 csum, csum_expected);
5744                                 num_copies = btrfs_num_copies(
5745                                                 &root->fs_info->mapping_tree,
5746                                                 bytenr, num_bytes);
5747                                 if (mirror < num_copies - 1) {
5748                                         mirror += 1;
5749                                         goto again;
5750                                 }
5751                         }
5752                         data_checked += root->sectorsize;
5753                 }
5754                 offset += read_len;
5755         }
5756 out:
5757         free(data);
5758         return ret;
5759 }
5760
5761 static int check_extent_exists(struct btrfs_root *root, u64 bytenr,
5762                                u64 num_bytes)
5763 {
5764         struct btrfs_path path;
5765         struct extent_buffer *leaf;
5766         struct btrfs_key key;
5767         int ret;
5768
5769         btrfs_init_path(&path);
5770         key.objectid = bytenr;
5771         key.type = BTRFS_EXTENT_ITEM_KEY;
5772         key.offset = (u64)-1;
5773
5774 again:
5775         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, &path,
5776                                 0, 0);
5777         if (ret < 0) {
5778                 fprintf(stderr, "Error looking up extent record %d\n", ret);
5779                 btrfs_release_path(&path);
5780                 return ret;
5781         } else if (ret) {
5782                 if (path.slots[0] > 0) {
5783                         path.slots[0]--;
5784                 } else {
5785                         ret = btrfs_prev_leaf(root, &path);
5786                         if (ret < 0) {
5787                                 goto out;
5788                         } else if (ret > 0) {
5789                                 ret = 0;
5790                                 goto out;
5791                         }
5792                 }
5793         }
5794
5795         btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
5796
5797         /*
5798          * Block group items come before extent items if they have the same
5799          * bytenr, so walk back one more just in case.  Dear future traveller,
5800          * first congrats on mastering time travel.  Now if it's not too much
5801          * trouble could you go back to 2006 and tell Chris to make the
5802          * BLOCK_GROUP_ITEM_KEY (and BTRFS_*_REF_KEY) lower than the
5803          * EXTENT_ITEM_KEY please?
5804          */
5805         while (key.type > BTRFS_EXTENT_ITEM_KEY) {
5806                 if (path.slots[0] > 0) {
5807                         path.slots[0]--;
5808                 } else {
5809                         ret = btrfs_prev_leaf(root, &path);
5810                         if (ret < 0) {
5811                                 goto out;
5812                         } else if (ret > 0) {
5813                                 ret = 0;
5814                                 goto out;
5815                         }
5816                 }
5817                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
5818         }
5819
5820         while (num_bytes) {
5821                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
5822                         ret = btrfs_next_leaf(root, &path);
5823                         if (ret < 0) {
5824                                 fprintf(stderr, "Error going to next leaf "
5825                                         "%d\n", ret);
5826                                 btrfs_release_path(&path);
5827                                 return ret;
5828                         } else if (ret) {
5829                                 break;
5830                         }
5831                 }
5832                 leaf = path.nodes[0];
5833                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
5834                 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
5835                         path.slots[0]++;
5836                         continue;
5837                 }
5838                 if (key.objectid + key.offset < bytenr) {
5839                         path.slots[0]++;
5840                         continue;
5841                 }
5842                 if (key.objectid > bytenr + num_bytes)
5843                         break;
5844
5845                 if (key.objectid == bytenr) {
5846                         if (key.offset >= num_bytes) {
5847                                 num_bytes = 0;
5848                                 break;
5849                         }
5850                         num_bytes -= key.offset;
5851                         bytenr += key.offset;
5852                 } else if (key.objectid < bytenr) {
5853                         if (key.objectid + key.offset >= bytenr + num_bytes) {
5854                                 num_bytes = 0;
5855                                 break;
5856                         }
5857                         num_bytes = (bytenr + num_bytes) -
5858                                 (key.objectid + key.offset);
5859                         bytenr = key.objectid + key.offset;
5860                 } else {
5861                         if (key.objectid + key.offset < bytenr + num_bytes) {
5862                                 u64 new_start = key.objectid + key.offset;
5863                                 u64 new_bytes = bytenr + num_bytes - new_start;
5864
5865                                 /*
5866                                  * Weird case, the extent is in the middle of
5867                                  * our range, we'll have to search one side
5868                                  * and then the other.  Not sure if this happens
5869                                  * in real life, but no harm in coding it up
5870                                  * anyway just in case.
5871                                  */
5872                                 btrfs_release_path(&path);
5873                                 ret = check_extent_exists(root, new_start,
5874                                                           new_bytes);
5875                                 if (ret) {
5876                                         fprintf(stderr, "Right section didn't "
5877                                                 "have a record\n");
5878                                         break;
5879                                 }
5880                                 num_bytes = key.objectid - bytenr;
5881                                 goto again;
5882                         }
5883                         num_bytes = key.objectid - bytenr;
5884                 }
5885                 path.slots[0]++;
5886         }
5887         ret = 0;
5888
5889 out:
5890         if (num_bytes && !ret) {
5891                 fprintf(stderr, "There are no extents for csum range "
5892                         "%Lu-%Lu\n", bytenr, bytenr+num_bytes);
5893                 ret = 1;
5894         }
5895
5896         btrfs_release_path(&path);
5897         return ret;
5898 }
5899
5900 static int check_csums(struct btrfs_root *root)
5901 {
5902         struct btrfs_path *path;
5903         struct extent_buffer *leaf;
5904         struct btrfs_key key;
5905         u64 offset = 0, num_bytes = 0;
5906         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
5907         int errors = 0;
5908         int ret;
5909         u64 data_len;
5910         unsigned long leaf_offset;
5911
5912         root = root->fs_info->csum_root;
5913         if (!extent_buffer_uptodate(root->node)) {
5914                 fprintf(stderr, "No valid csum tree found\n");
5915                 return -ENOENT;
5916         }
5917
5918         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
5919         key.type = BTRFS_EXTENT_CSUM_KEY;
5920         key.offset = 0;
5921
5922         path = btrfs_alloc_path();
5923         if (!path)
5924                 return -ENOMEM;
5925
5926         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5927         if (ret < 0) {
5928                 fprintf(stderr, "Error searching csum tree %d\n", ret);
5929                 btrfs_free_path(path);
5930                 return ret;
5931         }
5932
5933         if (ret > 0 && path->slots[0])
5934                 path->slots[0]--;
5935         ret = 0;
5936
5937         while (1) {
5938                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5939                         ret = btrfs_next_leaf(root, path);
5940                         if (ret < 0) {
5941                                 fprintf(stderr, "Error going to next leaf "
5942                                         "%d\n", ret);
5943                                 break;
5944                         }
5945                         if (ret)
5946                                 break;
5947                 }
5948                 leaf = path->nodes[0];
5949
5950                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5951                 if (key.type != BTRFS_EXTENT_CSUM_KEY) {
5952                         path->slots[0]++;
5953                         continue;
5954                 }
5955
5956                 data_len = (btrfs_item_size_nr(leaf, path->slots[0]) /
5957                               csum_size) * root->sectorsize;
5958                 if (!check_data_csum)
5959                         goto skip_csum_check;
5960                 leaf_offset = btrfs_item_ptr_offset(leaf, path->slots[0]);
5961                 ret = check_extent_csums(root, key.offset, data_len,
5962                                          leaf_offset, leaf);
5963                 if (ret)
5964                         break;
5965 skip_csum_check:
5966                 if (!num_bytes) {
5967                         offset = key.offset;
5968                 } else if (key.offset != offset + num_bytes) {
5969                         ret = check_extent_exists(root, offset, num_bytes);
5970                         if (ret) {
5971                                 fprintf(stderr, "Csum exists for %Lu-%Lu but "
5972                                         "there is no extent record\n",
5973                                         offset, offset+num_bytes);
5974                                 errors++;
5975                         }
5976                         offset = key.offset;
5977                         num_bytes = 0;
5978                 }
5979                 num_bytes += data_len;
5980                 path->slots[0]++;
5981         }
5982
5983         btrfs_free_path(path);
5984         return errors;
5985 }
5986
5987 static int is_dropped_key(struct btrfs_key *key,
5988                           struct btrfs_key *drop_key) {
5989         if (key->objectid < drop_key->objectid)
5990                 return 1;
5991         else if (key->objectid == drop_key->objectid) {
5992                 if (key->type < drop_key->type)
5993                         return 1;
5994                 else if (key->type == drop_key->type) {
5995                         if (key->offset < drop_key->offset)
5996                                 return 1;
5997                 }
5998         }
5999         return 0;
6000 }
6001
6002 /*
6003  * Here are the rules for FULL_BACKREF.
6004  *
6005  * 1) If BTRFS_HEADER_FLAG_RELOC is set then we have FULL_BACKREF set.
6006  * 2) If btrfs_header_owner(buf) no longer points to buf then we have
6007  *      FULL_BACKREF set.
6008  * 3) We cowed the block walking down a reloc tree.  This is impossible to tell
6009  *    if it happened after the relocation occurred since we'll have dropped the
6010  *    reloc root, so it's entirely possible to have FULL_BACKREF set on buf and
6011  *    have no real way to know for sure.
6012  *
6013  * We process the blocks one root at a time, and we start from the lowest root
6014  * objectid and go to the highest.  So we can just lookup the owner backref for
6015  * the record and if we don't find it then we know it doesn't exist and we have
6016  * a FULL BACKREF.
6017  *
6018  * FIXME: if we ever start reclaiming root objectid's then we need to fix this
6019  * assumption and simply indicate that we _think_ that the FULL BACKREF needs to
6020  * be set or not and then we can check later once we've gathered all the refs.
6021  */
6022 static int calc_extent_flag(struct btrfs_root *root,
6023                            struct cache_tree *extent_cache,
6024                            struct extent_buffer *buf,
6025                            struct root_item_record *ri,
6026                            u64 *flags)
6027 {
6028         struct extent_record *rec;
6029         struct cache_extent *cache;
6030         struct tree_backref *tback;
6031         u64 owner = 0;
6032
6033         cache = lookup_cache_extent(extent_cache, buf->start, 1);
6034         /* we have added this extent before */
6035         if (!cache)
6036                 return -ENOENT;
6037
6038         rec = container_of(cache, struct extent_record, cache);
6039
6040         /*
6041          * Except file/reloc tree, we can not have
6042          * FULL BACKREF MODE
6043          */
6044         if (ri->objectid < BTRFS_FIRST_FREE_OBJECTID)
6045                 goto normal;
6046         /*
6047          * root node
6048          */
6049         if (buf->start == ri->bytenr)
6050                 goto normal;
6051
6052         if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
6053                 goto full_backref;
6054
6055         owner = btrfs_header_owner(buf);
6056         if (owner == ri->objectid)
6057                 goto normal;
6058
6059         tback = find_tree_backref(rec, 0, owner);
6060         if (!tback)
6061                 goto full_backref;
6062 normal:
6063         *flags = 0;
6064         if (rec->flag_block_full_backref != FLAG_UNSET &&
6065             rec->flag_block_full_backref != 0)
6066                 rec->bad_full_backref = 1;
6067         return 0;
6068 full_backref:
6069         *flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6070         if (rec->flag_block_full_backref != FLAG_UNSET &&
6071             rec->flag_block_full_backref != 1)
6072                 rec->bad_full_backref = 1;
6073         return 0;
6074 }
6075
6076 static void report_mismatch_key_root(u8 key_type, u64 rootid)
6077 {
6078         fprintf(stderr, "Invalid key type(");
6079         print_key_type(stderr, 0, key_type);
6080         fprintf(stderr, ") found in root(");
6081         print_objectid(stderr, rootid, 0);
6082         fprintf(stderr, ")\n");
6083 }
6084
6085 /*
6086  * Check if the key is valid with its extent buffer.
6087  *
6088  * This is a early check in case invalid key exists in a extent buffer
6089  * This is not comprehensive yet, but should prevent wrong key/item passed
6090  * further
6091  */
6092 static int check_type_with_root(u64 rootid, u8 key_type)
6093 {
6094         switch (key_type) {
6095         /* Only valid in chunk tree */
6096         case BTRFS_DEV_ITEM_KEY:
6097         case BTRFS_CHUNK_ITEM_KEY:
6098                 if (rootid != BTRFS_CHUNK_TREE_OBJECTID)
6099                         goto err;
6100                 break;
6101         /* valid in csum and log tree */
6102         case BTRFS_CSUM_TREE_OBJECTID:
6103                 if (!(rootid == BTRFS_TREE_LOG_OBJECTID ||
6104                       is_fstree(rootid)))
6105                         goto err;
6106                 break;
6107         case BTRFS_EXTENT_ITEM_KEY:
6108         case BTRFS_METADATA_ITEM_KEY:
6109         case BTRFS_BLOCK_GROUP_ITEM_KEY:
6110                 if (rootid != BTRFS_EXTENT_TREE_OBJECTID)
6111                         goto err;
6112                 break;
6113         case BTRFS_ROOT_ITEM_KEY:
6114                 if (rootid != BTRFS_ROOT_TREE_OBJECTID)
6115                         goto err;
6116                 break;
6117         case BTRFS_DEV_EXTENT_KEY:
6118                 if (rootid != BTRFS_DEV_TREE_OBJECTID)
6119                         goto err;
6120                 break;
6121         }
6122         return 0;
6123 err:
6124         report_mismatch_key_root(key_type, rootid);
6125         return -EINVAL;
6126 }
6127
6128 static int run_next_block(struct btrfs_root *root,
6129                           struct block_info *bits,
6130                           int bits_nr,
6131                           u64 *last,
6132                           struct cache_tree *pending,
6133                           struct cache_tree *seen,
6134                           struct cache_tree *reada,
6135                           struct cache_tree *nodes,
6136                           struct cache_tree *extent_cache,
6137                           struct cache_tree *chunk_cache,
6138                           struct rb_root *dev_cache,
6139                           struct block_group_tree *block_group_cache,
6140                           struct device_extent_tree *dev_extent_cache,
6141                           struct root_item_record *ri)
6142 {
6143         struct extent_buffer *buf;
6144         struct extent_record *rec = NULL;
6145         u64 bytenr;
6146         u32 size;
6147         u64 parent;
6148         u64 owner;
6149         u64 flags;
6150         u64 ptr;
6151         u64 gen = 0;
6152         int ret = 0;
6153         int i;
6154         int nritems;
6155         struct btrfs_key key;
6156         struct cache_extent *cache;
6157         int reada_bits;
6158
6159         nritems = pick_next_pending(pending, reada, nodes, *last, bits,
6160                                     bits_nr, &reada_bits);
6161         if (nritems == 0)
6162                 return 1;
6163
6164         if (!reada_bits) {
6165                 for(i = 0; i < nritems; i++) {
6166                         ret = add_cache_extent(reada, bits[i].start,
6167                                                bits[i].size);
6168                         if (ret == -EEXIST)
6169                                 continue;
6170
6171                         /* fixme, get the parent transid */
6172                         readahead_tree_block(root, bits[i].start,
6173                                              bits[i].size, 0);
6174                 }
6175         }
6176         *last = bits[0].start;
6177         bytenr = bits[0].start;
6178         size = bits[0].size;
6179
6180         cache = lookup_cache_extent(pending, bytenr, size);
6181         if (cache) {
6182                 remove_cache_extent(pending, cache);
6183                 free(cache);
6184         }
6185         cache = lookup_cache_extent(reada, bytenr, size);
6186         if (cache) {
6187                 remove_cache_extent(reada, cache);
6188                 free(cache);
6189         }
6190         cache = lookup_cache_extent(nodes, bytenr, size);
6191         if (cache) {
6192                 remove_cache_extent(nodes, cache);
6193                 free(cache);
6194         }
6195         cache = lookup_cache_extent(extent_cache, bytenr, size);
6196         if (cache) {
6197                 rec = container_of(cache, struct extent_record, cache);
6198                 gen = rec->parent_generation;
6199         }
6200
6201         /* fixme, get the real parent transid */
6202         buf = read_tree_block(root, bytenr, size, gen);
6203         if (!extent_buffer_uptodate(buf)) {
6204                 record_bad_block_io(root->fs_info,
6205                                     extent_cache, bytenr, size);
6206                 goto out;
6207         }
6208
6209         nritems = btrfs_header_nritems(buf);
6210
6211         flags = 0;
6212         if (!init_extent_tree) {
6213                 ret = btrfs_lookup_extent_info(NULL, root, bytenr,
6214                                        btrfs_header_level(buf), 1, NULL,
6215                                        &flags);
6216                 if (ret < 0) {
6217                         ret = calc_extent_flag(root, extent_cache, buf, ri, &flags);
6218                         if (ret < 0) {
6219                                 fprintf(stderr, "Couldn't calc extent flags\n");
6220                                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6221                         }
6222                 }
6223         } else {
6224                 flags = 0;
6225                 ret = calc_extent_flag(root, extent_cache, buf, ri, &flags);
6226                 if (ret < 0) {
6227                         fprintf(stderr, "Couldn't calc extent flags\n");
6228                         flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6229                 }
6230         }
6231
6232         if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6233                 if (ri != NULL &&
6234                     ri->objectid != BTRFS_TREE_RELOC_OBJECTID &&
6235                     ri->objectid == btrfs_header_owner(buf)) {
6236                         /*
6237                          * Ok we got to this block from it's original owner and
6238                          * we have FULL_BACKREF set.  Relocation can leave
6239                          * converted blocks over so this is altogether possible,
6240                          * however it's not possible if the generation > the
6241                          * last snapshot, so check for this case.
6242                          */
6243                         if (!btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC) &&
6244                             btrfs_header_generation(buf) > ri->last_snapshot) {
6245                                 flags &= ~BTRFS_BLOCK_FLAG_FULL_BACKREF;
6246                                 rec->bad_full_backref = 1;
6247                         }
6248                 }
6249         } else {
6250                 if (ri != NULL &&
6251                     (ri->objectid == BTRFS_TREE_RELOC_OBJECTID ||
6252                      btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) {
6253                         flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6254                         rec->bad_full_backref = 1;
6255                 }
6256         }
6257
6258         if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6259                 rec->flag_block_full_backref = 1;
6260                 parent = bytenr;
6261                 owner = 0;
6262         } else {
6263                 rec->flag_block_full_backref = 0;
6264                 parent = 0;
6265                 owner = btrfs_header_owner(buf);
6266         }
6267
6268         ret = check_block(root, extent_cache, buf, flags);
6269         if (ret)
6270                 goto out;
6271
6272         if (btrfs_is_leaf(buf)) {
6273                 btree_space_waste += btrfs_leaf_free_space(root, buf);
6274                 for (i = 0; i < nritems; i++) {
6275                         struct btrfs_file_extent_item *fi;
6276                         btrfs_item_key_to_cpu(buf, &key, i);
6277                         /*
6278                          * Check key type against the leaf owner.
6279                          * Could filter quite a lot of early error if
6280                          * owner is correct
6281                          */
6282                         if (check_type_with_root(btrfs_header_owner(buf),
6283                                                  key.type)) {
6284                                 fprintf(stderr, "ignoring invalid key\n");
6285                                 continue;
6286                         }
6287                         if (key.type == BTRFS_EXTENT_ITEM_KEY) {
6288                                 process_extent_item(root, extent_cache, buf,
6289                                                     i);
6290                                 continue;
6291                         }
6292                         if (key.type == BTRFS_METADATA_ITEM_KEY) {
6293                                 process_extent_item(root, extent_cache, buf,
6294                                                     i);
6295                                 continue;
6296                         }
6297                         if (key.type == BTRFS_EXTENT_CSUM_KEY) {
6298                                 total_csum_bytes +=
6299                                         btrfs_item_size_nr(buf, i);
6300                                 continue;
6301                         }
6302                         if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6303                                 process_chunk_item(chunk_cache, &key, buf, i);
6304                                 continue;
6305                         }
6306                         if (key.type == BTRFS_DEV_ITEM_KEY) {
6307                                 process_device_item(dev_cache, &key, buf, i);
6308                                 continue;
6309                         }
6310                         if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
6311                                 process_block_group_item(block_group_cache,
6312                                         &key, buf, i);
6313                                 continue;
6314                         }
6315                         if (key.type == BTRFS_DEV_EXTENT_KEY) {
6316                                 process_device_extent_item(dev_extent_cache,
6317                                         &key, buf, i);
6318                                 continue;
6319
6320                         }
6321                         if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
6322 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6323                                 process_extent_ref_v0(extent_cache, buf, i);
6324 #else
6325                                 BUG();
6326 #endif
6327                                 continue;
6328                         }
6329
6330                         if (key.type == BTRFS_TREE_BLOCK_REF_KEY) {
6331                                 ret = add_tree_backref(extent_cache,
6332                                                 key.objectid, 0, key.offset, 0);
6333                                 if (ret < 0)
6334                                         error("add_tree_backref failed: %s",
6335                                               strerror(-ret));
6336                                 continue;
6337                         }
6338                         if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
6339                                 ret = add_tree_backref(extent_cache,
6340                                                 key.objectid, key.offset, 0, 0);
6341                                 if (ret < 0)
6342                                         error("add_tree_backref failed: %s",
6343                                               strerror(-ret));
6344                                 continue;
6345                         }
6346                         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
6347                                 struct btrfs_extent_data_ref *ref;
6348                                 ref = btrfs_item_ptr(buf, i,
6349                                                 struct btrfs_extent_data_ref);
6350                                 add_data_backref(extent_cache,
6351                                         key.objectid, 0,
6352                                         btrfs_extent_data_ref_root(buf, ref),
6353                                         btrfs_extent_data_ref_objectid(buf,
6354                                                                        ref),
6355                                         btrfs_extent_data_ref_offset(buf, ref),
6356                                         btrfs_extent_data_ref_count(buf, ref),
6357                                         0, root->sectorsize);
6358                                 continue;
6359                         }
6360                         if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
6361                                 struct btrfs_shared_data_ref *ref;
6362                                 ref = btrfs_item_ptr(buf, i,
6363                                                 struct btrfs_shared_data_ref);
6364                                 add_data_backref(extent_cache,
6365                                         key.objectid, key.offset, 0, 0, 0,
6366                                         btrfs_shared_data_ref_count(buf, ref),
6367                                         0, root->sectorsize);
6368                                 continue;
6369                         }
6370                         if (key.type == BTRFS_ORPHAN_ITEM_KEY) {
6371                                 struct bad_item *bad;
6372
6373                                 if (key.objectid == BTRFS_ORPHAN_OBJECTID)
6374                                         continue;
6375                                 if (!owner)
6376                                         continue;
6377                                 bad = malloc(sizeof(struct bad_item));
6378                                 if (!bad)
6379                                         continue;
6380                                 INIT_LIST_HEAD(&bad->list);
6381                                 memcpy(&bad->key, &key,
6382                                        sizeof(struct btrfs_key));
6383                                 bad->root_id = owner;
6384                                 list_add_tail(&bad->list, &delete_items);
6385                                 continue;
6386                         }
6387                         if (key.type != BTRFS_EXTENT_DATA_KEY)
6388                                 continue;
6389                         fi = btrfs_item_ptr(buf, i,
6390                                             struct btrfs_file_extent_item);
6391                         if (btrfs_file_extent_type(buf, fi) ==
6392                             BTRFS_FILE_EXTENT_INLINE)
6393                                 continue;
6394                         if (btrfs_file_extent_disk_bytenr(buf, fi) == 0)
6395                                 continue;
6396
6397                         data_bytes_allocated +=
6398                                 btrfs_file_extent_disk_num_bytes(buf, fi);
6399                         if (data_bytes_allocated < root->sectorsize) {
6400                                 abort();
6401                         }
6402                         data_bytes_referenced +=
6403                                 btrfs_file_extent_num_bytes(buf, fi);
6404                         add_data_backref(extent_cache,
6405                                 btrfs_file_extent_disk_bytenr(buf, fi),
6406                                 parent, owner, key.objectid, key.offset -
6407                                 btrfs_file_extent_offset(buf, fi), 1, 1,
6408                                 btrfs_file_extent_disk_num_bytes(buf, fi));
6409                 }
6410         } else {
6411                 int level;
6412                 struct btrfs_key first_key;
6413
6414                 first_key.objectid = 0;
6415
6416                 if (nritems > 0)
6417                         btrfs_item_key_to_cpu(buf, &first_key, 0);
6418                 level = btrfs_header_level(buf);
6419                 for (i = 0; i < nritems; i++) {
6420                         struct extent_record tmpl;
6421
6422                         ptr = btrfs_node_blockptr(buf, i);
6423                         size = root->nodesize;
6424                         btrfs_node_key_to_cpu(buf, &key, i);
6425                         if (ri != NULL) {
6426                                 if ((level == ri->drop_level)
6427                                     && is_dropped_key(&key, &ri->drop_key)) {
6428                                         continue;
6429                                 }
6430                         }
6431
6432                         memset(&tmpl, 0, sizeof(tmpl));
6433                         btrfs_cpu_key_to_disk(&tmpl.parent_key, &key);
6434                         tmpl.parent_generation = btrfs_node_ptr_generation(buf, i);
6435                         tmpl.start = ptr;
6436                         tmpl.nr = size;
6437                         tmpl.refs = 1;
6438                         tmpl.metadata = 1;
6439                         tmpl.max_size = size;
6440                         ret = add_extent_rec(extent_cache, &tmpl);
6441                         if (ret < 0)
6442                                 goto out;
6443
6444                         ret = add_tree_backref(extent_cache, ptr, parent,
6445                                         owner, 1);
6446                         if (ret < 0) {
6447                                 error("add_tree_backref failed: %s",
6448                                       strerror(-ret));
6449                                 continue;
6450                         }
6451
6452                         if (level > 1) {
6453                                 add_pending(nodes, seen, ptr, size);
6454                         } else {
6455                                 add_pending(pending, seen, ptr, size);
6456                         }
6457                 }
6458                 btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) -
6459                                       nritems) * sizeof(struct btrfs_key_ptr);
6460         }
6461         total_btree_bytes += buf->len;
6462         if (fs_root_objectid(btrfs_header_owner(buf)))
6463                 total_fs_tree_bytes += buf->len;
6464         if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID)
6465                 total_extent_tree_bytes += buf->len;
6466         if (!found_old_backref &&
6467             btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID &&
6468             btrfs_header_backref_rev(buf) == BTRFS_MIXED_BACKREF_REV &&
6469             !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
6470                 found_old_backref = 1;
6471 out:
6472         free_extent_buffer(buf);
6473         return ret;
6474 }
6475
6476 static int add_root_to_pending(struct extent_buffer *buf,
6477                                struct cache_tree *extent_cache,
6478                                struct cache_tree *pending,
6479                                struct cache_tree *seen,
6480                                struct cache_tree *nodes,
6481                                u64 objectid)
6482 {
6483         struct extent_record tmpl;
6484         int ret;
6485
6486         if (btrfs_header_level(buf) > 0)
6487                 add_pending(nodes, seen, buf->start, buf->len);
6488         else
6489                 add_pending(pending, seen, buf->start, buf->len);
6490
6491         memset(&tmpl, 0, sizeof(tmpl));
6492         tmpl.start = buf->start;
6493         tmpl.nr = buf->len;
6494         tmpl.is_root = 1;
6495         tmpl.refs = 1;
6496         tmpl.metadata = 1;
6497         tmpl.max_size = buf->len;
6498         add_extent_rec(extent_cache, &tmpl);
6499
6500         if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
6501             btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
6502                 ret = add_tree_backref(extent_cache, buf->start, buf->start,
6503                                 0, 1);
6504         else
6505                 ret = add_tree_backref(extent_cache, buf->start, 0, objectid,
6506                                 1);
6507         return ret;
6508 }
6509
6510 /* as we fix the tree, we might be deleting blocks that
6511  * we're tracking for repair.  This hook makes sure we
6512  * remove any backrefs for blocks as we are fixing them.
6513  */
6514 static int free_extent_hook(struct btrfs_trans_handle *trans,
6515                             struct btrfs_root *root,
6516                             u64 bytenr, u64 num_bytes, u64 parent,
6517                             u64 root_objectid, u64 owner, u64 offset,
6518                             int refs_to_drop)
6519 {
6520         struct extent_record *rec;
6521         struct cache_extent *cache;
6522         int is_data;
6523         struct cache_tree *extent_cache = root->fs_info->fsck_extent_cache;
6524
6525         is_data = owner >= BTRFS_FIRST_FREE_OBJECTID;
6526         cache = lookup_cache_extent(extent_cache, bytenr, num_bytes);
6527         if (!cache)
6528                 return 0;
6529
6530         rec = container_of(cache, struct extent_record, cache);
6531         if (is_data) {
6532                 struct data_backref *back;
6533                 back = find_data_backref(rec, parent, root_objectid, owner,
6534                                          offset, 1, bytenr, num_bytes);
6535                 if (!back)
6536                         goto out;
6537                 if (back->node.found_ref) {
6538                         back->found_ref -= refs_to_drop;
6539                         if (rec->refs)
6540                                 rec->refs -= refs_to_drop;
6541                 }
6542                 if (back->node.found_extent_tree) {
6543                         back->num_refs -= refs_to_drop;
6544                         if (rec->extent_item_refs)
6545                                 rec->extent_item_refs -= refs_to_drop;
6546                 }
6547                 if (back->found_ref == 0)
6548                         back->node.found_ref = 0;
6549                 if (back->num_refs == 0)
6550                         back->node.found_extent_tree = 0;
6551
6552                 if (!back->node.found_extent_tree && back->node.found_ref) {
6553                         list_del(&back->node.list);
6554                         free(back);
6555                 }
6556         } else {
6557                 struct tree_backref *back;
6558                 back = find_tree_backref(rec, parent, root_objectid);
6559                 if (!back)
6560                         goto out;
6561                 if (back->node.found_ref) {
6562                         if (rec->refs)
6563                                 rec->refs--;
6564                         back->node.found_ref = 0;
6565                 }
6566                 if (back->node.found_extent_tree) {
6567                         if (rec->extent_item_refs)
6568                                 rec->extent_item_refs--;
6569                         back->node.found_extent_tree = 0;
6570                 }
6571                 if (!back->node.found_extent_tree && back->node.found_ref) {
6572                         list_del(&back->node.list);
6573                         free(back);
6574                 }
6575         }
6576         maybe_free_extent_rec(extent_cache, rec);
6577 out:
6578         return 0;
6579 }
6580
6581 static int delete_extent_records(struct btrfs_trans_handle *trans,
6582                                  struct btrfs_root *root,
6583                                  struct btrfs_path *path,
6584                                  u64 bytenr, u64 new_len)
6585 {
6586         struct btrfs_key key;
6587         struct btrfs_key found_key;
6588         struct extent_buffer *leaf;
6589         int ret;
6590         int slot;
6591
6592
6593         key.objectid = bytenr;
6594         key.type = (u8)-1;
6595         key.offset = (u64)-1;
6596
6597         while(1) {
6598                 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
6599                                         &key, path, 0, 1);
6600                 if (ret < 0)
6601                         break;
6602
6603                 if (ret > 0) {
6604                         ret = 0;
6605                         if (path->slots[0] == 0)
6606                                 break;
6607                         path->slots[0]--;
6608                 }
6609                 ret = 0;
6610
6611                 leaf = path->nodes[0];
6612                 slot = path->slots[0];
6613
6614                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6615                 if (found_key.objectid != bytenr)
6616                         break;
6617
6618                 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
6619                     found_key.type != BTRFS_METADATA_ITEM_KEY &&
6620                     found_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
6621                     found_key.type != BTRFS_EXTENT_DATA_REF_KEY &&
6622                     found_key.type != BTRFS_EXTENT_REF_V0_KEY &&
6623                     found_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
6624                     found_key.type != BTRFS_SHARED_DATA_REF_KEY) {
6625                         btrfs_release_path(path);
6626                         if (found_key.type == 0) {
6627                                 if (found_key.offset == 0)
6628                                         break;
6629                                 key.offset = found_key.offset - 1;
6630                                 key.type = found_key.type;
6631                         }
6632                         key.type = found_key.type - 1;
6633                         key.offset = (u64)-1;
6634                         continue;
6635                 }
6636
6637                 fprintf(stderr, "repair deleting extent record: key %Lu %u %Lu\n",
6638                         found_key.objectid, found_key.type, found_key.offset);
6639
6640                 ret = btrfs_del_item(trans, root->fs_info->extent_root, path);
6641                 if (ret)
6642                         break;
6643                 btrfs_release_path(path);
6644
6645                 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
6646                     found_key.type == BTRFS_METADATA_ITEM_KEY) {
6647                         u64 bytes = (found_key.type == BTRFS_EXTENT_ITEM_KEY) ?
6648                                 found_key.offset : root->nodesize;
6649
6650                         ret = btrfs_update_block_group(trans, root, bytenr,
6651                                                        bytes, 0, 0);
6652                         if (ret)
6653                                 break;
6654                 }
6655         }
6656
6657         btrfs_release_path(path);
6658         return ret;
6659 }
6660
6661 /*
6662  * for a single backref, this will allocate a new extent
6663  * and add the backref to it.
6664  */
6665 static int record_extent(struct btrfs_trans_handle *trans,
6666                          struct btrfs_fs_info *info,
6667                          struct btrfs_path *path,
6668                          struct extent_record *rec,
6669                          struct extent_backref *back,
6670                          int allocated, u64 flags)
6671 {
6672         int ret;
6673         struct btrfs_root *extent_root = info->extent_root;
6674         struct extent_buffer *leaf;
6675         struct btrfs_key ins_key;
6676         struct btrfs_extent_item *ei;
6677         struct tree_backref *tback;
6678         struct data_backref *dback;
6679         struct btrfs_tree_block_info *bi;
6680
6681         if (!back->is_data)
6682                 rec->max_size = max_t(u64, rec->max_size,
6683                                     info->extent_root->nodesize);
6684
6685         if (!allocated) {
6686                 u32 item_size = sizeof(*ei);
6687
6688                 if (!back->is_data)
6689                         item_size += sizeof(*bi);
6690
6691                 ins_key.objectid = rec->start;
6692                 ins_key.offset = rec->max_size;
6693                 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
6694
6695                 ret = btrfs_insert_empty_item(trans, extent_root, path,
6696                                         &ins_key, item_size);
6697                 if (ret)
6698                         goto fail;
6699
6700                 leaf = path->nodes[0];
6701                 ei = btrfs_item_ptr(leaf, path->slots[0],
6702                                     struct btrfs_extent_item);
6703
6704                 btrfs_set_extent_refs(leaf, ei, 0);
6705                 btrfs_set_extent_generation(leaf, ei, rec->generation);
6706
6707                 if (back->is_data) {
6708                         btrfs_set_extent_flags(leaf, ei,
6709                                                BTRFS_EXTENT_FLAG_DATA);
6710                 } else {
6711                         struct btrfs_disk_key copy_key;;
6712
6713                         tback = to_tree_backref(back);
6714                         bi = (struct btrfs_tree_block_info *)(ei + 1);
6715                         memset_extent_buffer(leaf, 0, (unsigned long)bi,
6716                                              sizeof(*bi));
6717
6718                         btrfs_set_disk_key_objectid(&copy_key,
6719                                                     rec->info_objectid);
6720                         btrfs_set_disk_key_type(&copy_key, 0);
6721                         btrfs_set_disk_key_offset(&copy_key, 0);
6722
6723                         btrfs_set_tree_block_level(leaf, bi, rec->info_level);
6724                         btrfs_set_tree_block_key(leaf, bi, &copy_key);
6725
6726                         btrfs_set_extent_flags(leaf, ei,
6727                                                BTRFS_EXTENT_FLAG_TREE_BLOCK | flags);
6728                 }
6729
6730                 btrfs_mark_buffer_dirty(leaf);
6731                 ret = btrfs_update_block_group(trans, extent_root, rec->start,
6732                                                rec->max_size, 1, 0);
6733                 if (ret)
6734                         goto fail;
6735                 btrfs_release_path(path);
6736         }
6737
6738         if (back->is_data) {
6739                 u64 parent;
6740                 int i;
6741
6742                 dback = to_data_backref(back);
6743                 if (back->full_backref)
6744                         parent = dback->parent;
6745                 else
6746                         parent = 0;
6747
6748                 for (i = 0; i < dback->found_ref; i++) {
6749                         /* if parent != 0, we're doing a full backref
6750                          * passing BTRFS_FIRST_FREE_OBJECTID as the owner
6751                          * just makes the backref allocator create a data
6752                          * backref
6753                          */
6754                         ret = btrfs_inc_extent_ref(trans, info->extent_root,
6755                                                    rec->start, rec->max_size,
6756                                                    parent,
6757                                                    dback->root,
6758                                                    parent ?
6759                                                    BTRFS_FIRST_FREE_OBJECTID :
6760                                                    dback->owner,
6761                                                    dback->offset);
6762                         if (ret)
6763                                 break;
6764                 }
6765                 fprintf(stderr, "adding new data backref"
6766                                 " on %llu %s %llu owner %llu"
6767                                 " offset %llu found %d\n",
6768                                 (unsigned long long)rec->start,
6769                                 back->full_backref ?
6770                                 "parent" : "root",
6771                                 back->full_backref ?
6772                                 (unsigned long long)parent :
6773                                 (unsigned long long)dback->root,
6774                                 (unsigned long long)dback->owner,
6775                                 (unsigned long long)dback->offset,
6776                                 dback->found_ref);
6777         } else {
6778                 u64 parent;
6779
6780                 tback = to_tree_backref(back);
6781                 if (back->full_backref)
6782                         parent = tback->parent;
6783                 else
6784                         parent = 0;
6785
6786                 ret = btrfs_inc_extent_ref(trans, info->extent_root,
6787                                            rec->start, rec->max_size,
6788                                            parent, tback->root, 0, 0);
6789                 fprintf(stderr, "adding new tree backref on "
6790                         "start %llu len %llu parent %llu root %llu\n",
6791                         rec->start, rec->max_size, parent, tback->root);
6792         }
6793 fail:
6794         btrfs_release_path(path);
6795         return ret;
6796 }
6797
6798 static struct extent_entry *find_entry(struct list_head *entries,
6799                                        u64 bytenr, u64 bytes)
6800 {
6801         struct extent_entry *entry = NULL;
6802
6803         list_for_each_entry(entry, entries, list) {
6804                 if (entry->bytenr == bytenr && entry->bytes == bytes)
6805                         return entry;
6806         }
6807
6808         return NULL;
6809 }
6810
6811 static struct extent_entry *find_most_right_entry(struct list_head *entries)
6812 {
6813         struct extent_entry *entry, *best = NULL, *prev = NULL;
6814
6815         list_for_each_entry(entry, entries, list) {
6816                 if (!prev) {
6817                         prev = entry;
6818                         continue;
6819                 }
6820
6821                 /*
6822                  * If there are as many broken entries as entries then we know
6823                  * not to trust this particular entry.
6824                  */
6825                 if (entry->broken == entry->count)
6826                         continue;
6827
6828                 /*
6829                  * If our current entry == best then we can't be sure our best
6830                  * is really the best, so we need to keep searching.
6831                  */
6832                 if (best && best->count == entry->count) {
6833                         prev = entry;
6834                         best = NULL;
6835                         continue;
6836                 }
6837
6838                 /* Prev == entry, not good enough, have to keep searching */
6839                 if (!prev->broken && prev->count == entry->count)
6840                         continue;
6841
6842                 if (!best)
6843                         best = (prev->count > entry->count) ? prev : entry;
6844                 else if (best->count < entry->count)
6845                         best = entry;
6846                 prev = entry;
6847         }
6848
6849         return best;
6850 }
6851
6852 static int repair_ref(struct btrfs_fs_info *info, struct btrfs_path *path,
6853                       struct data_backref *dback, struct extent_entry *entry)
6854 {
6855         struct btrfs_trans_handle *trans;
6856         struct btrfs_root *root;
6857         struct btrfs_file_extent_item *fi;
6858         struct extent_buffer *leaf;
6859         struct btrfs_key key;
6860         u64 bytenr, bytes;
6861         int ret, err;
6862
6863         key.objectid = dback->root;
6864         key.type = BTRFS_ROOT_ITEM_KEY;
6865         key.offset = (u64)-1;
6866         root = btrfs_read_fs_root(info, &key);
6867         if (IS_ERR(root)) {
6868                 fprintf(stderr, "Couldn't find root for our ref\n");
6869                 return -EINVAL;
6870         }
6871
6872         /*
6873          * The backref points to the original offset of the extent if it was
6874          * split, so we need to search down to the offset we have and then walk
6875          * forward until we find the backref we're looking for.
6876          */
6877         key.objectid = dback->owner;
6878         key.type = BTRFS_EXTENT_DATA_KEY;
6879         key.offset = dback->offset;
6880         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6881         if (ret < 0) {
6882                 fprintf(stderr, "Error looking up ref %d\n", ret);
6883                 return ret;
6884         }
6885
6886         while (1) {
6887                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
6888                         ret = btrfs_next_leaf(root, path);
6889                         if (ret) {
6890                                 fprintf(stderr, "Couldn't find our ref, next\n");
6891                                 return -EINVAL;
6892                         }
6893                 }
6894                 leaf = path->nodes[0];
6895                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6896                 if (key.objectid != dback->owner ||
6897                     key.type != BTRFS_EXTENT_DATA_KEY) {
6898                         fprintf(stderr, "Couldn't find our ref, search\n");
6899                         return -EINVAL;
6900                 }
6901                 fi = btrfs_item_ptr(leaf, path->slots[0],
6902                                     struct btrfs_file_extent_item);
6903                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6904                 bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
6905
6906                 if (bytenr == dback->disk_bytenr && bytes == dback->bytes)
6907                         break;
6908                 path->slots[0]++;
6909         }
6910
6911         btrfs_release_path(path);
6912
6913         trans = btrfs_start_transaction(root, 1);
6914         if (IS_ERR(trans))
6915                 return PTR_ERR(trans);
6916
6917         /*
6918          * Ok we have the key of the file extent we want to fix, now we can cow
6919          * down to the thing and fix it.
6920          */
6921         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6922         if (ret < 0) {
6923                 fprintf(stderr, "Error cowing down to ref [%Lu, %u, %Lu]: %d\n",
6924                         key.objectid, key.type, key.offset, ret);
6925                 goto out;
6926         }
6927         if (ret > 0) {
6928                 fprintf(stderr, "Well that's odd, we just found this key "
6929                         "[%Lu, %u, %Lu]\n", key.objectid, key.type,
6930                         key.offset);
6931                 ret = -EINVAL;
6932                 goto out;
6933         }
6934         leaf = path->nodes[0];
6935         fi = btrfs_item_ptr(leaf, path->slots[0],
6936                             struct btrfs_file_extent_item);
6937
6938         if (btrfs_file_extent_compression(leaf, fi) &&
6939             dback->disk_bytenr != entry->bytenr) {
6940                 fprintf(stderr, "Ref doesn't match the record start and is "
6941                         "compressed, please take a btrfs-image of this file "
6942                         "system and send it to a btrfs developer so they can "
6943                         "complete this functionality for bytenr %Lu\n",
6944                         dback->disk_bytenr);
6945                 ret = -EINVAL;
6946                 goto out;
6947         }
6948
6949         if (dback->node.broken && dback->disk_bytenr != entry->bytenr) {
6950                 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
6951         } else if (dback->disk_bytenr > entry->bytenr) {
6952                 u64 off_diff, offset;
6953
6954                 off_diff = dback->disk_bytenr - entry->bytenr;
6955                 offset = btrfs_file_extent_offset(leaf, fi);
6956                 if (dback->disk_bytenr + offset +
6957                     btrfs_file_extent_num_bytes(leaf, fi) >
6958                     entry->bytenr + entry->bytes) {
6959                         fprintf(stderr, "Ref is past the entry end, please "
6960                                 "take a btrfs-image of this file system and "
6961                                 "send it to a btrfs developer, ref %Lu\n",
6962                                 dback->disk_bytenr);
6963                         ret = -EINVAL;
6964                         goto out;
6965                 }
6966                 offset += off_diff;
6967                 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
6968                 btrfs_set_file_extent_offset(leaf, fi, offset);
6969         } else if (dback->disk_bytenr < entry->bytenr) {
6970                 u64 offset;
6971
6972                 offset = btrfs_file_extent_offset(leaf, fi);
6973                 if (dback->disk_bytenr + offset < entry->bytenr) {
6974                         fprintf(stderr, "Ref is before the entry start, please"
6975                                 " take a btrfs-image of this file system and "
6976                                 "send it to a btrfs developer, ref %Lu\n",
6977                                 dback->disk_bytenr);
6978                         ret = -EINVAL;
6979                         goto out;
6980                 }
6981
6982                 offset += dback->disk_bytenr;
6983                 offset -= entry->bytenr;
6984                 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
6985                 btrfs_set_file_extent_offset(leaf, fi, offset);
6986         }
6987
6988         btrfs_set_file_extent_disk_num_bytes(leaf, fi, entry->bytes);
6989
6990         /*
6991          * Chances are if disk_num_bytes were wrong then so is ram_bytes, but
6992          * only do this if we aren't using compression, otherwise it's a
6993          * trickier case.
6994          */
6995         if (!btrfs_file_extent_compression(leaf, fi))
6996                 btrfs_set_file_extent_ram_bytes(leaf, fi, entry->bytes);
6997         else
6998                 printf("ram bytes may be wrong?\n");
6999         btrfs_mark_buffer_dirty(leaf);
7000 out:
7001         err = btrfs_commit_transaction(trans, root);
7002         btrfs_release_path(path);
7003         return ret ? ret : err;
7004 }
7005
7006 static int verify_backrefs(struct btrfs_fs_info *info, struct btrfs_path *path,
7007                            struct extent_record *rec)
7008 {
7009         struct extent_backref *back;
7010         struct data_backref *dback;
7011         struct extent_entry *entry, *best = NULL;
7012         LIST_HEAD(entries);
7013         int nr_entries = 0;
7014         int broken_entries = 0;
7015         int ret = 0;
7016         short mismatch = 0;
7017
7018         /*
7019          * Metadata is easy and the backrefs should always agree on bytenr and
7020          * size, if not we've got bigger issues.
7021          */
7022         if (rec->metadata)
7023                 return 0;
7024
7025         list_for_each_entry(back, &rec->backrefs, list) {
7026                 if (back->full_backref || !back->is_data)
7027                         continue;
7028
7029                 dback = to_data_backref(back);
7030
7031                 /*
7032                  * We only pay attention to backrefs that we found a real
7033                  * backref for.
7034                  */
7035                 if (dback->found_ref == 0)
7036                         continue;
7037
7038                 /*
7039                  * For now we only catch when the bytes don't match, not the
7040                  * bytenr.  We can easily do this at the same time, but I want
7041                  * to have a fs image to test on before we just add repair
7042                  * functionality willy-nilly so we know we won't screw up the
7043                  * repair.
7044                  */
7045
7046                 entry = find_entry(&entries, dback->disk_bytenr,
7047                                    dback->bytes);
7048                 if (!entry) {
7049                         entry = malloc(sizeof(struct extent_entry));
7050                         if (!entry) {
7051                                 ret = -ENOMEM;
7052                                 goto out;
7053                         }
7054                         memset(entry, 0, sizeof(*entry));
7055                         entry->bytenr = dback->disk_bytenr;
7056                         entry->bytes = dback->bytes;
7057                         list_add_tail(&entry->list, &entries);
7058                         nr_entries++;
7059                 }
7060
7061                 /*
7062                  * If we only have on entry we may think the entries agree when
7063                  * in reality they don't so we have to do some extra checking.
7064                  */
7065                 if (dback->disk_bytenr != rec->start ||
7066                     dback->bytes != rec->nr || back->broken)
7067                         mismatch = 1;
7068
7069                 if (back->broken) {
7070                         entry->broken++;
7071                         broken_entries++;
7072                 }
7073
7074                 entry->count++;
7075         }
7076
7077         /* Yay all the backrefs agree, carry on good sir */
7078         if (nr_entries <= 1 && !mismatch)
7079                 goto out;
7080
7081         fprintf(stderr, "attempting to repair backref discrepency for bytenr "
7082                 "%Lu\n", rec->start);
7083
7084         /*
7085          * First we want to see if the backrefs can agree amongst themselves who
7086          * is right, so figure out which one of the entries has the highest
7087          * count.
7088          */
7089         best = find_most_right_entry(&entries);
7090
7091         /*
7092          * Ok so we may have an even split between what the backrefs think, so
7093          * this is where we use the extent ref to see what it thinks.
7094          */
7095         if (!best) {
7096                 entry = find_entry(&entries, rec->start, rec->nr);
7097                 if (!entry && (!broken_entries || !rec->found_rec)) {
7098                         fprintf(stderr, "Backrefs don't agree with each other "
7099                                 "and extent record doesn't agree with anybody,"
7100                                 " so we can't fix bytenr %Lu bytes %Lu\n",
7101                                 rec->start, rec->nr);
7102                         ret = -EINVAL;
7103                         goto out;
7104                 } else if (!entry) {
7105                         /*
7106                          * Ok our backrefs were broken, we'll assume this is the
7107                          * correct value and add an entry for this range.
7108                          */
7109                         entry = malloc(sizeof(struct extent_entry));
7110                         if (!entry) {
7111                                 ret = -ENOMEM;
7112                                 goto out;
7113                         }
7114                         memset(entry, 0, sizeof(*entry));
7115                         entry->bytenr = rec->start;
7116                         entry->bytes = rec->nr;
7117                         list_add_tail(&entry->list, &entries);
7118                         nr_entries++;
7119                 }
7120                 entry->count++;
7121                 best = find_most_right_entry(&entries);
7122                 if (!best) {
7123                         fprintf(stderr, "Backrefs and extent record evenly "
7124                                 "split on who is right, this is going to "
7125                                 "require user input to fix bytenr %Lu bytes "
7126                                 "%Lu\n", rec->start, rec->nr);
7127                         ret = -EINVAL;
7128                         goto out;
7129                 }
7130         }
7131
7132         /*
7133          * I don't think this can happen currently as we'll abort() if we catch
7134          * this case higher up, but in case somebody removes that we still can't
7135          * deal with it properly here yet, so just bail out of that's the case.
7136          */
7137         if (best->bytenr != rec->start) {
7138                 fprintf(stderr, "Extent start and backref starts don't match, "
7139                         "please use btrfs-image on this file system and send "
7140                         "it to a btrfs developer so they can make fsck fix "
7141                         "this particular case.  bytenr is %Lu, bytes is %Lu\n",
7142                         rec->start, rec->nr);
7143                 ret = -EINVAL;
7144                 goto out;
7145         }
7146
7147         /*
7148          * Ok great we all agreed on an extent record, let's go find the real
7149          * references and fix up the ones that don't match.
7150          */
7151         list_for_each_entry(back, &rec->backrefs, list) {
7152                 if (back->full_backref || !back->is_data)
7153                         continue;
7154
7155                 dback = to_data_backref(back);
7156
7157                 /*
7158                  * Still ignoring backrefs that don't have a real ref attached
7159                  * to them.
7160                  */
7161                 if (dback->found_ref == 0)
7162                         continue;
7163
7164                 if (dback->bytes == best->bytes &&
7165                     dback->disk_bytenr == best->bytenr)
7166                         continue;
7167
7168                 ret = repair_ref(info, path, dback, best);
7169                 if (ret)
7170                         goto out;
7171         }
7172
7173         /*
7174          * Ok we messed with the actual refs, which means we need to drop our
7175          * entire cache and go back and rescan.  I know this is a huge pain and
7176          * adds a lot of extra work, but it's the only way to be safe.  Once all
7177          * the backrefs agree we may not need to do anything to the extent
7178          * record itself.
7179          */
7180         ret = -EAGAIN;
7181 out:
7182         while (!list_empty(&entries)) {
7183                 entry = list_entry(entries.next, struct extent_entry, list);
7184                 list_del_init(&entry->list);
7185                 free(entry);
7186         }
7187         return ret;
7188 }
7189
7190 static int process_duplicates(struct btrfs_root *root,
7191                               struct cache_tree *extent_cache,
7192                               struct extent_record *rec)
7193 {
7194         struct extent_record *good, *tmp;
7195         struct cache_extent *cache;
7196         int ret;
7197
7198         /*
7199          * If we found a extent record for this extent then return, or if we
7200          * have more than one duplicate we are likely going to need to delete
7201          * something.
7202          */
7203         if (rec->found_rec || rec->num_duplicates > 1)
7204                 return 0;
7205
7206         /* Shouldn't happen but just in case */
7207         BUG_ON(!rec->num_duplicates);
7208
7209         /*
7210          * So this happens if we end up with a backref that doesn't match the
7211          * actual extent entry.  So either the backref is bad or the extent
7212          * entry is bad.  Either way we want to have the extent_record actually
7213          * reflect what we found in the extent_tree, so we need to take the
7214          * duplicate out and use that as the extent_record since the only way we
7215          * get a duplicate is if we find a real life BTRFS_EXTENT_ITEM_KEY.
7216          */
7217         remove_cache_extent(extent_cache, &rec->cache);
7218
7219         good = to_extent_record(rec->dups.next);
7220         list_del_init(&good->list);
7221         INIT_LIST_HEAD(&good->backrefs);
7222         INIT_LIST_HEAD(&good->dups);
7223         good->cache.start = good->start;
7224         good->cache.size = good->nr;
7225         good->content_checked = 0;
7226         good->owner_ref_checked = 0;
7227         good->num_duplicates = 0;
7228         good->refs = rec->refs;
7229         list_splice_init(&rec->backrefs, &good->backrefs);
7230         while (1) {
7231                 cache = lookup_cache_extent(extent_cache, good->start,
7232                                             good->nr);
7233                 if (!cache)
7234                         break;
7235                 tmp = container_of(cache, struct extent_record, cache);
7236
7237                 /*
7238                  * If we find another overlapping extent and it's found_rec is
7239                  * set then it's a duplicate and we need to try and delete
7240                  * something.
7241                  */
7242                 if (tmp->found_rec || tmp->num_duplicates > 0) {
7243                         if (list_empty(&good->list))
7244                                 list_add_tail(&good->list,
7245                                               &duplicate_extents);
7246                         good->num_duplicates += tmp->num_duplicates + 1;
7247                         list_splice_init(&tmp->dups, &good->dups);
7248                         list_del_init(&tmp->list);
7249                         list_add_tail(&tmp->list, &good->dups);
7250                         remove_cache_extent(extent_cache, &tmp->cache);
7251                         continue;
7252                 }
7253
7254                 /*
7255                  * Ok we have another non extent item backed extent rec, so lets
7256                  * just add it to this extent and carry on like we did above.
7257                  */
7258                 good->refs += tmp->refs;
7259                 list_splice_init(&tmp->backrefs, &good->backrefs);
7260                 remove_cache_extent(extent_cache, &tmp->cache);
7261                 free(tmp);
7262         }
7263         ret = insert_cache_extent(extent_cache, &good->cache);
7264         BUG_ON(ret);
7265         free(rec);
7266         return good->num_duplicates ? 0 : 1;
7267 }
7268
7269 static int delete_duplicate_records(struct btrfs_root *root,
7270                                     struct extent_record *rec)
7271 {
7272         struct btrfs_trans_handle *trans;
7273         LIST_HEAD(delete_list);
7274         struct btrfs_path *path;
7275         struct extent_record *tmp, *good, *n;
7276         int nr_del = 0;
7277         int ret = 0, err;
7278         struct btrfs_key key;
7279
7280         path = btrfs_alloc_path();
7281         if (!path) {
7282                 ret = -ENOMEM;
7283                 goto out;
7284         }
7285
7286         good = rec;
7287         /* Find the record that covers all of the duplicates. */
7288         list_for_each_entry(tmp, &rec->dups, list) {
7289                 if (good->start < tmp->start)
7290                         continue;
7291                 if (good->nr > tmp->nr)
7292                         continue;
7293
7294                 if (tmp->start + tmp->nr < good->start + good->nr) {
7295                         fprintf(stderr, "Ok we have overlapping extents that "
7296                                 "aren't completely covered by each other, this "
7297                                 "is going to require more careful thought.  "
7298                                 "The extents are [%Lu-%Lu] and [%Lu-%Lu]\n",
7299                                 tmp->start, tmp->nr, good->start, good->nr);
7300                         abort();
7301                 }
7302                 good = tmp;
7303         }
7304
7305         if (good != rec)
7306                 list_add_tail(&rec->list, &delete_list);
7307
7308         list_for_each_entry_safe(tmp, n, &rec->dups, list) {
7309                 if (tmp == good)
7310                         continue;
7311                 list_move_tail(&tmp->list, &delete_list);
7312         }
7313
7314         root = root->fs_info->extent_root;
7315         trans = btrfs_start_transaction(root, 1);
7316         if (IS_ERR(trans)) {
7317                 ret = PTR_ERR(trans);
7318                 goto out;
7319         }
7320
7321         list_for_each_entry(tmp, &delete_list, list) {
7322                 if (tmp->found_rec == 0)
7323                         continue;
7324                 key.objectid = tmp->start;
7325                 key.type = BTRFS_EXTENT_ITEM_KEY;
7326                 key.offset = tmp->nr;
7327
7328                 /* Shouldn't happen but just in case */
7329                 if (tmp->metadata) {
7330                         fprintf(stderr, "Well this shouldn't happen, extent "
7331                                 "record overlaps but is metadata? "
7332                                 "[%Lu, %Lu]\n", tmp->start, tmp->nr);
7333                         abort();
7334                 }
7335
7336                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7337                 if (ret) {
7338                         if (ret > 0)
7339                                 ret = -EINVAL;
7340                         break;
7341                 }
7342                 ret = btrfs_del_item(trans, root, path);
7343                 if (ret)
7344                         break;
7345                 btrfs_release_path(path);
7346                 nr_del++;
7347         }
7348         err = btrfs_commit_transaction(trans, root);
7349         if (err && !ret)
7350                 ret = err;
7351 out:
7352         while (!list_empty(&delete_list)) {
7353                 tmp = to_extent_record(delete_list.next);
7354                 list_del_init(&tmp->list);
7355                 if (tmp == rec)
7356                         continue;
7357                 free(tmp);
7358         }
7359
7360         while (!list_empty(&rec->dups)) {
7361                 tmp = to_extent_record(rec->dups.next);
7362                 list_del_init(&tmp->list);
7363                 free(tmp);
7364         }
7365
7366         btrfs_free_path(path);
7367
7368         if (!ret && !nr_del)
7369                 rec->num_duplicates = 0;
7370
7371         return ret ? ret : nr_del;
7372 }
7373
7374 static int find_possible_backrefs(struct btrfs_fs_info *info,
7375                                   struct btrfs_path *path,
7376                                   struct cache_tree *extent_cache,
7377                                   struct extent_record *rec)
7378 {
7379         struct btrfs_root *root;
7380         struct extent_backref *back;
7381         struct data_backref *dback;
7382         struct cache_extent *cache;
7383         struct btrfs_file_extent_item *fi;
7384         struct btrfs_key key;
7385         u64 bytenr, bytes;
7386         int ret;
7387
7388         list_for_each_entry(back, &rec->backrefs, list) {
7389                 /* Don't care about full backrefs (poor unloved backrefs) */
7390                 if (back->full_backref || !back->is_data)
7391                         continue;
7392
7393                 dback = to_data_backref(back);
7394
7395                 /* We found this one, we don't need to do a lookup */
7396                 if (dback->found_ref)
7397                         continue;
7398
7399                 key.objectid = dback->root;
7400                 key.type = BTRFS_ROOT_ITEM_KEY;
7401                 key.offset = (u64)-1;
7402
7403                 root = btrfs_read_fs_root(info, &key);
7404
7405                 /* No root, definitely a bad ref, skip */
7406                 if (IS_ERR(root) && PTR_ERR(root) == -ENOENT)
7407                         continue;
7408                 /* Other err, exit */
7409                 if (IS_ERR(root))
7410                         return PTR_ERR(root);
7411
7412                 key.objectid = dback->owner;
7413                 key.type = BTRFS_EXTENT_DATA_KEY;
7414                 key.offset = dback->offset;
7415                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7416                 if (ret) {
7417                         btrfs_release_path(path);
7418                         if (ret < 0)
7419                                 return ret;
7420                         /* Didn't find it, we can carry on */
7421                         ret = 0;
7422                         continue;
7423                 }
7424
7425                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
7426                                     struct btrfs_file_extent_item);
7427                 bytenr = btrfs_file_extent_disk_bytenr(path->nodes[0], fi);
7428                 bytes = btrfs_file_extent_disk_num_bytes(path->nodes[0], fi);
7429                 btrfs_release_path(path);
7430                 cache = lookup_cache_extent(extent_cache, bytenr, 1);
7431                 if (cache) {
7432                         struct extent_record *tmp;
7433                         tmp = container_of(cache, struct extent_record, cache);
7434
7435                         /*
7436                          * If we found an extent record for the bytenr for this
7437                          * particular backref then we can't add it to our
7438                          * current extent record.  We only want to add backrefs
7439                          * that don't have a corresponding extent item in the
7440                          * extent tree since they likely belong to this record
7441                          * and we need to fix it if it doesn't match bytenrs.
7442                          */
7443                         if  (tmp->found_rec)
7444                                 continue;
7445                 }
7446
7447                 dback->found_ref += 1;
7448                 dback->disk_bytenr = bytenr;
7449                 dback->bytes = bytes;
7450
7451                 /*
7452                  * Set this so the verify backref code knows not to trust the
7453                  * values in this backref.
7454                  */
7455                 back->broken = 1;
7456         }
7457
7458         return 0;
7459 }
7460
7461 /*
7462  * Record orphan data ref into corresponding root.
7463  *
7464  * Return 0 if the extent item contains data ref and recorded.
7465  * Return 1 if the extent item contains no useful data ref
7466  *   On that case, it may contains only shared_dataref or metadata backref
7467  *   or the file extent exists(this should be handled by the extent bytenr
7468  *   recovery routine)
7469  * Return <0 if something goes wrong.
7470  */
7471 static int record_orphan_data_extents(struct btrfs_fs_info *fs_info,
7472                                       struct extent_record *rec)
7473 {
7474         struct btrfs_key key;
7475         struct btrfs_root *dest_root;
7476         struct extent_backref *back;
7477         struct data_backref *dback;
7478         struct orphan_data_extent *orphan;
7479         struct btrfs_path *path;
7480         int recorded_data_ref = 0;
7481         int ret = 0;
7482
7483         if (rec->metadata)
7484                 return 1;
7485         path = btrfs_alloc_path();
7486         if (!path)
7487                 return -ENOMEM;
7488         list_for_each_entry(back, &rec->backrefs, list) {
7489                 if (back->full_backref || !back->is_data ||
7490                     !back->found_extent_tree)
7491                         continue;
7492                 dback = to_data_backref(back);
7493                 if (dback->found_ref)
7494                         continue;
7495                 key.objectid = dback->root;
7496                 key.type = BTRFS_ROOT_ITEM_KEY;
7497                 key.offset = (u64)-1;
7498
7499                 dest_root = btrfs_read_fs_root(fs_info, &key);
7500
7501                 /* For non-exist root we just skip it */
7502                 if (IS_ERR(dest_root) || !dest_root)
7503                         continue;
7504
7505                 key.objectid = dback->owner;
7506                 key.type = BTRFS_EXTENT_DATA_KEY;
7507                 key.offset = dback->offset;
7508
7509                 ret = btrfs_search_slot(NULL, dest_root, &key, path, 0, 0);
7510                 btrfs_release_path(path);
7511                 /*
7512                  * For ret < 0, it's OK since the fs-tree may be corrupted,
7513                  * we need to record it for inode/file extent rebuild.
7514                  * For ret > 0, we record it only for file extent rebuild.
7515                  * For ret == 0, the file extent exists but only bytenr
7516                  * mismatch, let the original bytenr fix routine to handle,
7517                  * don't record it.
7518                  */
7519                 if (ret == 0)
7520                         continue;
7521                 ret = 0;
7522                 orphan = malloc(sizeof(*orphan));
7523                 if (!orphan) {
7524                         ret = -ENOMEM;
7525                         goto out;
7526                 }
7527                 INIT_LIST_HEAD(&orphan->list);
7528                 orphan->root = dback->root;
7529                 orphan->objectid = dback->owner;
7530                 orphan->offset = dback->offset;
7531                 orphan->disk_bytenr = rec->cache.start;
7532                 orphan->disk_len = rec->cache.size;
7533                 list_add(&dest_root->orphan_data_extents, &orphan->list);
7534                 recorded_data_ref = 1;
7535         }
7536 out:
7537         btrfs_free_path(path);
7538         if (!ret)
7539                 return !recorded_data_ref;
7540         else
7541                 return ret;
7542 }
7543
7544 /*
7545  * when an incorrect extent item is found, this will delete
7546  * all of the existing entries for it and recreate them
7547  * based on what the tree scan found.
7548  */
7549 static int fixup_extent_refs(struct btrfs_fs_info *info,
7550                              struct cache_tree *extent_cache,
7551                              struct extent_record *rec)
7552 {
7553         struct btrfs_trans_handle *trans = NULL;
7554         int ret;
7555         struct btrfs_path *path;
7556         struct list_head *cur = rec->backrefs.next;
7557         struct cache_extent *cache;
7558         struct extent_backref *back;
7559         int allocated = 0;
7560         u64 flags = 0;
7561
7562         if (rec->flag_block_full_backref)
7563                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7564
7565         path = btrfs_alloc_path();
7566         if (!path)
7567                 return -ENOMEM;
7568
7569         if (rec->refs != rec->extent_item_refs && !rec->metadata) {
7570                 /*
7571                  * Sometimes the backrefs themselves are so broken they don't
7572                  * get attached to any meaningful rec, so first go back and
7573                  * check any of our backrefs that we couldn't find and throw
7574                  * them into the list if we find the backref so that
7575                  * verify_backrefs can figure out what to do.
7576                  */
7577                 ret = find_possible_backrefs(info, path, extent_cache, rec);
7578                 if (ret < 0)
7579                         goto out;
7580         }
7581
7582         /* step one, make sure all of the backrefs agree */
7583         ret = verify_backrefs(info, path, rec);
7584         if (ret < 0)
7585                 goto out;
7586
7587         trans = btrfs_start_transaction(info->extent_root, 1);
7588         if (IS_ERR(trans)) {
7589                 ret = PTR_ERR(trans);
7590                 goto out;
7591         }
7592
7593         /* step two, delete all the existing records */
7594         ret = delete_extent_records(trans, info->extent_root, path,
7595                                     rec->start, rec->max_size);
7596
7597         if (ret < 0)
7598                 goto out;
7599
7600         /* was this block corrupt?  If so, don't add references to it */
7601         cache = lookup_cache_extent(info->corrupt_blocks,
7602                                     rec->start, rec->max_size);
7603         if (cache) {
7604                 ret = 0;
7605                 goto out;
7606         }
7607
7608         /* step three, recreate all the refs we did find */
7609         while(cur != &rec->backrefs) {
7610                 back = to_extent_backref(cur);
7611                 cur = cur->next;
7612
7613                 /*
7614                  * if we didn't find any references, don't create a
7615                  * new extent record
7616                  */
7617                 if (!back->found_ref)
7618                         continue;
7619
7620                 rec->bad_full_backref = 0;
7621                 ret = record_extent(trans, info, path, rec, back, allocated, flags);
7622                 allocated = 1;
7623
7624                 if (ret)
7625                         goto out;
7626         }
7627 out:
7628         if (trans) {
7629                 int err = btrfs_commit_transaction(trans, info->extent_root);
7630                 if (!ret)
7631                         ret = err;
7632         }
7633
7634         btrfs_free_path(path);
7635         return ret;
7636 }
7637
7638 static int fixup_extent_flags(struct btrfs_fs_info *fs_info,
7639                               struct extent_record *rec)
7640 {
7641         struct btrfs_trans_handle *trans;
7642         struct btrfs_root *root = fs_info->extent_root;
7643         struct btrfs_path *path;
7644         struct btrfs_extent_item *ei;
7645         struct btrfs_key key;
7646         u64 flags;
7647         int ret = 0;
7648
7649         key.objectid = rec->start;
7650         if (rec->metadata) {
7651                 key.type = BTRFS_METADATA_ITEM_KEY;
7652                 key.offset = rec->info_level;
7653         } else {
7654                 key.type = BTRFS_EXTENT_ITEM_KEY;
7655                 key.offset = rec->max_size;
7656         }
7657
7658         path = btrfs_alloc_path();
7659         if (!path)
7660                 return -ENOMEM;
7661
7662         trans = btrfs_start_transaction(root, 0);
7663         if (IS_ERR(trans)) {
7664                 btrfs_free_path(path);
7665                 return PTR_ERR(trans);
7666         }
7667
7668         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
7669         if (ret < 0) {
7670                 btrfs_free_path(path);
7671                 btrfs_commit_transaction(trans, root);
7672                 return ret;
7673         } else if (ret) {
7674                 fprintf(stderr, "Didn't find extent for %llu\n",
7675                         (unsigned long long)rec->start);
7676                 btrfs_free_path(path);
7677                 btrfs_commit_transaction(trans, root);
7678                 return -ENOENT;
7679         }
7680
7681         ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
7682                             struct btrfs_extent_item);
7683         flags = btrfs_extent_flags(path->nodes[0], ei);
7684         if (rec->flag_block_full_backref) {
7685                 fprintf(stderr, "setting full backref on %llu\n",
7686                         (unsigned long long)key.objectid);
7687                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7688         } else {
7689                 fprintf(stderr, "clearing full backref on %llu\n",
7690                         (unsigned long long)key.objectid);
7691                 flags &= ~BTRFS_BLOCK_FLAG_FULL_BACKREF;
7692         }
7693         btrfs_set_extent_flags(path->nodes[0], ei, flags);
7694         btrfs_mark_buffer_dirty(path->nodes[0]);
7695         btrfs_free_path(path);
7696         return btrfs_commit_transaction(trans, root);
7697 }
7698
7699 /* right now we only prune from the extent allocation tree */
7700 static int prune_one_block(struct btrfs_trans_handle *trans,
7701                            struct btrfs_fs_info *info,
7702                            struct btrfs_corrupt_block *corrupt)
7703 {
7704         int ret;
7705         struct btrfs_path path;
7706         struct extent_buffer *eb;
7707         u64 found;
7708         int slot;
7709         int nritems;
7710         int level = corrupt->level + 1;
7711
7712         btrfs_init_path(&path);
7713 again:
7714         /* we want to stop at the parent to our busted block */
7715         path.lowest_level = level;
7716
7717         ret = btrfs_search_slot(trans, info->extent_root,
7718                                 &corrupt->key, &path, -1, 1);
7719
7720         if (ret < 0)
7721                 goto out;
7722
7723         eb = path.nodes[level];
7724         if (!eb) {
7725                 ret = -ENOENT;
7726                 goto out;
7727         }
7728
7729         /*
7730          * hopefully the search gave us the block we want to prune,
7731          * lets try that first
7732          */
7733         slot = path.slots[level];
7734         found =  btrfs_node_blockptr(eb, slot);
7735         if (found == corrupt->cache.start)
7736                 goto del_ptr;
7737
7738         nritems = btrfs_header_nritems(eb);
7739
7740         /* the search failed, lets scan this node and hope we find it */
7741         for (slot = 0; slot < nritems; slot++) {
7742                 found =  btrfs_node_blockptr(eb, slot);
7743                 if (found == corrupt->cache.start)
7744                         goto del_ptr;
7745         }
7746         /*
7747          * we couldn't find the bad block.  TODO, search all the nodes for pointers
7748          * to this block
7749          */
7750         if (eb == info->extent_root->node) {
7751                 ret = -ENOENT;
7752                 goto out;
7753         } else {
7754                 level++;
7755                 btrfs_release_path(&path);
7756                 goto again;
7757         }
7758
7759 del_ptr:
7760         printk("deleting pointer to block %Lu\n", corrupt->cache.start);
7761         ret = btrfs_del_ptr(trans, info->extent_root, &path, level, slot);
7762
7763 out:
7764         btrfs_release_path(&path);
7765         return ret;
7766 }
7767
7768 static int prune_corrupt_blocks(struct btrfs_fs_info *info)
7769 {
7770         struct btrfs_trans_handle *trans = NULL;
7771         struct cache_extent *cache;
7772         struct btrfs_corrupt_block *corrupt;
7773
7774         while (1) {
7775                 cache = search_cache_extent(info->corrupt_blocks, 0);
7776                 if (!cache)
7777                         break;
7778                 if (!trans) {
7779                         trans = btrfs_start_transaction(info->extent_root, 1);
7780                         if (IS_ERR(trans))
7781                                 return PTR_ERR(trans);
7782                 }
7783                 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
7784                 prune_one_block(trans, info, corrupt);
7785                 remove_cache_extent(info->corrupt_blocks, cache);
7786         }
7787         if (trans)
7788                 return btrfs_commit_transaction(trans, info->extent_root);
7789         return 0;
7790 }
7791
7792 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info)
7793 {
7794         struct btrfs_block_group_cache *cache;
7795         u64 start, end;
7796         int ret;
7797
7798         while (1) {
7799                 ret = find_first_extent_bit(&fs_info->free_space_cache, 0,
7800                                             &start, &end, EXTENT_DIRTY);
7801                 if (ret)
7802                         break;
7803                 clear_extent_dirty(&fs_info->free_space_cache, start, end,
7804                                    GFP_NOFS);
7805         }
7806
7807         start = 0;
7808         while (1) {
7809                 cache = btrfs_lookup_first_block_group(fs_info, start);
7810                 if (!cache)
7811                         break;
7812                 if (cache->cached)
7813                         cache->cached = 0;
7814                 start = cache->key.objectid + cache->key.offset;
7815         }
7816 }
7817
7818 static int check_extent_refs(struct btrfs_root *root,
7819                              struct cache_tree *extent_cache)
7820 {
7821         struct extent_record *rec;
7822         struct cache_extent *cache;
7823         int err = 0;
7824         int ret = 0;
7825         int fixed = 0;
7826         int had_dups = 0;
7827         int recorded = 0;
7828
7829         if (repair) {
7830                 /*
7831                  * if we're doing a repair, we have to make sure
7832                  * we don't allocate from the problem extents.
7833                  * In the worst case, this will be all the
7834                  * extents in the FS
7835                  */
7836                 cache = search_cache_extent(extent_cache, 0);
7837                 while(cache) {
7838                         rec = container_of(cache, struct extent_record, cache);
7839                         set_extent_dirty(root->fs_info->excluded_extents,
7840                                          rec->start,
7841                                          rec->start + rec->max_size - 1,
7842                                          GFP_NOFS);
7843                         cache = next_cache_extent(cache);
7844                 }
7845
7846                 /* pin down all the corrupted blocks too */
7847                 cache = search_cache_extent(root->fs_info->corrupt_blocks, 0);
7848                 while(cache) {
7849                         set_extent_dirty(root->fs_info->excluded_extents,
7850                                          cache->start,
7851                                          cache->start + cache->size - 1,
7852                                          GFP_NOFS);
7853                         cache = next_cache_extent(cache);
7854                 }
7855                 prune_corrupt_blocks(root->fs_info);
7856                 reset_cached_block_groups(root->fs_info);
7857         }
7858
7859         reset_cached_block_groups(root->fs_info);
7860
7861         /*
7862          * We need to delete any duplicate entries we find first otherwise we
7863          * could mess up the extent tree when we have backrefs that actually
7864          * belong to a different extent item and not the weird duplicate one.
7865          */
7866         while (repair && !list_empty(&duplicate_extents)) {
7867                 rec = to_extent_record(duplicate_extents.next);
7868                 list_del_init(&rec->list);
7869
7870                 /* Sometimes we can find a backref before we find an actual
7871                  * extent, so we need to process it a little bit to see if there
7872                  * truly are multiple EXTENT_ITEM_KEY's for the same range, or
7873                  * if this is a backref screwup.  If we need to delete stuff
7874                  * process_duplicates() will return 0, otherwise it will return
7875                  * 1 and we
7876                  */
7877                 if (process_duplicates(root, extent_cache, rec))
7878                         continue;
7879                 ret = delete_duplicate_records(root, rec);
7880                 if (ret < 0)
7881                         return ret;
7882                 /*
7883                  * delete_duplicate_records will return the number of entries
7884                  * deleted, so if it's greater than 0 then we know we actually
7885                  * did something and we need to remove.
7886                  */
7887                 if (ret)
7888                         had_dups = 1;
7889         }
7890
7891         if (had_dups)
7892                 return -EAGAIN;
7893
7894         while(1) {
7895                 int cur_err = 0;
7896
7897                 fixed = 0;
7898                 recorded = 0;
7899                 cache = search_cache_extent(extent_cache, 0);
7900                 if (!cache)
7901                         break;
7902                 rec = container_of(cache, struct extent_record, cache);
7903                 if (rec->num_duplicates) {
7904                         fprintf(stderr, "extent item %llu has multiple extent "
7905                                 "items\n", (unsigned long long)rec->start);
7906                         err = 1;
7907                         cur_err = 1;
7908                 }
7909
7910                 if (rec->refs != rec->extent_item_refs) {
7911                         fprintf(stderr, "ref mismatch on [%llu %llu] ",
7912                                 (unsigned long long)rec->start,
7913                                 (unsigned long long)rec->nr);
7914                         fprintf(stderr, "extent item %llu, found %llu\n",
7915                                 (unsigned long long)rec->extent_item_refs,
7916                                 (unsigned long long)rec->refs);
7917                         ret = record_orphan_data_extents(root->fs_info, rec);
7918                         if (ret < 0)
7919                                 goto repair_abort;
7920                         if (ret == 0) {
7921                                 recorded = 1;
7922                         } else {
7923                                 /*
7924                                  * we can't use the extent to repair file
7925                                  * extent, let the fallback method handle it.
7926                                  */
7927                                 if (!fixed && repair) {
7928                                         ret = fixup_extent_refs(
7929                                                         root->fs_info,
7930                                                         extent_cache, rec);
7931                                         if (ret)
7932                                                 goto repair_abort;
7933                                         fixed = 1;
7934                                 }
7935                         }
7936                         err = 1;
7937                         cur_err = 1;
7938                 }
7939                 if (all_backpointers_checked(rec, 1)) {
7940                         fprintf(stderr, "backpointer mismatch on [%llu %llu]\n",
7941                                 (unsigned long long)rec->start,
7942                                 (unsigned long long)rec->nr);
7943
7944                         if (!fixed && !recorded && repair) {
7945                                 ret = fixup_extent_refs(root->fs_info,
7946                                                         extent_cache, rec);
7947                                 if (ret)
7948                                         goto repair_abort;
7949                                 fixed = 1;
7950                         }
7951                         cur_err = 1;
7952                         err = 1;
7953                 }
7954                 if (!rec->owner_ref_checked) {
7955                         fprintf(stderr, "owner ref check failed [%llu %llu]\n",
7956                                 (unsigned long long)rec->start,
7957                                 (unsigned long long)rec->nr);
7958                         if (!fixed && !recorded && repair) {
7959                                 ret = fixup_extent_refs(root->fs_info,
7960                                                         extent_cache, rec);
7961                                 if (ret)
7962                                         goto repair_abort;
7963                                 fixed = 1;
7964                         }
7965                         err = 1;
7966                         cur_err = 1;
7967                 }
7968                 if (rec->bad_full_backref) {
7969                         fprintf(stderr, "bad full backref, on [%llu]\n",
7970                                 (unsigned long long)rec->start);
7971                         if (repair) {
7972                                 ret = fixup_extent_flags(root->fs_info, rec);
7973                                 if (ret)
7974                                         goto repair_abort;
7975                                 fixed = 1;
7976                         }
7977                         err = 1;
7978                         cur_err = 1;
7979                 }
7980                 /*
7981                  * Although it's not a extent ref's problem, we reuse this
7982                  * routine for error reporting.
7983                  * No repair function yet.
7984                  */
7985                 if (rec->crossing_stripes) {
7986                         fprintf(stderr,
7987                                 "bad metadata [%llu, %llu) crossing stripe boundary\n",
7988                                 rec->start, rec->start + rec->max_size);
7989                         err = 1;
7990                         cur_err = 1;
7991                 }
7992
7993                 if (rec->wrong_chunk_type) {
7994                         fprintf(stderr,
7995                                 "bad extent [%llu, %llu), type mismatch with chunk\n",
7996                                 rec->start, rec->start + rec->max_size);
7997                         err = 1;
7998                         cur_err = 1;
7999                 }
8000
8001                 remove_cache_extent(extent_cache, cache);
8002                 free_all_extent_backrefs(rec);
8003                 if (!init_extent_tree && repair && (!cur_err || fixed))
8004                         clear_extent_dirty(root->fs_info->excluded_extents,
8005                                            rec->start,
8006                                            rec->start + rec->max_size - 1,
8007                                            GFP_NOFS);
8008                 free(rec);
8009         }
8010 repair_abort:
8011         if (repair) {
8012                 if (ret && ret != -EAGAIN) {
8013                         fprintf(stderr, "failed to repair damaged filesystem, aborting\n");
8014                         exit(1);
8015                 } else if (!ret) {
8016                         struct btrfs_trans_handle *trans;
8017
8018                         root = root->fs_info->extent_root;
8019                         trans = btrfs_start_transaction(root, 1);
8020                         if (IS_ERR(trans)) {
8021                                 ret = PTR_ERR(trans);
8022                                 goto repair_abort;
8023                         }
8024
8025                         btrfs_fix_block_accounting(trans, root);
8026                         ret = btrfs_commit_transaction(trans, root);
8027                         if (ret)
8028                                 goto repair_abort;
8029                 }
8030                 if (err)
8031                         fprintf(stderr, "repaired damaged extent references\n");
8032                 return ret;
8033         }
8034         return err;
8035 }
8036
8037 u64 calc_stripe_length(u64 type, u64 length, int num_stripes)
8038 {
8039         u64 stripe_size;
8040
8041         if (type & BTRFS_BLOCK_GROUP_RAID0) {
8042                 stripe_size = length;
8043                 stripe_size /= num_stripes;
8044         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
8045                 stripe_size = length * 2;
8046                 stripe_size /= num_stripes;
8047         } else if (type & BTRFS_BLOCK_GROUP_RAID5) {
8048                 stripe_size = length;
8049                 stripe_size /= (num_stripes - 1);
8050         } else if (type & BTRFS_BLOCK_GROUP_RAID6) {
8051                 stripe_size = length;
8052                 stripe_size /= (num_stripes - 2);
8053         } else {
8054                 stripe_size = length;
8055         }
8056         return stripe_size;
8057 }
8058
8059 /*
8060  * Check the chunk with its block group/dev list ref:
8061  * Return 0 if all refs seems valid.
8062  * Return 1 if part of refs seems valid, need later check for rebuild ref
8063  * like missing block group and needs to search extent tree to rebuild them.
8064  * Return -1 if essential refs are missing and unable to rebuild.
8065  */
8066 static int check_chunk_refs(struct chunk_record *chunk_rec,
8067                             struct block_group_tree *block_group_cache,
8068                             struct device_extent_tree *dev_extent_cache,
8069                             int silent)
8070 {
8071         struct cache_extent *block_group_item;
8072         struct block_group_record *block_group_rec;
8073         struct cache_extent *dev_extent_item;
8074         struct device_extent_record *dev_extent_rec;
8075         u64 devid;
8076         u64 offset;
8077         u64 length;
8078         int metadump_v2 = 0;
8079         int i;
8080         int ret = 0;
8081
8082         block_group_item = lookup_cache_extent(&block_group_cache->tree,
8083                                                chunk_rec->offset,
8084                                                chunk_rec->length);
8085         if (block_group_item) {
8086                 block_group_rec = container_of(block_group_item,
8087                                                struct block_group_record,
8088                                                cache);
8089                 if (chunk_rec->length != block_group_rec->offset ||
8090                     chunk_rec->offset != block_group_rec->objectid ||
8091                     (!metadump_v2 &&
8092                      chunk_rec->type_flags != block_group_rec->flags)) {
8093                         if (!silent)
8094                                 fprintf(stderr,
8095                                         "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) mismatch with block group[%llu, %u, %llu]: offset(%llu), objectid(%llu), flags(%llu)\n",
8096                                         chunk_rec->objectid,
8097                                         chunk_rec->type,
8098                                         chunk_rec->offset,
8099                                         chunk_rec->length,
8100                                         chunk_rec->offset,
8101                                         chunk_rec->type_flags,
8102                                         block_group_rec->objectid,
8103                                         block_group_rec->type,
8104                                         block_group_rec->offset,
8105                                         block_group_rec->offset,
8106                                         block_group_rec->objectid,
8107                                         block_group_rec->flags);
8108                         ret = -1;
8109                 } else {
8110                         list_del_init(&block_group_rec->list);
8111                         chunk_rec->bg_rec = block_group_rec;
8112                 }
8113         } else {
8114                 if (!silent)
8115                         fprintf(stderr,
8116                                 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) is not found in block group\n",
8117                                 chunk_rec->objectid,
8118                                 chunk_rec->type,
8119                                 chunk_rec->offset,
8120                                 chunk_rec->length,
8121                                 chunk_rec->offset,
8122                                 chunk_rec->type_flags);
8123                 ret = 1;
8124         }
8125
8126         if (metadump_v2)
8127                 return ret;
8128
8129         length = calc_stripe_length(chunk_rec->type_flags, chunk_rec->length,
8130                                     chunk_rec->num_stripes);
8131         for (i = 0; i < chunk_rec->num_stripes; ++i) {
8132                 devid = chunk_rec->stripes[i].devid;
8133                 offset = chunk_rec->stripes[i].offset;
8134                 dev_extent_item = lookup_cache_extent2(&dev_extent_cache->tree,
8135                                                        devid, offset, length);
8136                 if (dev_extent_item) {
8137                         dev_extent_rec = container_of(dev_extent_item,
8138                                                 struct device_extent_record,
8139                                                 cache);
8140                         if (dev_extent_rec->objectid != devid ||
8141                             dev_extent_rec->offset != offset ||
8142                             dev_extent_rec->chunk_offset != chunk_rec->offset ||
8143                             dev_extent_rec->length != length) {
8144                                 if (!silent)
8145                                         fprintf(stderr,
8146                                                 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] dismatch dev extent[%llu, %llu, %llu]\n",
8147                                                 chunk_rec->objectid,
8148                                                 chunk_rec->type,
8149                                                 chunk_rec->offset,
8150                                                 chunk_rec->stripes[i].devid,
8151                                                 chunk_rec->stripes[i].offset,
8152                                                 dev_extent_rec->objectid,
8153                                                 dev_extent_rec->offset,
8154                                                 dev_extent_rec->length);
8155                                 ret = -1;
8156                         } else {
8157                                 list_move(&dev_extent_rec->chunk_list,
8158                                           &chunk_rec->dextents);
8159                         }
8160                 } else {
8161                         if (!silent)
8162                                 fprintf(stderr,
8163                                         "Chunk[%llu, %u, %llu] stripe[%llu, %llu] is not found in dev extent\n",
8164                                         chunk_rec->objectid,
8165                                         chunk_rec->type,
8166                                         chunk_rec->offset,
8167                                         chunk_rec->stripes[i].devid,
8168                                         chunk_rec->stripes[i].offset);
8169                         ret = -1;
8170                 }
8171         }
8172         return ret;
8173 }
8174
8175 /* check btrfs_chunk -> btrfs_dev_extent / btrfs_block_group_item */
8176 int check_chunks(struct cache_tree *chunk_cache,
8177                  struct block_group_tree *block_group_cache,
8178                  struct device_extent_tree *dev_extent_cache,
8179                  struct list_head *good, struct list_head *bad,
8180                  struct list_head *rebuild, int silent)
8181 {
8182         struct cache_extent *chunk_item;
8183         struct chunk_record *chunk_rec;
8184         struct block_group_record *bg_rec;
8185         struct device_extent_record *dext_rec;
8186         int err;
8187         int ret = 0;
8188
8189         chunk_item = first_cache_extent(chunk_cache);
8190         while (chunk_item) {
8191                 chunk_rec = container_of(chunk_item, struct chunk_record,
8192                                          cache);
8193                 err = check_chunk_refs(chunk_rec, block_group_cache,
8194                                        dev_extent_cache, silent);
8195                 if (err < 0)
8196                         ret = err;
8197                 if (err == 0 && good)
8198                         list_add_tail(&chunk_rec->list, good);
8199                 if (err > 0 && rebuild)
8200                         list_add_tail(&chunk_rec->list, rebuild);
8201                 if (err < 0 && bad)
8202                         list_add_tail(&chunk_rec->list, bad);
8203                 chunk_item = next_cache_extent(chunk_item);
8204         }
8205
8206         list_for_each_entry(bg_rec, &block_group_cache->block_groups, list) {
8207                 if (!silent)
8208                         fprintf(stderr,
8209                                 "Block group[%llu, %llu] (flags = %llu) didn't find the relative chunk.\n",
8210                                 bg_rec->objectid,
8211                                 bg_rec->offset,
8212                                 bg_rec->flags);
8213                 if (!ret)
8214                         ret = 1;
8215         }
8216
8217         list_for_each_entry(dext_rec, &dev_extent_cache->no_chunk_orphans,
8218                             chunk_list) {
8219                 if (!silent)
8220                         fprintf(stderr,
8221                                 "Device extent[%llu, %llu, %llu] didn't find the relative chunk.\n",
8222                                 dext_rec->objectid,
8223                                 dext_rec->offset,
8224                                 dext_rec->length);
8225                 if (!ret)
8226                         ret = 1;
8227         }
8228         return ret;
8229 }
8230
8231
8232 static int check_device_used(struct device_record *dev_rec,
8233                              struct device_extent_tree *dext_cache)
8234 {
8235         struct cache_extent *cache;
8236         struct device_extent_record *dev_extent_rec;
8237         u64 total_byte = 0;
8238
8239         cache = search_cache_extent2(&dext_cache->tree, dev_rec->devid, 0);
8240         while (cache) {
8241                 dev_extent_rec = container_of(cache,
8242                                               struct device_extent_record,
8243                                               cache);
8244                 if (dev_extent_rec->objectid != dev_rec->devid)
8245                         break;
8246
8247                 list_del_init(&dev_extent_rec->device_list);
8248                 total_byte += dev_extent_rec->length;
8249                 cache = next_cache_extent(cache);
8250         }
8251
8252         if (total_byte != dev_rec->byte_used) {
8253                 fprintf(stderr,
8254                         "Dev extent's total-byte(%llu) is not equal to byte-used(%llu) in dev[%llu, %u, %llu]\n",
8255                         total_byte, dev_rec->byte_used, dev_rec->objectid,
8256                         dev_rec->type, dev_rec->offset);
8257                 return -1;
8258         } else {
8259                 return 0;
8260         }
8261 }
8262
8263 /* check btrfs_dev_item -> btrfs_dev_extent */
8264 static int check_devices(struct rb_root *dev_cache,
8265                          struct device_extent_tree *dev_extent_cache)
8266 {
8267         struct rb_node *dev_node;
8268         struct device_record *dev_rec;
8269         struct device_extent_record *dext_rec;
8270         int err;
8271         int ret = 0;
8272
8273         dev_node = rb_first(dev_cache);
8274         while (dev_node) {
8275                 dev_rec = container_of(dev_node, struct device_record, node);
8276                 err = check_device_used(dev_rec, dev_extent_cache);
8277                 if (err)
8278                         ret = err;
8279
8280                 dev_node = rb_next(dev_node);
8281         }
8282         list_for_each_entry(dext_rec, &dev_extent_cache->no_device_orphans,
8283                             device_list) {
8284                 fprintf(stderr,
8285                         "Device extent[%llu, %llu, %llu] didn't find its device.\n",
8286                         dext_rec->objectid, dext_rec->offset, dext_rec->length);
8287                 if (!ret)
8288                         ret = 1;
8289         }
8290         return ret;
8291 }
8292
8293 static int add_root_item_to_list(struct list_head *head,
8294                                   u64 objectid, u64 bytenr, u64 last_snapshot,
8295                                   u8 level, u8 drop_level,
8296                                   int level_size, struct btrfs_key *drop_key)
8297 {
8298
8299         struct root_item_record *ri_rec;
8300         ri_rec = malloc(sizeof(*ri_rec));
8301         if (!ri_rec)
8302                 return -ENOMEM;
8303         ri_rec->bytenr = bytenr;
8304         ri_rec->objectid = objectid;
8305         ri_rec->level = level;
8306         ri_rec->level_size = level_size;
8307         ri_rec->drop_level = drop_level;
8308         ri_rec->last_snapshot = last_snapshot;
8309         if (drop_key)
8310                 memcpy(&ri_rec->drop_key, drop_key, sizeof(*drop_key));
8311         list_add_tail(&ri_rec->list, head);
8312
8313         return 0;
8314 }
8315
8316 static void free_root_item_list(struct list_head *list)
8317 {
8318         struct root_item_record *ri_rec;
8319
8320         while (!list_empty(list)) {
8321                 ri_rec = list_first_entry(list, struct root_item_record,
8322                                           list);
8323                 list_del_init(&ri_rec->list);
8324                 free(ri_rec);
8325         }
8326 }
8327
8328 static int deal_root_from_list(struct list_head *list,
8329                                struct btrfs_root *root,
8330                                struct block_info *bits,
8331                                int bits_nr,
8332                                struct cache_tree *pending,
8333                                struct cache_tree *seen,
8334                                struct cache_tree *reada,
8335                                struct cache_tree *nodes,
8336                                struct cache_tree *extent_cache,
8337                                struct cache_tree *chunk_cache,
8338                                struct rb_root *dev_cache,
8339                                struct block_group_tree *block_group_cache,
8340                                struct device_extent_tree *dev_extent_cache)
8341 {
8342         int ret = 0;
8343         u64 last;
8344
8345         while (!list_empty(list)) {
8346                 struct root_item_record *rec;
8347                 struct extent_buffer *buf;
8348                 rec = list_entry(list->next,
8349                                  struct root_item_record, list);
8350                 last = 0;
8351                 buf = read_tree_block(root->fs_info->tree_root,
8352                                       rec->bytenr, rec->level_size, 0);
8353                 if (!extent_buffer_uptodate(buf)) {
8354                         free_extent_buffer(buf);
8355                         ret = -EIO;
8356                         break;
8357                 }
8358                 ret = add_root_to_pending(buf, extent_cache, pending,
8359                                     seen, nodes, rec->objectid);
8360                 if (ret < 0)
8361                         break;
8362                 /*
8363                  * To rebuild extent tree, we need deal with snapshot
8364                  * one by one, otherwise we deal with node firstly which
8365                  * can maximize readahead.
8366                  */
8367                 while (1) {
8368                         ret = run_next_block(root, bits, bits_nr, &last,
8369                                              pending, seen, reada, nodes,
8370                                              extent_cache, chunk_cache,
8371                                              dev_cache, block_group_cache,
8372                                              dev_extent_cache, rec);
8373                         if (ret != 0)
8374                                 break;
8375                 }
8376                 free_extent_buffer(buf);
8377                 list_del(&rec->list);
8378                 free(rec);
8379                 if (ret < 0)
8380                         break;
8381         }
8382         while (ret >= 0) {
8383                 ret = run_next_block(root, bits, bits_nr, &last, pending, seen,
8384                                      reada, nodes, extent_cache, chunk_cache,
8385                                      dev_cache, block_group_cache,
8386                                      dev_extent_cache, NULL);
8387                 if (ret != 0) {
8388                         if (ret > 0)
8389                                 ret = 0;
8390                         break;
8391                 }
8392         }
8393         return ret;
8394 }
8395
8396 static int check_chunks_and_extents(struct btrfs_root *root)
8397 {
8398         struct rb_root dev_cache;
8399         struct cache_tree chunk_cache;
8400         struct block_group_tree block_group_cache;
8401         struct device_extent_tree dev_extent_cache;
8402         struct cache_tree extent_cache;
8403         struct cache_tree seen;
8404         struct cache_tree pending;
8405         struct cache_tree reada;
8406         struct cache_tree nodes;
8407         struct extent_io_tree excluded_extents;
8408         struct cache_tree corrupt_blocks;
8409         struct btrfs_path path;
8410         struct btrfs_key key;
8411         struct btrfs_key found_key;
8412         int ret, err = 0;
8413         struct block_info *bits;
8414         int bits_nr;
8415         struct extent_buffer *leaf;
8416         int slot;
8417         struct btrfs_root_item ri;
8418         struct list_head dropping_trees;
8419         struct list_head normal_trees;
8420         struct btrfs_root *root1;
8421         u64 objectid;
8422         u32 level_size;
8423         u8 level;
8424
8425         dev_cache = RB_ROOT;
8426         cache_tree_init(&chunk_cache);
8427         block_group_tree_init(&block_group_cache);
8428         device_extent_tree_init(&dev_extent_cache);
8429
8430         cache_tree_init(&extent_cache);
8431         cache_tree_init(&seen);
8432         cache_tree_init(&pending);
8433         cache_tree_init(&nodes);
8434         cache_tree_init(&reada);
8435         cache_tree_init(&corrupt_blocks);
8436         extent_io_tree_init(&excluded_extents);
8437         INIT_LIST_HEAD(&dropping_trees);
8438         INIT_LIST_HEAD(&normal_trees);
8439
8440         if (repair) {
8441                 root->fs_info->excluded_extents = &excluded_extents;
8442                 root->fs_info->fsck_extent_cache = &extent_cache;
8443                 root->fs_info->free_extent_hook = free_extent_hook;
8444                 root->fs_info->corrupt_blocks = &corrupt_blocks;
8445         }
8446
8447         bits_nr = 1024;
8448         bits = malloc(bits_nr * sizeof(struct block_info));
8449         if (!bits) {
8450                 perror("malloc");
8451                 exit(1);
8452         }
8453
8454         if (ctx.progress_enabled) {
8455                 ctx.tp = TASK_EXTENTS;
8456                 task_start(ctx.info);
8457         }
8458
8459 again:
8460         root1 = root->fs_info->tree_root;
8461         level = btrfs_header_level(root1->node);
8462         ret = add_root_item_to_list(&normal_trees, root1->root_key.objectid,
8463                                     root1->node->start, 0, level, 0,
8464                                     root1->nodesize, NULL);
8465         if (ret < 0)
8466                 goto out;
8467         root1 = root->fs_info->chunk_root;
8468         level = btrfs_header_level(root1->node);
8469         ret = add_root_item_to_list(&normal_trees, root1->root_key.objectid,
8470                                     root1->node->start, 0, level, 0,
8471                                     root1->nodesize, NULL);
8472         if (ret < 0)
8473                 goto out;
8474         btrfs_init_path(&path);
8475         key.offset = 0;
8476         key.objectid = 0;
8477         key.type = BTRFS_ROOT_ITEM_KEY;
8478         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
8479                                         &key, &path, 0, 0);
8480         if (ret < 0)
8481                 goto out;
8482         while(1) {
8483                 leaf = path.nodes[0];
8484                 slot = path.slots[0];
8485                 if (slot >= btrfs_header_nritems(path.nodes[0])) {
8486                         ret = btrfs_next_leaf(root, &path);
8487                         if (ret != 0)
8488                                 break;
8489                         leaf = path.nodes[0];
8490                         slot = path.slots[0];
8491                 }
8492                 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
8493                 if (found_key.type == BTRFS_ROOT_ITEM_KEY) {
8494                         unsigned long offset;
8495                         u64 last_snapshot;
8496
8497                         offset = btrfs_item_ptr_offset(leaf, path.slots[0]);
8498                         read_extent_buffer(leaf, &ri, offset, sizeof(ri));
8499                         last_snapshot = btrfs_root_last_snapshot(&ri);
8500                         if (btrfs_disk_key_objectid(&ri.drop_progress) == 0) {
8501                                 level = btrfs_root_level(&ri);
8502                                 level_size = root->nodesize;
8503                                 ret = add_root_item_to_list(&normal_trees,
8504                                                 found_key.objectid,
8505                                                 btrfs_root_bytenr(&ri),
8506                                                 last_snapshot, level,
8507                                                 0, level_size, NULL);
8508                                 if (ret < 0)
8509                                         goto out;
8510                         } else {
8511                                 level = btrfs_root_level(&ri);
8512                                 level_size = root->nodesize;
8513                                 objectid = found_key.objectid;
8514                                 btrfs_disk_key_to_cpu(&found_key,
8515                                                       &ri.drop_progress);
8516                                 ret = add_root_item_to_list(&dropping_trees,
8517                                                 objectid,
8518                                                 btrfs_root_bytenr(&ri),
8519                                                 last_snapshot, level,
8520                                                 ri.drop_level,
8521                                                 level_size, &found_key);
8522                                 if (ret < 0)
8523                                         goto out;
8524                         }
8525                 }
8526                 path.slots[0]++;
8527         }
8528         btrfs_release_path(&path);
8529
8530         /*
8531          * check_block can return -EAGAIN if it fixes something, please keep
8532          * this in mind when dealing with return values from these functions, if
8533          * we get -EAGAIN we want to fall through and restart the loop.
8534          */
8535         ret = deal_root_from_list(&normal_trees, root, bits, bits_nr, &pending,
8536                                   &seen, &reada, &nodes, &extent_cache,
8537                                   &chunk_cache, &dev_cache, &block_group_cache,
8538                                   &dev_extent_cache);
8539         if (ret < 0) {
8540                 if (ret == -EAGAIN)
8541                         goto loop;
8542                 goto out;
8543         }
8544         ret = deal_root_from_list(&dropping_trees, root, bits, bits_nr,
8545                                   &pending, &seen, &reada, &nodes,
8546                                   &extent_cache, &chunk_cache, &dev_cache,
8547                                   &block_group_cache, &dev_extent_cache);
8548         if (ret < 0) {
8549                 if (ret == -EAGAIN)
8550                         goto loop;
8551                 goto out;
8552         }
8553
8554         ret = check_chunks(&chunk_cache, &block_group_cache,
8555                            &dev_extent_cache, NULL, NULL, NULL, 0);
8556         if (ret) {
8557                 if (ret == -EAGAIN)
8558                         goto loop;
8559                 err = ret;
8560         }
8561
8562         ret = check_extent_refs(root, &extent_cache);
8563         if (ret < 0) {
8564                 if (ret == -EAGAIN)
8565                         goto loop;
8566                 goto out;
8567         }
8568
8569         ret = check_devices(&dev_cache, &dev_extent_cache);
8570         if (ret && err)
8571                 ret = err;
8572
8573 out:
8574         task_stop(ctx.info);
8575         if (repair) {
8576                 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
8577                 extent_io_tree_cleanup(&excluded_extents);
8578                 root->fs_info->fsck_extent_cache = NULL;
8579                 root->fs_info->free_extent_hook = NULL;
8580                 root->fs_info->corrupt_blocks = NULL;
8581                 root->fs_info->excluded_extents = NULL;
8582         }
8583         free(bits);
8584         free_chunk_cache_tree(&chunk_cache);
8585         free_device_cache_tree(&dev_cache);
8586         free_block_group_tree(&block_group_cache);
8587         free_device_extent_tree(&dev_extent_cache);
8588         free_extent_cache_tree(&seen);
8589         free_extent_cache_tree(&pending);
8590         free_extent_cache_tree(&reada);
8591         free_extent_cache_tree(&nodes);
8592         return ret;
8593 loop:
8594         free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
8595         free_extent_cache_tree(&seen);
8596         free_extent_cache_tree(&pending);
8597         free_extent_cache_tree(&reada);
8598         free_extent_cache_tree(&nodes);
8599         free_chunk_cache_tree(&chunk_cache);
8600         free_block_group_tree(&block_group_cache);
8601         free_device_cache_tree(&dev_cache);
8602         free_device_extent_tree(&dev_extent_cache);
8603         free_extent_record_cache(root->fs_info, &extent_cache);
8604         free_root_item_list(&normal_trees);
8605         free_root_item_list(&dropping_trees);
8606         extent_io_tree_cleanup(&excluded_extents);
8607         goto again;
8608 }
8609
8610 /*
8611  * Check backrefs of a tree block given by @bytenr or @eb.
8612  *
8613  * @root:       the root containing the @bytenr or @eb
8614  * @eb:         tree block extent buffer, can be NULL
8615  * @bytenr:     bytenr of the tree block to search
8616  * @level:      tree level of the tree block
8617  * @owner:      owner of the tree block
8618  *
8619  * Return >0 for any error found and output error message
8620  * Return 0 for no error found
8621  */
8622 static int check_tree_block_ref(struct btrfs_root *root,
8623                                 struct extent_buffer *eb, u64 bytenr,
8624                                 int level, u64 owner)
8625 {
8626         struct btrfs_key key;
8627         struct btrfs_root *extent_root = root->fs_info->extent_root;
8628         struct btrfs_path path;
8629         struct btrfs_extent_item *ei;
8630         struct btrfs_extent_inline_ref *iref;
8631         struct extent_buffer *leaf;
8632         unsigned long end;
8633         unsigned long ptr;
8634         int slot;
8635         int skinny_level;
8636         int type;
8637         u32 nodesize = root->nodesize;
8638         u32 item_size;
8639         u64 offset;
8640         int found_ref = 0;
8641         int err = 0;
8642         int ret;
8643
8644         btrfs_init_path(&path);
8645         key.objectid = bytenr;
8646         if (btrfs_fs_incompat(root->fs_info,
8647                               BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA))
8648                 key.type = BTRFS_METADATA_ITEM_KEY;
8649         else
8650                 key.type = BTRFS_EXTENT_ITEM_KEY;
8651         key.offset = (u64)-1;
8652
8653         /* Search for the backref in extent tree */
8654         ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
8655         if (ret < 0) {
8656                 err |= BACKREF_MISSING;
8657                 goto out;
8658         }
8659         ret = btrfs_previous_extent_item(extent_root, &path, bytenr);
8660         if (ret) {
8661                 err |= BACKREF_MISSING;
8662                 goto out;
8663         }
8664
8665         leaf = path.nodes[0];
8666         slot = path.slots[0];
8667         btrfs_item_key_to_cpu(leaf, &key, slot);
8668
8669         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
8670
8671         if (key.type == BTRFS_METADATA_ITEM_KEY) {
8672                 skinny_level = (int)key.offset;
8673                 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
8674         } else {
8675                 struct btrfs_tree_block_info *info;
8676
8677                 info = (struct btrfs_tree_block_info *)(ei + 1);
8678                 skinny_level = btrfs_tree_block_level(leaf, info);
8679                 iref = (struct btrfs_extent_inline_ref *)(info + 1);
8680         }
8681
8682         if (eb) {
8683                 u64 header_gen;
8684                 u64 extent_gen;
8685
8686                 if (!(btrfs_extent_flags(leaf, ei) &
8687                       BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
8688                         error(
8689                 "extent[%llu %u] backref type mismatch, missing bit: %llx",
8690                                 key.objectid, nodesize,
8691                                 BTRFS_EXTENT_FLAG_TREE_BLOCK);
8692                         err = BACKREF_MISMATCH;
8693                 }
8694                 header_gen = btrfs_header_generation(eb);
8695                 extent_gen = btrfs_extent_generation(leaf, ei);
8696                 if (header_gen != extent_gen) {
8697                         error(
8698         "extent[%llu %u] backref generation mismatch, wanted: %llu, have: %llu",
8699                                 key.objectid, nodesize, header_gen,
8700                                 extent_gen);
8701                         err = BACKREF_MISMATCH;
8702                 }
8703                 if (level != skinny_level) {
8704                         error(
8705                         "extent[%llu %u] level mismatch, wanted: %u, have: %u",
8706                                 key.objectid, nodesize, level, skinny_level);
8707                         err = BACKREF_MISMATCH;
8708                 }
8709                 if (!is_fstree(owner) && btrfs_extent_refs(leaf, ei) != 1) {
8710                         error(
8711                         "extent[%llu %u] is referred by other roots than %llu",
8712                                 key.objectid, nodesize, root->objectid);
8713                         err = BACKREF_MISMATCH;
8714                 }
8715         }
8716
8717         /*
8718          * Iterate the extent/metadata item to find the exact backref
8719          */
8720         item_size = btrfs_item_size_nr(leaf, slot);
8721         ptr = (unsigned long)iref;
8722         end = (unsigned long)ei + item_size;
8723         while (ptr < end) {
8724                 iref = (struct btrfs_extent_inline_ref *)ptr;
8725                 type = btrfs_extent_inline_ref_type(leaf, iref);
8726                 offset = btrfs_extent_inline_ref_offset(leaf, iref);
8727
8728                 if (type == BTRFS_TREE_BLOCK_REF_KEY &&
8729                         (offset == root->objectid || offset == owner)) {
8730                         found_ref = 1;
8731                 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
8732                         /* Check if the backref points to valid referencer */
8733                         found_ref = !check_tree_block_ref(root, NULL, offset,
8734                                                           level + 1, owner);
8735                 }
8736
8737                 if (found_ref)
8738                         break;
8739                 ptr += btrfs_extent_inline_ref_size(type);
8740         }
8741
8742         /*
8743          * Inlined extent item doesn't have what we need, check
8744          * TREE_BLOCK_REF_KEY
8745          */
8746         if (!found_ref) {
8747                 btrfs_release_path(&path);
8748                 key.objectid = bytenr;
8749                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
8750                 key.offset = root->objectid;
8751
8752                 ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
8753                 if (!ret)
8754                         found_ref = 1;
8755         }
8756         if (!found_ref)
8757                 err |= BACKREF_MISSING;
8758 out:
8759         btrfs_release_path(&path);
8760         if (eb && (err & BACKREF_MISSING))
8761                 error("extent[%llu %u] backref lost (owner: %llu, level: %u)",
8762                         bytenr, nodesize, owner, level);
8763         return err;
8764 }
8765
8766 /*
8767  * Check EXTENT_DATA item, mainly for its dbackref in extent tree
8768  *
8769  * Return >0 any error found and output error message
8770  * Return 0 for no error found
8771  */
8772 static int check_extent_data_item(struct btrfs_root *root,
8773                                   struct extent_buffer *eb, int slot)
8774 {
8775         struct btrfs_file_extent_item *fi;
8776         struct btrfs_path path;
8777         struct btrfs_root *extent_root = root->fs_info->extent_root;
8778         struct btrfs_key fi_key;
8779         struct btrfs_key dbref_key;
8780         struct extent_buffer *leaf;
8781         struct btrfs_extent_item *ei;
8782         struct btrfs_extent_inline_ref *iref;
8783         struct btrfs_extent_data_ref *dref;
8784         u64 owner;
8785         u64 file_extent_gen;
8786         u64 disk_bytenr;
8787         u64 disk_num_bytes;
8788         u64 extent_num_bytes;
8789         u64 extent_flags;
8790         u64 extent_gen;
8791         u32 item_size;
8792         unsigned long end;
8793         unsigned long ptr;
8794         int type;
8795         u64 ref_root;
8796         int found_dbackref = 0;
8797         int err = 0;
8798         int ret;
8799
8800         btrfs_item_key_to_cpu(eb, &fi_key, slot);
8801         fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
8802         file_extent_gen = btrfs_file_extent_generation(eb, fi);
8803
8804         /* Nothing to check for hole and inline data extents */
8805         if (btrfs_file_extent_type(eb, fi) == BTRFS_FILE_EXTENT_INLINE ||
8806             btrfs_file_extent_disk_bytenr(eb, fi) == 0)
8807                 return 0;
8808
8809         disk_bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
8810         disk_num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
8811         extent_num_bytes = btrfs_file_extent_num_bytes(eb, fi);
8812
8813         /* Check unaligned disk_num_bytes and num_bytes */
8814         if (!IS_ALIGNED(disk_num_bytes, root->sectorsize)) {
8815                 error(
8816 "file extent [%llu, %llu] has unaligned disk num bytes: %llu, should be aligned to %u",
8817                         fi_key.objectid, fi_key.offset, disk_num_bytes,
8818                         root->sectorsize);
8819                 err |= BYTES_UNALIGNED;
8820         } else {
8821                 data_bytes_allocated += disk_num_bytes;
8822         }
8823         if (!IS_ALIGNED(extent_num_bytes, root->sectorsize)) {
8824                 error(
8825 "file extent [%llu, %llu] has unaligned num bytes: %llu, should be aligned to %u",
8826                         fi_key.objectid, fi_key.offset, extent_num_bytes,
8827                         root->sectorsize);
8828                 err |= BYTES_UNALIGNED;
8829         } else {
8830                 data_bytes_referenced += extent_num_bytes;
8831         }
8832         owner = btrfs_header_owner(eb);
8833
8834         /* Check the extent item of the file extent in extent tree */
8835         btrfs_init_path(&path);
8836         dbref_key.objectid = btrfs_file_extent_disk_bytenr(eb, fi);
8837         dbref_key.type = BTRFS_EXTENT_ITEM_KEY;
8838         dbref_key.offset = btrfs_file_extent_disk_num_bytes(eb, fi);
8839
8840         ret = btrfs_search_slot(NULL, extent_root, &dbref_key, &path, 0, 0);
8841         if (ret) {
8842                 err |= BACKREF_MISSING;
8843                 goto error;
8844         }
8845
8846         leaf = path.nodes[0];
8847         slot = path.slots[0];
8848         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
8849
8850         extent_flags = btrfs_extent_flags(leaf, ei);
8851         extent_gen = btrfs_extent_generation(leaf, ei);
8852
8853         if (!(extent_flags & BTRFS_EXTENT_FLAG_DATA)) {
8854                 error(
8855                     "extent[%llu %llu] backref type mismatch, wanted bit: %llx",
8856                     disk_bytenr, disk_num_bytes,
8857                     BTRFS_EXTENT_FLAG_DATA);
8858                 err |= BACKREF_MISMATCH;
8859         }
8860
8861         if (file_extent_gen < extent_gen) {
8862                 error(
8863 "extent[%llu %llu] backref generation mismatch, wanted: <=%llu, have: %llu",
8864                         disk_bytenr, disk_num_bytes, file_extent_gen,
8865                         extent_gen);
8866                 err |= BACKREF_MISMATCH;
8867         }
8868
8869         /* Check data backref inside that extent item */
8870         item_size = btrfs_item_size_nr(leaf, path.slots[0]);
8871         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
8872         ptr = (unsigned long)iref;
8873         end = (unsigned long)ei + item_size;
8874         while (ptr < end) {
8875                 iref = (struct btrfs_extent_inline_ref *)ptr;
8876                 type = btrfs_extent_inline_ref_type(leaf, iref);
8877                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
8878
8879                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
8880                         ref_root = btrfs_extent_data_ref_root(leaf, dref);
8881                         if (ref_root == owner || ref_root == root->objectid)
8882                                 found_dbackref = 1;
8883                 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
8884                         found_dbackref = !check_tree_block_ref(root, NULL,
8885                                 btrfs_extent_inline_ref_offset(leaf, iref),
8886                                 0, owner);
8887                 }
8888
8889                 if (found_dbackref)
8890                         break;
8891                 ptr += btrfs_extent_inline_ref_size(type);
8892         }
8893
8894         /* Didn't found inlined data backref, try EXTENT_DATA_REF_KEY */
8895         if (!found_dbackref) {
8896                 btrfs_release_path(&path);
8897
8898                 btrfs_init_path(&path);
8899                 dbref_key.objectid = btrfs_file_extent_disk_bytenr(eb, fi);
8900                 dbref_key.type = BTRFS_EXTENT_DATA_REF_KEY;
8901                 dbref_key.offset = hash_extent_data_ref(root->objectid,
8902                                 fi_key.objectid, fi_key.offset);
8903
8904                 ret = btrfs_search_slot(NULL, root->fs_info->extent_root,
8905                                         &dbref_key, &path, 0, 0);
8906                 if (!ret)
8907                         found_dbackref = 1;
8908         }
8909
8910         if (!found_dbackref)
8911                 err |= BACKREF_MISSING;
8912 error:
8913         btrfs_release_path(&path);
8914         if (err & BACKREF_MISSING) {
8915                 error("data extent[%llu %llu] backref lost",
8916                       disk_bytenr, disk_num_bytes);
8917         }
8918         return err;
8919 }
8920
8921 /*
8922  * Get real tree block level for the case like shared block
8923  * Return >= 0 as tree level
8924  * Return <0 for error
8925  */
8926 static int query_tree_block_level(struct btrfs_fs_info *fs_info, u64 bytenr)
8927 {
8928         struct extent_buffer *eb;
8929         struct btrfs_path path;
8930         struct btrfs_key key;
8931         struct btrfs_extent_item *ei;
8932         u64 flags;
8933         u64 transid;
8934         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
8935         u8 backref_level;
8936         u8 header_level;
8937         int ret;
8938
8939         /* Search extent tree for extent generation and level */
8940         key.objectid = bytenr;
8941         key.type = BTRFS_METADATA_ITEM_KEY;
8942         key.offset = (u64)-1;
8943
8944         btrfs_init_path(&path);
8945         ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, &path, 0, 0);
8946         if (ret < 0)
8947                 goto release_out;
8948         ret = btrfs_previous_extent_item(fs_info->extent_root, &path, bytenr);
8949         if (ret < 0)
8950                 goto release_out;
8951         if (ret > 0) {
8952                 ret = -ENOENT;
8953                 goto release_out;
8954         }
8955
8956         btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
8957         ei = btrfs_item_ptr(path.nodes[0], path.slots[0],
8958                             struct btrfs_extent_item);
8959         flags = btrfs_extent_flags(path.nodes[0], ei);
8960         if (!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
8961                 ret = -ENOENT;
8962                 goto release_out;
8963         }
8964
8965         /* Get transid for later read_tree_block() check */
8966         transid = btrfs_extent_generation(path.nodes[0], ei);
8967
8968         /* Get backref level as one source */
8969         if (key.type == BTRFS_METADATA_ITEM_KEY) {
8970                 backref_level = key.offset;
8971         } else {
8972                 struct btrfs_tree_block_info *info;
8973
8974                 info = (struct btrfs_tree_block_info *)(ei + 1);
8975                 backref_level = btrfs_tree_block_level(path.nodes[0], info);
8976         }
8977         btrfs_release_path(&path);
8978
8979         /* Get level from tree block as an alternative source */
8980         eb = read_tree_block_fs_info(fs_info, bytenr, nodesize, transid);
8981         if (!extent_buffer_uptodate(eb)) {
8982                 free_extent_buffer(eb);
8983                 return -EIO;
8984         }
8985         header_level = btrfs_header_level(eb);
8986         free_extent_buffer(eb);
8987
8988         if (header_level != backref_level)
8989                 return -EIO;
8990         return header_level;
8991
8992 release_out:
8993         btrfs_release_path(&path);
8994         return ret;
8995 }
8996
8997 /*
8998  * Check if a tree block backref is valid (points to a valid tree block)
8999  * if level == -1, level will be resolved
9000  * Return >0 for any error found and print error message
9001  */
9002 static int check_tree_block_backref(struct btrfs_fs_info *fs_info, u64 root_id,
9003                                     u64 bytenr, int level)
9004 {
9005         struct btrfs_root *root;
9006         struct btrfs_key key;
9007         struct btrfs_path path;
9008         struct extent_buffer *eb;
9009         struct extent_buffer *node;
9010         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
9011         int err = 0;
9012         int ret;
9013
9014         /* Query level for level == -1 special case */
9015         if (level == -1)
9016                 level = query_tree_block_level(fs_info, bytenr);
9017         if (level < 0) {
9018                 err |= REFERENCER_MISSING;
9019                 goto out;
9020         }
9021
9022         key.objectid = root_id;
9023         key.type = BTRFS_ROOT_ITEM_KEY;
9024         key.offset = (u64)-1;
9025
9026         root = btrfs_read_fs_root(fs_info, &key);
9027         if (IS_ERR(root)) {
9028                 err |= REFERENCER_MISSING;
9029                 goto out;
9030         }
9031
9032         /* Read out the tree block to get item/node key */
9033         eb = read_tree_block(root, bytenr, root->nodesize, 0);
9034         if (!extent_buffer_uptodate(eb)) {
9035                 err |= REFERENCER_MISSING;
9036                 free_extent_buffer(eb);
9037                 goto out;
9038         }
9039
9040         /* Empty tree, no need to check key */
9041         if (!btrfs_header_nritems(eb) && !level) {
9042                 free_extent_buffer(eb);
9043                 goto out;
9044         }
9045
9046         if (level)
9047                 btrfs_node_key_to_cpu(eb, &key, 0);
9048         else
9049                 btrfs_item_key_to_cpu(eb, &key, 0);
9050
9051         free_extent_buffer(eb);
9052
9053         btrfs_init_path(&path);
9054         path.lowest_level = level;
9055         /* Search with the first key, to ensure we can reach it */
9056         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
9057         if (ret < 0) {
9058                 err |= REFERENCER_MISSING;
9059                 goto release_out;
9060         }
9061
9062         node = path.nodes[level];
9063         if (btrfs_header_bytenr(node) != bytenr) {
9064                 error(
9065         "extent [%llu %d] referencer bytenr mismatch, wanted: %llu, have: %llu",
9066                         bytenr, nodesize, bytenr,
9067                         btrfs_header_bytenr(node));
9068                 err |= REFERENCER_MISMATCH;
9069         }
9070         if (btrfs_header_level(node) != level) {
9071                 error(
9072         "extent [%llu %d] referencer level mismatch, wanted: %d, have: %d",
9073                         bytenr, nodesize, level,
9074                         btrfs_header_level(node));
9075                 err |= REFERENCER_MISMATCH;
9076         }
9077
9078 release_out:
9079         btrfs_release_path(&path);
9080 out:
9081         if (err & REFERENCER_MISSING) {
9082                 if (level < 0)
9083                         error("extent [%llu %d] lost referencer (owner: %llu)",
9084                                 bytenr, nodesize, root_id);
9085                 else
9086                         error(
9087                 "extent [%llu %d] lost referencer (owner: %llu, level: %u)",
9088                                 bytenr, nodesize, root_id, level);
9089         }
9090
9091         return err;
9092 }
9093
9094 /*
9095  * Check referencer for shared block backref
9096  * If level == -1, this function will resolve the level.
9097  */
9098 static int check_shared_block_backref(struct btrfs_fs_info *fs_info,
9099                                      u64 parent, u64 bytenr, int level)
9100 {
9101         struct extent_buffer *eb;
9102         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
9103         u32 nr;
9104         int found_parent = 0;
9105         int i;
9106
9107         eb = read_tree_block_fs_info(fs_info, parent, nodesize, 0);
9108         if (!extent_buffer_uptodate(eb))
9109                 goto out;
9110
9111         if (level == -1)
9112                 level = query_tree_block_level(fs_info, bytenr);
9113         if (level < 0)
9114                 goto out;
9115
9116         if (level + 1 != btrfs_header_level(eb))
9117                 goto out;
9118
9119         nr = btrfs_header_nritems(eb);
9120         for (i = 0; i < nr; i++) {
9121                 if (bytenr == btrfs_node_blockptr(eb, i)) {
9122                         found_parent = 1;
9123                         break;
9124                 }
9125         }
9126 out:
9127         free_extent_buffer(eb);
9128         if (!found_parent) {
9129                 error(
9130         "shared extent[%llu %u] lost its parent (parent: %llu, level: %u)",
9131                         bytenr, nodesize, parent, level);
9132                 return REFERENCER_MISSING;
9133         }
9134         return 0;
9135 }
9136
9137 /*
9138  * Check referencer for normal (inlined) data ref
9139  * If len == 0, it will be resolved by searching in extent tree
9140  */
9141 static int check_extent_data_backref(struct btrfs_fs_info *fs_info,
9142                                      u64 root_id, u64 objectid, u64 offset,
9143                                      u64 bytenr, u64 len, u32 count)
9144 {
9145         struct btrfs_root *root;
9146         struct btrfs_root *extent_root = fs_info->extent_root;
9147         struct btrfs_key key;
9148         struct btrfs_path path;
9149         struct extent_buffer *leaf;
9150         struct btrfs_file_extent_item *fi;
9151         u32 found_count = 0;
9152         int slot;
9153         int ret = 0;
9154
9155         if (!len) {
9156                 key.objectid = bytenr;
9157                 key.type = BTRFS_EXTENT_ITEM_KEY;
9158                 key.offset = (u64)-1;
9159
9160                 btrfs_init_path(&path);
9161                 ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
9162                 if (ret < 0)
9163                         goto out;
9164                 ret = btrfs_previous_extent_item(extent_root, &path, bytenr);
9165                 if (ret)
9166                         goto out;
9167                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
9168                 if (key.objectid != bytenr ||
9169                     key.type != BTRFS_EXTENT_ITEM_KEY)
9170                         goto out;
9171                 len = key.offset;
9172                 btrfs_release_path(&path);
9173         }
9174         key.objectid = root_id;
9175         key.type = BTRFS_ROOT_ITEM_KEY;
9176         key.offset = (u64)-1;
9177         btrfs_init_path(&path);
9178
9179         root = btrfs_read_fs_root(fs_info, &key);
9180         if (IS_ERR(root))
9181                 goto out;
9182
9183         key.objectid = objectid;
9184         key.type = BTRFS_EXTENT_DATA_KEY;
9185         /*
9186          * It can be nasty as data backref offset is
9187          * file offset - file extent offset, which is smaller or
9188          * equal to original backref offset.  The only special case is
9189          * overflow.  So we need to special check and do further search.
9190          */
9191         key.offset = offset & (1ULL << 63) ? 0 : offset;
9192
9193         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
9194         if (ret < 0)
9195                 goto out;
9196
9197         /*
9198          * Search afterwards to get correct one
9199          * NOTE: As we must do a comprehensive check on the data backref to
9200          * make sure the dref count also matches, we must iterate all file
9201          * extents for that inode.
9202          */
9203         while (1) {
9204                 leaf = path.nodes[0];
9205                 slot = path.slots[0];
9206
9207                 btrfs_item_key_to_cpu(leaf, &key, slot);
9208                 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
9209                         break;
9210                 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
9211                 /*
9212                  * Except normal disk bytenr and disk num bytes, we still
9213                  * need to do extra check on dbackref offset as
9214                  * dbackref offset = file_offset - file_extent_offset
9215                  */
9216                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == bytenr &&
9217                     btrfs_file_extent_disk_num_bytes(leaf, fi) == len &&
9218                     (u64)(key.offset - btrfs_file_extent_offset(leaf, fi)) ==
9219                     offset)
9220                         found_count++;
9221
9222                 ret = btrfs_next_item(root, &path);
9223                 if (ret)
9224                         break;
9225         }
9226 out:
9227         btrfs_release_path(&path);
9228         if (found_count != count) {
9229                 error(
9230 "extent[%llu, %llu] referencer count mismatch (root: %llu, owner: %llu, offset: %llu) wanted: %u, have: %u",
9231                         bytenr, len, root_id, objectid, offset, count, found_count);
9232                 return REFERENCER_MISSING;
9233         }
9234         return 0;
9235 }
9236
9237 /*
9238  * Check if the referencer of a shared data backref exists
9239  */
9240 static int check_shared_data_backref(struct btrfs_fs_info *fs_info,
9241                                      u64 parent, u64 bytenr)
9242 {
9243         struct extent_buffer *eb;
9244         struct btrfs_key key;
9245         struct btrfs_file_extent_item *fi;
9246         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
9247         u32 nr;
9248         int found_parent = 0;
9249         int i;
9250
9251         eb = read_tree_block_fs_info(fs_info, parent, nodesize, 0);
9252         if (!extent_buffer_uptodate(eb))
9253                 goto out;
9254
9255         nr = btrfs_header_nritems(eb);
9256         for (i = 0; i < nr; i++) {
9257                 btrfs_item_key_to_cpu(eb, &key, i);
9258                 if (key.type != BTRFS_EXTENT_DATA_KEY)
9259                         continue;
9260
9261                 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
9262                 if (btrfs_file_extent_type(eb, fi) == BTRFS_FILE_EXTENT_INLINE)
9263                         continue;
9264
9265                 if (btrfs_file_extent_disk_bytenr(eb, fi) == bytenr) {
9266                         found_parent = 1;
9267                         break;
9268                 }
9269         }
9270
9271 out:
9272         free_extent_buffer(eb);
9273         if (!found_parent) {
9274                 error("shared extent %llu referencer lost (parent: %llu)",
9275                         bytenr, parent);
9276                 return REFERENCER_MISSING;
9277         }
9278         return 0;
9279 }
9280
9281 /*
9282  * This function will check a given extent item, including its backref and
9283  * itself (like crossing stripe boundary and type)
9284  *
9285  * Since we don't use extent_record anymore, introduce new error bit
9286  */
9287 static int check_extent_item(struct btrfs_fs_info *fs_info,
9288                              struct extent_buffer *eb, int slot)
9289 {
9290         struct btrfs_extent_item *ei;
9291         struct btrfs_extent_inline_ref *iref;
9292         struct btrfs_extent_data_ref *dref;
9293         unsigned long end;
9294         unsigned long ptr;
9295         int type;
9296         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
9297         u32 item_size = btrfs_item_size_nr(eb, slot);
9298         u64 flags;
9299         u64 offset;
9300         int metadata = 0;
9301         int level;
9302         struct btrfs_key key;
9303         int ret;
9304         int err = 0;
9305
9306         btrfs_item_key_to_cpu(eb, &key, slot);
9307         if (key.type == BTRFS_EXTENT_ITEM_KEY)
9308                 bytes_used += key.offset;
9309         else
9310                 bytes_used += nodesize;
9311
9312         if (item_size < sizeof(*ei)) {
9313                 /*
9314                  * COMPAT_EXTENT_TREE_V0 case, but it's already a super
9315                  * old thing when on disk format is still un-determined.
9316                  * No need to care about it anymore
9317                  */
9318                 error("unsupported COMPAT_EXTENT_TREE_V0 detected");
9319                 return -ENOTTY;
9320         }
9321
9322         ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
9323         flags = btrfs_extent_flags(eb, ei);
9324
9325         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
9326                 metadata = 1;
9327         if (metadata && check_crossing_stripes(global_info, key.objectid,
9328                                                eb->len)) {
9329                 error("bad metadata [%llu, %llu) crossing stripe boundary",
9330                       key.objectid, key.objectid + nodesize);
9331                 err |= CROSSING_STRIPE_BOUNDARY;
9332         }
9333
9334         ptr = (unsigned long)(ei + 1);
9335
9336         if (metadata && key.type == BTRFS_EXTENT_ITEM_KEY) {
9337                 /* Old EXTENT_ITEM metadata */
9338                 struct btrfs_tree_block_info *info;
9339
9340                 info = (struct btrfs_tree_block_info *)ptr;
9341                 level = btrfs_tree_block_level(eb, info);
9342                 ptr += sizeof(struct btrfs_tree_block_info);
9343         } else {
9344                 /* New METADATA_ITEM */
9345                 level = key.offset;
9346         }
9347         end = (unsigned long)ei + item_size;
9348
9349         if (ptr >= end) {
9350                 err |= ITEM_SIZE_MISMATCH;
9351                 goto out;
9352         }
9353
9354         /* Now check every backref in this extent item */
9355 next:
9356         iref = (struct btrfs_extent_inline_ref *)ptr;
9357         type = btrfs_extent_inline_ref_type(eb, iref);
9358         offset = btrfs_extent_inline_ref_offset(eb, iref);
9359         switch (type) {
9360         case BTRFS_TREE_BLOCK_REF_KEY:
9361                 ret = check_tree_block_backref(fs_info, offset, key.objectid,
9362                                                level);
9363                 err |= ret;
9364                 break;
9365         case BTRFS_SHARED_BLOCK_REF_KEY:
9366                 ret = check_shared_block_backref(fs_info, offset, key.objectid,
9367                                                  level);
9368                 err |= ret;
9369                 break;
9370         case BTRFS_EXTENT_DATA_REF_KEY:
9371                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
9372                 ret = check_extent_data_backref(fs_info,
9373                                 btrfs_extent_data_ref_root(eb, dref),
9374                                 btrfs_extent_data_ref_objectid(eb, dref),
9375                                 btrfs_extent_data_ref_offset(eb, dref),
9376                                 key.objectid, key.offset,
9377                                 btrfs_extent_data_ref_count(eb, dref));
9378                 err |= ret;
9379                 break;
9380         case BTRFS_SHARED_DATA_REF_KEY:
9381                 ret = check_shared_data_backref(fs_info, offset, key.objectid);
9382                 err |= ret;
9383                 break;
9384         default:
9385                 error("extent[%llu %d %llu] has unknown ref type: %d",
9386                         key.objectid, key.type, key.offset, type);
9387                 err |= UNKNOWN_TYPE;
9388                 goto out;
9389         }
9390
9391         ptr += btrfs_extent_inline_ref_size(type);
9392         if (ptr < end)
9393                 goto next;
9394
9395 out:
9396         return err;
9397 }
9398
9399 /*
9400  * Check if a dev extent item is referred correctly by its chunk
9401  */
9402 static int check_dev_extent_item(struct btrfs_fs_info *fs_info,
9403                                  struct extent_buffer *eb, int slot)
9404 {
9405         struct btrfs_root *chunk_root = fs_info->chunk_root;
9406         struct btrfs_dev_extent *ptr;
9407         struct btrfs_path path;
9408         struct btrfs_key chunk_key;
9409         struct btrfs_key devext_key;
9410         struct btrfs_chunk *chunk;
9411         struct extent_buffer *l;
9412         int num_stripes;
9413         u64 length;
9414         int i;
9415         int found_chunk = 0;
9416         int ret;
9417
9418         btrfs_item_key_to_cpu(eb, &devext_key, slot);
9419         ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_extent);
9420         length = btrfs_dev_extent_length(eb, ptr);
9421
9422         chunk_key.objectid = btrfs_dev_extent_chunk_objectid(eb, ptr);
9423         chunk_key.type = BTRFS_CHUNK_ITEM_KEY;
9424         chunk_key.offset = btrfs_dev_extent_chunk_offset(eb, ptr);
9425
9426         btrfs_init_path(&path);
9427         ret = btrfs_search_slot(NULL, chunk_root, &chunk_key, &path, 0, 0);
9428         if (ret)
9429                 goto out;
9430
9431         l = path.nodes[0];
9432         chunk = btrfs_item_ptr(l, path.slots[0], struct btrfs_chunk);
9433         if (btrfs_chunk_length(l, chunk) != length)
9434                 goto out;
9435
9436         num_stripes = btrfs_chunk_num_stripes(l, chunk);
9437         for (i = 0; i < num_stripes; i++) {
9438                 u64 devid = btrfs_stripe_devid_nr(l, chunk, i);
9439                 u64 offset = btrfs_stripe_offset_nr(l, chunk, i);
9440
9441                 if (devid == devext_key.objectid &&
9442                     offset == devext_key.offset) {
9443                         found_chunk = 1;
9444                         break;
9445                 }
9446         }
9447 out:
9448         btrfs_release_path(&path);
9449         if (!found_chunk) {
9450                 error(
9451                 "device extent[%llu, %llu, %llu] did not find the related chunk",
9452                         devext_key.objectid, devext_key.offset, length);
9453                 return REFERENCER_MISSING;
9454         }
9455         return 0;
9456 }
9457
9458 /*
9459  * Check if the used space is correct with the dev item
9460  */
9461 static int check_dev_item(struct btrfs_fs_info *fs_info,
9462                           struct extent_buffer *eb, int slot)
9463 {
9464         struct btrfs_root *dev_root = fs_info->dev_root;
9465         struct btrfs_dev_item *dev_item;
9466         struct btrfs_path path;
9467         struct btrfs_key key;
9468         struct btrfs_dev_extent *ptr;
9469         u64 dev_id;
9470         u64 used;
9471         u64 total = 0;
9472         int ret;
9473
9474         dev_item = btrfs_item_ptr(eb, slot, struct btrfs_dev_item);
9475         dev_id = btrfs_device_id(eb, dev_item);
9476         used = btrfs_device_bytes_used(eb, dev_item);
9477
9478         key.objectid = dev_id;
9479         key.type = BTRFS_DEV_EXTENT_KEY;
9480         key.offset = 0;
9481
9482         btrfs_init_path(&path);
9483         ret = btrfs_search_slot(NULL, dev_root, &key, &path, 0, 0);
9484         if (ret < 0) {
9485                 btrfs_item_key_to_cpu(eb, &key, slot);
9486                 error("cannot find any related dev extent for dev[%llu, %u, %llu]",
9487                         key.objectid, key.type, key.offset);
9488                 btrfs_release_path(&path);
9489                 return REFERENCER_MISSING;
9490         }
9491
9492         /* Iterate dev_extents to calculate the used space of a device */
9493         while (1) {
9494                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
9495
9496                 if (key.objectid > dev_id)
9497                         break;
9498                 if (key.type != BTRFS_DEV_EXTENT_KEY || key.objectid != dev_id)
9499                         goto next;
9500
9501                 ptr = btrfs_item_ptr(path.nodes[0], path.slots[0],
9502                                      struct btrfs_dev_extent);
9503                 total += btrfs_dev_extent_length(path.nodes[0], ptr);
9504 next:
9505                 ret = btrfs_next_item(dev_root, &path);
9506                 if (ret)
9507                         break;
9508         }
9509         btrfs_release_path(&path);
9510
9511         if (used != total) {
9512                 btrfs_item_key_to_cpu(eb, &key, slot);
9513                 error(
9514 "Dev extent's total-byte %llu is not equal to bytes-used %llu in dev[%llu, %u, %llu]",
9515                         total, used, BTRFS_ROOT_TREE_OBJECTID,
9516                         BTRFS_DEV_EXTENT_KEY, dev_id);
9517                 return ACCOUNTING_MISMATCH;
9518         }
9519         return 0;
9520 }
9521
9522 /*
9523  * Check a block group item with its referener (chunk) and its used space
9524  * with extent/metadata item
9525  */
9526 static int check_block_group_item(struct btrfs_fs_info *fs_info,
9527                                   struct extent_buffer *eb, int slot)
9528 {
9529         struct btrfs_root *extent_root = fs_info->extent_root;
9530         struct btrfs_root *chunk_root = fs_info->chunk_root;
9531         struct btrfs_block_group_item *bi;
9532         struct btrfs_block_group_item bg_item;
9533         struct btrfs_path path;
9534         struct btrfs_key bg_key;
9535         struct btrfs_key chunk_key;
9536         struct btrfs_key extent_key;
9537         struct btrfs_chunk *chunk;
9538         struct extent_buffer *leaf;
9539         struct btrfs_extent_item *ei;
9540         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
9541         u64 flags;
9542         u64 bg_flags;
9543         u64 used;
9544         u64 total = 0;
9545         int ret;
9546         int err = 0;
9547
9548         btrfs_item_key_to_cpu(eb, &bg_key, slot);
9549         bi = btrfs_item_ptr(eb, slot, struct btrfs_block_group_item);
9550         read_extent_buffer(eb, &bg_item, (unsigned long)bi, sizeof(bg_item));
9551         used = btrfs_block_group_used(&bg_item);
9552         bg_flags = btrfs_block_group_flags(&bg_item);
9553
9554         chunk_key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
9555         chunk_key.type = BTRFS_CHUNK_ITEM_KEY;
9556         chunk_key.offset = bg_key.objectid;
9557
9558         btrfs_init_path(&path);
9559         /* Search for the referencer chunk */
9560         ret = btrfs_search_slot(NULL, chunk_root, &chunk_key, &path, 0, 0);
9561         if (ret) {
9562                 error(
9563                 "block group[%llu %llu] did not find the related chunk item",
9564                         bg_key.objectid, bg_key.offset);
9565                 err |= REFERENCER_MISSING;
9566         } else {
9567                 chunk = btrfs_item_ptr(path.nodes[0], path.slots[0],
9568                                         struct btrfs_chunk);
9569                 if (btrfs_chunk_length(path.nodes[0], chunk) !=
9570                                                 bg_key.offset) {
9571                         error(
9572         "block group[%llu %llu] related chunk item length does not match",
9573                                 bg_key.objectid, bg_key.offset);
9574                         err |= REFERENCER_MISMATCH;
9575                 }
9576         }
9577         btrfs_release_path(&path);
9578
9579         /* Search from the block group bytenr */
9580         extent_key.objectid = bg_key.objectid;
9581         extent_key.type = 0;
9582         extent_key.offset = 0;
9583
9584         btrfs_init_path(&path);
9585         ret = btrfs_search_slot(NULL, extent_root, &extent_key, &path, 0, 0);
9586         if (ret < 0)
9587                 goto out;
9588
9589         /* Iterate extent tree to account used space */
9590         while (1) {
9591                 leaf = path.nodes[0];
9592                 btrfs_item_key_to_cpu(leaf, &extent_key, path.slots[0]);
9593                 if (extent_key.objectid >= bg_key.objectid + bg_key.offset)
9594                         break;
9595
9596                 if (extent_key.type != BTRFS_METADATA_ITEM_KEY &&
9597                     extent_key.type != BTRFS_EXTENT_ITEM_KEY)
9598                         goto next;
9599                 if (extent_key.objectid < bg_key.objectid)
9600                         goto next;
9601
9602                 if (extent_key.type == BTRFS_METADATA_ITEM_KEY)
9603                         total += nodesize;
9604                 else
9605                         total += extent_key.offset;
9606
9607                 ei = btrfs_item_ptr(leaf, path.slots[0],
9608                                     struct btrfs_extent_item);
9609                 flags = btrfs_extent_flags(leaf, ei);
9610                 if (flags & BTRFS_EXTENT_FLAG_DATA) {
9611                         if (!(bg_flags & BTRFS_BLOCK_GROUP_DATA)) {
9612                                 error(
9613                         "bad extent[%llu, %llu) type mismatch with chunk",
9614                                         extent_key.objectid,
9615                                         extent_key.objectid + extent_key.offset);
9616                                 err |= CHUNK_TYPE_MISMATCH;
9617                         }
9618                 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
9619                         if (!(bg_flags & (BTRFS_BLOCK_GROUP_SYSTEM |
9620                                     BTRFS_BLOCK_GROUP_METADATA))) {
9621                                 error(
9622                         "bad extent[%llu, %llu) type mismatch with chunk",
9623                                         extent_key.objectid,
9624                                         extent_key.objectid + nodesize);
9625                                 err |= CHUNK_TYPE_MISMATCH;
9626                         }
9627                 }
9628 next:
9629                 ret = btrfs_next_item(extent_root, &path);
9630                 if (ret)
9631                         break;
9632         }
9633
9634 out:
9635         btrfs_release_path(&path);
9636
9637         if (total != used) {
9638                 error(
9639                 "block group[%llu %llu] used %llu but extent items used %llu",
9640                         bg_key.objectid, bg_key.offset, used, total);
9641                 err |= ACCOUNTING_MISMATCH;
9642         }
9643         return err;
9644 }
9645
9646 /*
9647  * Check a chunk item.
9648  * Including checking all referred dev_extents and block group
9649  */
9650 static int check_chunk_item(struct btrfs_fs_info *fs_info,
9651                             struct extent_buffer *eb, int slot)
9652 {
9653         struct btrfs_root *extent_root = fs_info->extent_root;
9654         struct btrfs_root *dev_root = fs_info->dev_root;
9655         struct btrfs_path path;
9656         struct btrfs_key chunk_key;
9657         struct btrfs_key bg_key;
9658         struct btrfs_key devext_key;
9659         struct btrfs_chunk *chunk;
9660         struct extent_buffer *leaf;
9661         struct btrfs_block_group_item *bi;
9662         struct btrfs_block_group_item bg_item;
9663         struct btrfs_dev_extent *ptr;
9664         u32 sectorsize = btrfs_super_sectorsize(fs_info->super_copy);
9665         u64 length;
9666         u64 chunk_end;
9667         u64 type;
9668         u64 profile;
9669         int num_stripes;
9670         u64 offset;
9671         u64 objectid;
9672         int i;
9673         int ret;
9674         int err = 0;
9675
9676         btrfs_item_key_to_cpu(eb, &chunk_key, slot);
9677         chunk = btrfs_item_ptr(eb, slot, struct btrfs_chunk);
9678         length = btrfs_chunk_length(eb, chunk);
9679         chunk_end = chunk_key.offset + length;
9680         if (!IS_ALIGNED(length, sectorsize)) {
9681                 error("chunk[%llu %llu) not aligned to %u",
9682                         chunk_key.offset, chunk_end, sectorsize);
9683                 err |= BYTES_UNALIGNED;
9684                 goto out;
9685         }
9686
9687         type = btrfs_chunk_type(eb, chunk);
9688         profile = type & BTRFS_BLOCK_GROUP_PROFILE_MASK;
9689         if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
9690                 error("chunk[%llu %llu) has no chunk type",
9691                         chunk_key.offset, chunk_end);
9692                 err |= UNKNOWN_TYPE;
9693         }
9694         if (profile && (profile & (profile - 1))) {
9695                 error("chunk[%llu %llu) multiple profiles detected: %llx",
9696                         chunk_key.offset, chunk_end, profile);
9697                 err |= UNKNOWN_TYPE;
9698         }
9699
9700         bg_key.objectid = chunk_key.offset;
9701         bg_key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9702         bg_key.offset = length;
9703
9704         btrfs_init_path(&path);
9705         ret = btrfs_search_slot(NULL, extent_root, &bg_key, &path, 0, 0);
9706         if (ret) {
9707                 error(
9708                 "chunk[%llu %llu) did not find the related block group item",
9709                         chunk_key.offset, chunk_end);
9710                 err |= REFERENCER_MISSING;
9711         } else{
9712                 leaf = path.nodes[0];
9713                 bi = btrfs_item_ptr(leaf, path.slots[0],
9714                                     struct btrfs_block_group_item);
9715                 read_extent_buffer(leaf, &bg_item, (unsigned long)bi,
9716                                    sizeof(bg_item));
9717                 if (btrfs_block_group_flags(&bg_item) != type) {
9718                         error(
9719 "chunk[%llu %llu) related block group item flags mismatch, wanted: %llu, have: %llu",
9720                                 chunk_key.offset, chunk_end, type,
9721                                 btrfs_block_group_flags(&bg_item));
9722                         err |= REFERENCER_MISSING;
9723                 }
9724         }
9725
9726         num_stripes = btrfs_chunk_num_stripes(eb, chunk);
9727         for (i = 0; i < num_stripes; i++) {
9728                 btrfs_release_path(&path);
9729                 btrfs_init_path(&path);
9730                 devext_key.objectid = btrfs_stripe_devid_nr(eb, chunk, i);
9731                 devext_key.type = BTRFS_DEV_EXTENT_KEY;
9732                 devext_key.offset = btrfs_stripe_offset_nr(eb, chunk, i);
9733
9734                 ret = btrfs_search_slot(NULL, dev_root, &devext_key, &path,
9735                                         0, 0);
9736                 if (ret)
9737                         goto not_match_dev;
9738
9739                 leaf = path.nodes[0];
9740                 ptr = btrfs_item_ptr(leaf, path.slots[0],
9741                                      struct btrfs_dev_extent);
9742                 objectid = btrfs_dev_extent_chunk_objectid(leaf, ptr);
9743                 offset = btrfs_dev_extent_chunk_offset(leaf, ptr);
9744                 if (objectid != chunk_key.objectid ||
9745                     offset != chunk_key.offset ||
9746                     btrfs_dev_extent_length(leaf, ptr) != length)
9747                         goto not_match_dev;
9748                 continue;
9749 not_match_dev:
9750                 err |= BACKREF_MISSING;
9751                 error(
9752                 "chunk[%llu %llu) stripe %d did not find the related dev extent",
9753                         chunk_key.objectid, chunk_end, i);
9754                 continue;
9755         }
9756         btrfs_release_path(&path);
9757 out:
9758         return err;
9759 }
9760
9761 /*
9762  * Main entry function to check known items and update related accounting info
9763  */
9764 static int check_leaf_items(struct btrfs_root *root, struct extent_buffer *eb)
9765 {
9766         struct btrfs_fs_info *fs_info = root->fs_info;
9767         struct btrfs_key key;
9768         int slot = 0;
9769         int type;
9770         struct btrfs_extent_data_ref *dref;
9771         int ret;
9772         int err = 0;
9773
9774 next:
9775         btrfs_item_key_to_cpu(eb, &key, slot);
9776         type = key.type;
9777
9778         switch (type) {
9779         case BTRFS_EXTENT_DATA_KEY:
9780                 ret = check_extent_data_item(root, eb, slot);
9781                 err |= ret;
9782                 break;
9783         case BTRFS_BLOCK_GROUP_ITEM_KEY:
9784                 ret = check_block_group_item(fs_info, eb, slot);
9785                 err |= ret;
9786                 break;
9787         case BTRFS_DEV_ITEM_KEY:
9788                 ret = check_dev_item(fs_info, eb, slot);
9789                 err |= ret;
9790                 break;
9791         case BTRFS_CHUNK_ITEM_KEY:
9792                 ret = check_chunk_item(fs_info, eb, slot);
9793                 err |= ret;
9794                 break;
9795         case BTRFS_DEV_EXTENT_KEY:
9796                 ret = check_dev_extent_item(fs_info, eb, slot);
9797                 err |= ret;
9798                 break;
9799         case BTRFS_EXTENT_ITEM_KEY:
9800         case BTRFS_METADATA_ITEM_KEY:
9801                 ret = check_extent_item(fs_info, eb, slot);
9802                 err |= ret;
9803                 break;
9804         case BTRFS_EXTENT_CSUM_KEY:
9805                 total_csum_bytes += btrfs_item_size_nr(eb, slot);
9806                 break;
9807         case BTRFS_TREE_BLOCK_REF_KEY:
9808                 ret = check_tree_block_backref(fs_info, key.offset,
9809                                                key.objectid, -1);
9810                 err |= ret;
9811                 break;
9812         case BTRFS_EXTENT_DATA_REF_KEY:
9813                 dref = btrfs_item_ptr(eb, slot, struct btrfs_extent_data_ref);
9814                 ret = check_extent_data_backref(fs_info,
9815                                 btrfs_extent_data_ref_root(eb, dref),
9816                                 btrfs_extent_data_ref_objectid(eb, dref),
9817                                 btrfs_extent_data_ref_offset(eb, dref),
9818                                 key.objectid, 0,
9819                                 btrfs_extent_data_ref_count(eb, dref));
9820                 err |= ret;
9821                 break;
9822         case BTRFS_SHARED_BLOCK_REF_KEY:
9823                 ret = check_shared_block_backref(fs_info, key.offset,
9824                                                  key.objectid, -1);
9825                 err |= ret;
9826                 break;
9827         case BTRFS_SHARED_DATA_REF_KEY:
9828                 ret = check_shared_data_backref(fs_info, key.offset,
9829                                                 key.objectid);
9830                 err |= ret;
9831                 break;
9832         default:
9833                 break;
9834         }
9835
9836         if (++slot < btrfs_header_nritems(eb))
9837                 goto next;
9838
9839         return err;
9840 }
9841
9842 /*
9843  * Helper function for later fs/subvol tree check.  To determine if a tree
9844  * block should be checked.
9845  * This function will ensure only the direct referencer with lowest rootid to
9846  * check a fs/subvolume tree block.
9847  *
9848  * Backref check at extent tree would detect errors like missing subvolume
9849  * tree, so we can do aggressive check to reduce duplicated checks.
9850  */
9851 static int should_check(struct btrfs_root *root, struct extent_buffer *eb)
9852 {
9853         struct btrfs_root *extent_root = root->fs_info->extent_root;
9854         struct btrfs_key key;
9855         struct btrfs_path path;
9856         struct extent_buffer *leaf;
9857         int slot;
9858         struct btrfs_extent_item *ei;
9859         unsigned long ptr;
9860         unsigned long end;
9861         int type;
9862         u32 item_size;
9863         u64 offset;
9864         struct btrfs_extent_inline_ref *iref;
9865         int ret;
9866
9867         btrfs_init_path(&path);
9868         key.objectid = btrfs_header_bytenr(eb);
9869         key.type = BTRFS_METADATA_ITEM_KEY;
9870         key.offset = (u64)-1;
9871
9872         /*
9873          * Any failure in backref resolving means we can't determine
9874          * whom the tree block belongs to.
9875          * So in that case, we need to check that tree block
9876          */
9877         ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
9878         if (ret < 0)
9879                 goto need_check;
9880
9881         ret = btrfs_previous_extent_item(extent_root, &path,
9882                                          btrfs_header_bytenr(eb));
9883         if (ret)
9884                 goto need_check;
9885
9886         leaf = path.nodes[0];
9887         slot = path.slots[0];
9888         btrfs_item_key_to_cpu(leaf, &key, slot);
9889         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
9890
9891         if (key.type == BTRFS_METADATA_ITEM_KEY) {
9892                 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
9893         } else {
9894                 struct btrfs_tree_block_info *info;
9895
9896                 info = (struct btrfs_tree_block_info *)(ei + 1);
9897                 iref = (struct btrfs_extent_inline_ref *)(info + 1);
9898         }
9899
9900         item_size = btrfs_item_size_nr(leaf, slot);
9901         ptr = (unsigned long)iref;
9902         end = (unsigned long)ei + item_size;
9903         while (ptr < end) {
9904                 iref = (struct btrfs_extent_inline_ref *)ptr;
9905                 type = btrfs_extent_inline_ref_type(leaf, iref);
9906                 offset = btrfs_extent_inline_ref_offset(leaf, iref);
9907
9908                 /*
9909                  * We only check the tree block if current root is
9910                  * the lowest referencer of it.
9911                  */
9912                 if (type == BTRFS_TREE_BLOCK_REF_KEY &&
9913                     offset < root->objectid) {
9914                         btrfs_release_path(&path);
9915                         return 0;
9916                 }
9917
9918                 ptr += btrfs_extent_inline_ref_size(type);
9919         }
9920         /*
9921          * Normally we should also check keyed tree block ref, but that may be
9922          * very time consuming.  Inlined ref should already make us skip a lot
9923          * of refs now.  So skip search keyed tree block ref.
9924          */
9925
9926 need_check:
9927         btrfs_release_path(&path);
9928         return 1;
9929 }
9930
9931 /*
9932  * Traversal function for tree block. We will do:
9933  * 1) Skip shared fs/subvolume tree blocks
9934  * 2) Update related bytes accounting
9935  * 3) Pre-order traversal
9936  */
9937 static int traverse_tree_block(struct btrfs_root *root,
9938                                 struct extent_buffer *node)
9939 {
9940         struct extent_buffer *eb;
9941         struct btrfs_key key;
9942         struct btrfs_key drop_key;
9943         int level;
9944         u64 nr;
9945         int i;
9946         int err = 0;
9947         int ret;
9948
9949         /*
9950          * Skip shared fs/subvolume tree block, in that case they will
9951          * be checked by referencer with lowest rootid
9952          */
9953         if (is_fstree(root->objectid) && !should_check(root, node))
9954                 return 0;
9955
9956         /* Update bytes accounting */
9957         total_btree_bytes += node->len;
9958         if (fs_root_objectid(btrfs_header_owner(node)))
9959                 total_fs_tree_bytes += node->len;
9960         if (btrfs_header_owner(node) == BTRFS_EXTENT_TREE_OBJECTID)
9961                 total_extent_tree_bytes += node->len;
9962         if (!found_old_backref &&
9963             btrfs_header_owner(node) == BTRFS_TREE_RELOC_OBJECTID &&
9964             btrfs_header_backref_rev(node) == BTRFS_MIXED_BACKREF_REV &&
9965             !btrfs_header_flag(node, BTRFS_HEADER_FLAG_RELOC))
9966                 found_old_backref = 1;
9967
9968         /* pre-order tranversal, check itself first */
9969         level = btrfs_header_level(node);
9970         ret = check_tree_block_ref(root, node, btrfs_header_bytenr(node),
9971                                    btrfs_header_level(node),
9972                                    btrfs_header_owner(node));
9973         err |= ret;
9974         if (err)
9975                 error(
9976         "check %s failed root %llu bytenr %llu level %d, force continue check",
9977                         level ? "node":"leaf", root->objectid,
9978                         btrfs_header_bytenr(node), btrfs_header_level(node));
9979
9980         if (!level) {
9981                 btree_space_waste += btrfs_leaf_free_space(root, node);
9982                 ret = check_leaf_items(root, node);
9983                 err |= ret;
9984                 return err;
9985         }
9986
9987         nr = btrfs_header_nritems(node);
9988         btrfs_disk_key_to_cpu(&drop_key, &root->root_item.drop_progress);
9989         btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) - nr) *
9990                 sizeof(struct btrfs_key_ptr);
9991
9992         /* Then check all its children */
9993         for (i = 0; i < nr; i++) {
9994                 u64 blocknr = btrfs_node_blockptr(node, i);
9995
9996                 btrfs_node_key_to_cpu(node, &key, i);
9997                 if (level == root->root_item.drop_level &&
9998                     is_dropped_key(&key, &drop_key))
9999                         continue;
10000
10001                 /*
10002                  * As a btrfs tree has most 8 levels (0..7), so it's quite safe
10003                  * to call the function itself.
10004                  */
10005                 eb = read_tree_block(root, blocknr, root->nodesize, 0);
10006                 if (extent_buffer_uptodate(eb)) {
10007                         ret = traverse_tree_block(root, eb);
10008                         err |= ret;
10009                 }
10010                 free_extent_buffer(eb);
10011         }
10012
10013         return err;
10014 }
10015
10016 /*
10017  * Low memory usage version check_chunks_and_extents.
10018  */
10019 static int check_chunks_and_extents_v2(struct btrfs_root *root)
10020 {
10021         struct btrfs_path path;
10022         struct btrfs_key key;
10023         struct btrfs_root *root1;
10024         struct btrfs_root *cur_root;
10025         int err = 0;
10026         int ret;
10027
10028         root1 = root->fs_info->chunk_root;
10029         ret = traverse_tree_block(root1, root1->node);
10030         err |= ret;
10031
10032         root1 = root->fs_info->tree_root;
10033         ret = traverse_tree_block(root1, root1->node);
10034         err |= ret;
10035
10036         btrfs_init_path(&path);
10037         key.objectid = BTRFS_EXTENT_TREE_OBJECTID;
10038         key.offset = 0;
10039         key.type = BTRFS_ROOT_ITEM_KEY;
10040
10041         ret = btrfs_search_slot(NULL, root1, &key, &path, 0, 0);
10042         if (ret) {
10043                 error("cannot find extent treet in tree_root");
10044                 goto out;
10045         }
10046
10047         while (1) {
10048                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
10049                 if (key.type != BTRFS_ROOT_ITEM_KEY)
10050                         goto next;
10051                 key.offset = (u64)-1;
10052
10053                 cur_root = btrfs_read_fs_root(root->fs_info, &key);
10054                 if (IS_ERR(cur_root) || !cur_root) {
10055                         error("failed to read tree: %lld", key.objectid);
10056                         goto next;
10057                 }
10058
10059                 ret = traverse_tree_block(cur_root, cur_root->node);
10060                 err |= ret;
10061
10062 next:
10063                 ret = btrfs_next_item(root1, &path);
10064                 if (ret)
10065                         goto out;
10066         }
10067
10068 out:
10069         btrfs_release_path(&path);
10070         return err;
10071 }
10072
10073 static int btrfs_fsck_reinit_root(struct btrfs_trans_handle *trans,
10074                            struct btrfs_root *root, int overwrite)
10075 {
10076         struct extent_buffer *c;
10077         struct extent_buffer *old = root->node;
10078         int level;
10079         int ret;
10080         struct btrfs_disk_key disk_key = {0,0,0};
10081
10082         level = 0;
10083
10084         if (overwrite) {
10085                 c = old;
10086                 extent_buffer_get(c);
10087                 goto init;
10088         }
10089         c = btrfs_alloc_free_block(trans, root,
10090                                    root->nodesize,
10091                                    root->root_key.objectid,
10092                                    &disk_key, level, 0, 0);
10093         if (IS_ERR(c)) {
10094                 c = old;
10095                 extent_buffer_get(c);
10096                 overwrite = 1;
10097         }
10098 init:
10099         memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
10100         btrfs_set_header_level(c, level);
10101         btrfs_set_header_bytenr(c, c->start);
10102         btrfs_set_header_generation(c, trans->transid);
10103         btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
10104         btrfs_set_header_owner(c, root->root_key.objectid);
10105
10106         write_extent_buffer(c, root->fs_info->fsid,
10107                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
10108
10109         write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
10110                             btrfs_header_chunk_tree_uuid(c),
10111                             BTRFS_UUID_SIZE);
10112
10113         btrfs_mark_buffer_dirty(c);
10114         /*
10115          * this case can happen in the following case:
10116          *
10117          * 1.overwrite previous root.
10118          *
10119          * 2.reinit reloc data root, this is because we skip pin
10120          * down reloc data tree before which means we can allocate
10121          * same block bytenr here.
10122          */
10123         if (old->start == c->start) {
10124                 btrfs_set_root_generation(&root->root_item,
10125                                           trans->transid);
10126                 root->root_item.level = btrfs_header_level(root->node);
10127                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
10128                                         &root->root_key, &root->root_item);
10129                 if (ret) {
10130                         free_extent_buffer(c);
10131                         return ret;
10132                 }
10133         }
10134         free_extent_buffer(old);
10135         root->node = c;
10136         add_root_to_dirty_list(root);
10137         return 0;
10138 }
10139
10140 static int pin_down_tree_blocks(struct btrfs_fs_info *fs_info,
10141                                 struct extent_buffer *eb, int tree_root)
10142 {
10143         struct extent_buffer *tmp;
10144         struct btrfs_root_item *ri;
10145         struct btrfs_key key;
10146         u64 bytenr;
10147         u32 nodesize;
10148         int level = btrfs_header_level(eb);
10149         int nritems;
10150         int ret;
10151         int i;
10152
10153         /*
10154          * If we have pinned this block before, don't pin it again.
10155          * This can not only avoid forever loop with broken filesystem
10156          * but also give us some speedups.
10157          */
10158         if (test_range_bit(&fs_info->pinned_extents, eb->start,
10159                            eb->start + eb->len - 1, EXTENT_DIRTY, 0))
10160                 return 0;
10161
10162         btrfs_pin_extent(fs_info, eb->start, eb->len);
10163
10164         nodesize = btrfs_super_nodesize(fs_info->super_copy);
10165         nritems = btrfs_header_nritems(eb);
10166         for (i = 0; i < nritems; i++) {
10167                 if (level == 0) {
10168                         btrfs_item_key_to_cpu(eb, &key, i);
10169                         if (key.type != BTRFS_ROOT_ITEM_KEY)
10170                                 continue;
10171                         /* Skip the extent root and reloc roots */
10172                         if (key.objectid == BTRFS_EXTENT_TREE_OBJECTID ||
10173                             key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
10174                             key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
10175                                 continue;
10176                         ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
10177                         bytenr = btrfs_disk_root_bytenr(eb, ri);
10178
10179                         /*
10180                          * If at any point we start needing the real root we
10181                          * will have to build a stump root for the root we are
10182                          * in, but for now this doesn't actually use the root so
10183                          * just pass in extent_root.
10184                          */
10185                         tmp = read_tree_block(fs_info->extent_root, bytenr,
10186                                               nodesize, 0);
10187                         if (!extent_buffer_uptodate(tmp)) {
10188                                 fprintf(stderr, "Error reading root block\n");
10189                                 return -EIO;
10190                         }
10191                         ret = pin_down_tree_blocks(fs_info, tmp, 0);
10192                         free_extent_buffer(tmp);
10193                         if (ret)
10194                                 return ret;
10195                 } else {
10196                         bytenr = btrfs_node_blockptr(eb, i);
10197
10198                         /* If we aren't the tree root don't read the block */
10199                         if (level == 1 && !tree_root) {
10200                                 btrfs_pin_extent(fs_info, bytenr, nodesize);
10201                                 continue;
10202                         }
10203
10204                         tmp = read_tree_block(fs_info->extent_root, bytenr,
10205                                               nodesize, 0);
10206                         if (!extent_buffer_uptodate(tmp)) {
10207                                 fprintf(stderr, "Error reading tree block\n");
10208                                 return -EIO;
10209                         }
10210                         ret = pin_down_tree_blocks(fs_info, tmp, tree_root);
10211                         free_extent_buffer(tmp);
10212                         if (ret)
10213                                 return ret;
10214                 }
10215         }
10216
10217         return 0;
10218 }
10219
10220 static int pin_metadata_blocks(struct btrfs_fs_info *fs_info)
10221 {
10222         int ret;
10223
10224         ret = pin_down_tree_blocks(fs_info, fs_info->chunk_root->node, 0);
10225         if (ret)
10226                 return ret;
10227
10228         return pin_down_tree_blocks(fs_info, fs_info->tree_root->node, 1);
10229 }
10230
10231 static int reset_block_groups(struct btrfs_fs_info *fs_info)
10232 {
10233         struct btrfs_block_group_cache *cache;
10234         struct btrfs_path *path;
10235         struct extent_buffer *leaf;
10236         struct btrfs_chunk *chunk;
10237         struct btrfs_key key;
10238         int ret;
10239         u64 start;
10240
10241         path = btrfs_alloc_path();
10242         if (!path)
10243                 return -ENOMEM;
10244
10245         key.objectid = 0;
10246         key.type = BTRFS_CHUNK_ITEM_KEY;
10247         key.offset = 0;
10248
10249         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
10250         if (ret < 0) {
10251                 btrfs_free_path(path);
10252                 return ret;
10253         }
10254
10255         /*
10256          * We do this in case the block groups were screwed up and had alloc
10257          * bits that aren't actually set on the chunks.  This happens with
10258          * restored images every time and could happen in real life I guess.
10259          */
10260         fs_info->avail_data_alloc_bits = 0;
10261         fs_info->avail_metadata_alloc_bits = 0;
10262         fs_info->avail_system_alloc_bits = 0;
10263
10264         /* First we need to create the in-memory block groups */
10265         while (1) {
10266                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
10267                         ret = btrfs_next_leaf(fs_info->chunk_root, path);
10268                         if (ret < 0) {
10269                                 btrfs_free_path(path);
10270                                 return ret;
10271                         }
10272                         if (ret) {
10273                                 ret = 0;
10274                                 break;
10275                         }
10276                 }
10277                 leaf = path->nodes[0];
10278                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
10279                 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
10280                         path->slots[0]++;
10281                         continue;
10282                 }
10283
10284                 chunk = btrfs_item_ptr(leaf, path->slots[0],
10285                                        struct btrfs_chunk);
10286                 btrfs_add_block_group(fs_info, 0,
10287                                       btrfs_chunk_type(leaf, chunk),
10288                                       key.objectid, key.offset,
10289                                       btrfs_chunk_length(leaf, chunk));
10290                 set_extent_dirty(&fs_info->free_space_cache, key.offset,
10291                                  key.offset + btrfs_chunk_length(leaf, chunk),
10292                                  GFP_NOFS);
10293                 path->slots[0]++;
10294         }
10295         start = 0;
10296         while (1) {
10297                 cache = btrfs_lookup_first_block_group(fs_info, start);
10298                 if (!cache)
10299                         break;
10300                 cache->cached = 1;
10301                 start = cache->key.objectid + cache->key.offset;
10302         }
10303
10304         btrfs_free_path(path);
10305         return 0;
10306 }
10307
10308 static int reset_balance(struct btrfs_trans_handle *trans,
10309                          struct btrfs_fs_info *fs_info)
10310 {
10311         struct btrfs_root *root = fs_info->tree_root;
10312         struct btrfs_path *path;
10313         struct extent_buffer *leaf;
10314         struct btrfs_key key;
10315         int del_slot, del_nr = 0;
10316         int ret;
10317         int found = 0;
10318
10319         path = btrfs_alloc_path();
10320         if (!path)
10321                 return -ENOMEM;
10322
10323         key.objectid = BTRFS_BALANCE_OBJECTID;
10324         key.type = BTRFS_BALANCE_ITEM_KEY;
10325         key.offset = 0;
10326
10327         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10328         if (ret) {
10329                 if (ret > 0)
10330                         ret = 0;
10331                 if (!ret)
10332                         goto reinit_data_reloc;
10333                 else
10334                         goto out;
10335         }
10336
10337         ret = btrfs_del_item(trans, root, path);
10338         if (ret)
10339                 goto out;
10340         btrfs_release_path(path);
10341
10342         key.objectid = BTRFS_TREE_RELOC_OBJECTID;
10343         key.type = BTRFS_ROOT_ITEM_KEY;
10344         key.offset = 0;
10345
10346         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10347         if (ret < 0)
10348                 goto out;
10349         while (1) {
10350                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
10351                         if (!found)
10352                                 break;
10353
10354                         if (del_nr) {
10355                                 ret = btrfs_del_items(trans, root, path,
10356                                                       del_slot, del_nr);
10357                                 del_nr = 0;
10358                                 if (ret)
10359                                         goto out;
10360                         }
10361                         key.offset++;
10362                         btrfs_release_path(path);
10363
10364                         found = 0;
10365                         ret = btrfs_search_slot(trans, root, &key, path,
10366                                                 -1, 1);
10367                         if (ret < 0)
10368                                 goto out;
10369                         continue;
10370                 }
10371                 found = 1;
10372                 leaf = path->nodes[0];
10373                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
10374                 if (key.objectid > BTRFS_TREE_RELOC_OBJECTID)
10375                         break;
10376                 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
10377                         path->slots[0]++;
10378                         continue;
10379                 }
10380                 if (!del_nr) {
10381                         del_slot = path->slots[0];
10382                         del_nr = 1;
10383                 } else {
10384                         del_nr++;
10385                 }
10386                 path->slots[0]++;
10387         }
10388
10389         if (del_nr) {
10390                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
10391                 if (ret)
10392                         goto out;
10393         }
10394         btrfs_release_path(path);
10395
10396 reinit_data_reloc:
10397         key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
10398         key.type = BTRFS_ROOT_ITEM_KEY;
10399         key.offset = (u64)-1;
10400         root = btrfs_read_fs_root(fs_info, &key);
10401         if (IS_ERR(root)) {
10402                 fprintf(stderr, "Error reading data reloc tree\n");
10403                 ret = PTR_ERR(root);
10404                 goto out;
10405         }
10406         record_root_in_trans(trans, root);
10407         ret = btrfs_fsck_reinit_root(trans, root, 0);
10408         if (ret)
10409                 goto out;
10410         ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
10411 out:
10412         btrfs_free_path(path);
10413         return ret;
10414 }
10415
10416 static int reinit_extent_tree(struct btrfs_trans_handle *trans,
10417                               struct btrfs_fs_info *fs_info)
10418 {
10419         u64 start = 0;
10420         int ret;
10421
10422         /*
10423          * The only reason we don't do this is because right now we're just
10424          * walking the trees we find and pinning down their bytes, we don't look
10425          * at any of the leaves.  In order to do mixed groups we'd have to check
10426          * the leaves of any fs roots and pin down the bytes for any file
10427          * extents we find.  Not hard but why do it if we don't have to?
10428          */
10429         if (btrfs_fs_incompat(fs_info, BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)) {
10430                 fprintf(stderr, "We don't support re-initing the extent tree "
10431                         "for mixed block groups yet, please notify a btrfs "
10432                         "developer you want to do this so they can add this "
10433                         "functionality.\n");
10434                 return -EINVAL;
10435         }
10436
10437         /*
10438          * first we need to walk all of the trees except the extent tree and pin
10439          * down the bytes that are in use so we don't overwrite any existing
10440          * metadata.
10441          */
10442         ret = pin_metadata_blocks(fs_info);
10443         if (ret) {
10444                 fprintf(stderr, "error pinning down used bytes\n");
10445                 return ret;
10446         }
10447
10448         /*
10449          * Need to drop all the block groups since we're going to recreate all
10450          * of them again.
10451          */
10452         btrfs_free_block_groups(fs_info);
10453         ret = reset_block_groups(fs_info);
10454         if (ret) {
10455                 fprintf(stderr, "error resetting the block groups\n");
10456                 return ret;
10457         }
10458
10459         /* Ok we can allocate now, reinit the extent root */
10460         ret = btrfs_fsck_reinit_root(trans, fs_info->extent_root, 0);
10461         if (ret) {
10462                 fprintf(stderr, "extent root initialization failed\n");
10463                 /*
10464                  * When the transaction code is updated we should end the
10465                  * transaction, but for now progs only knows about commit so
10466                  * just return an error.
10467                  */
10468                 return ret;
10469         }
10470
10471         /*
10472          * Now we have all the in-memory block groups setup so we can make
10473          * allocations properly, and the metadata we care about is safe since we
10474          * pinned all of it above.
10475          */
10476         while (1) {
10477                 struct btrfs_block_group_cache *cache;
10478
10479                 cache = btrfs_lookup_first_block_group(fs_info, start);
10480                 if (!cache)
10481                         break;
10482                 start = cache->key.objectid + cache->key.offset;
10483                 ret = btrfs_insert_item(trans, fs_info->extent_root,
10484                                         &cache->key, &cache->item,
10485                                         sizeof(cache->item));
10486                 if (ret) {
10487                         fprintf(stderr, "Error adding block group\n");
10488                         return ret;
10489                 }
10490                 btrfs_extent_post_op(trans, fs_info->extent_root);
10491         }
10492
10493         ret = reset_balance(trans, fs_info);
10494         if (ret)
10495                 fprintf(stderr, "error resetting the pending balance\n");
10496
10497         return ret;
10498 }
10499
10500 static int recow_extent_buffer(struct btrfs_root *root, struct extent_buffer *eb)
10501 {
10502         struct btrfs_path *path;
10503         struct btrfs_trans_handle *trans;
10504         struct btrfs_key key;
10505         int ret;
10506
10507         printf("Recowing metadata block %llu\n", eb->start);
10508         key.objectid = btrfs_header_owner(eb);
10509         key.type = BTRFS_ROOT_ITEM_KEY;
10510         key.offset = (u64)-1;
10511
10512         root = btrfs_read_fs_root(root->fs_info, &key);
10513         if (IS_ERR(root)) {
10514                 fprintf(stderr, "Couldn't find owner root %llu\n",
10515                         key.objectid);
10516                 return PTR_ERR(root);
10517         }
10518
10519         path = btrfs_alloc_path();
10520         if (!path)
10521                 return -ENOMEM;
10522
10523         trans = btrfs_start_transaction(root, 1);
10524         if (IS_ERR(trans)) {
10525                 btrfs_free_path(path);
10526                 return PTR_ERR(trans);
10527         }
10528
10529         path->lowest_level = btrfs_header_level(eb);
10530         if (path->lowest_level)
10531                 btrfs_node_key_to_cpu(eb, &key, 0);
10532         else
10533                 btrfs_item_key_to_cpu(eb, &key, 0);
10534
10535         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
10536         btrfs_commit_transaction(trans, root);
10537         btrfs_free_path(path);
10538         return ret;
10539 }
10540
10541 static int delete_bad_item(struct btrfs_root *root, struct bad_item *bad)
10542 {
10543         struct btrfs_path *path;
10544         struct btrfs_trans_handle *trans;
10545         struct btrfs_key key;
10546         int ret;
10547
10548         printf("Deleting bad item [%llu,%u,%llu]\n", bad->key.objectid,
10549                bad->key.type, bad->key.offset);
10550         key.objectid = bad->root_id;
10551         key.type = BTRFS_ROOT_ITEM_KEY;
10552         key.offset = (u64)-1;
10553
10554         root = btrfs_read_fs_root(root->fs_info, &key);
10555         if (IS_ERR(root)) {
10556                 fprintf(stderr, "Couldn't find owner root %llu\n",
10557                         key.objectid);
10558                 return PTR_ERR(root);
10559         }
10560
10561         path = btrfs_alloc_path();
10562         if (!path)
10563                 return -ENOMEM;
10564
10565         trans = btrfs_start_transaction(root, 1);
10566         if (IS_ERR(trans)) {
10567                 btrfs_free_path(path);
10568                 return PTR_ERR(trans);
10569         }
10570
10571         ret = btrfs_search_slot(trans, root, &bad->key, path, -1, 1);
10572         if (ret) {
10573                 if (ret > 0)
10574                         ret = 0;
10575                 goto out;
10576         }
10577         ret = btrfs_del_item(trans, root, path);
10578 out:
10579         btrfs_commit_transaction(trans, root);
10580         btrfs_free_path(path);
10581         return ret;
10582 }
10583
10584 static int zero_log_tree(struct btrfs_root *root)
10585 {
10586         struct btrfs_trans_handle *trans;
10587         int ret;
10588
10589         trans = btrfs_start_transaction(root, 1);
10590         if (IS_ERR(trans)) {
10591                 ret = PTR_ERR(trans);
10592                 return ret;
10593         }
10594         btrfs_set_super_log_root(root->fs_info->super_copy, 0);
10595         btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
10596         ret = btrfs_commit_transaction(trans, root);
10597         return ret;
10598 }
10599
10600 static int populate_csum(struct btrfs_trans_handle *trans,
10601                          struct btrfs_root *csum_root, char *buf, u64 start,
10602                          u64 len)
10603 {
10604         u64 offset = 0;
10605         u64 sectorsize;
10606         int ret = 0;
10607
10608         while (offset < len) {
10609                 sectorsize = csum_root->sectorsize;
10610                 ret = read_extent_data(csum_root, buf, start + offset,
10611                                        &sectorsize, 0);
10612                 if (ret)
10613                         break;
10614                 ret = btrfs_csum_file_block(trans, csum_root, start + len,
10615                                             start + offset, buf, sectorsize);
10616                 if (ret)
10617                         break;
10618                 offset += sectorsize;
10619         }
10620         return ret;
10621 }
10622
10623 static int fill_csum_tree_from_one_fs_root(struct btrfs_trans_handle *trans,
10624                                       struct btrfs_root *csum_root,
10625                                       struct btrfs_root *cur_root)
10626 {
10627         struct btrfs_path *path;
10628         struct btrfs_key key;
10629         struct extent_buffer *node;
10630         struct btrfs_file_extent_item *fi;
10631         char *buf = NULL;
10632         u64 start = 0;
10633         u64 len = 0;
10634         int slot = 0;
10635         int ret = 0;
10636
10637         path = btrfs_alloc_path();
10638         if (!path)
10639                 return -ENOMEM;
10640         buf = malloc(cur_root->fs_info->csum_root->sectorsize);
10641         if (!buf) {
10642                 ret = -ENOMEM;
10643                 goto out;
10644         }
10645
10646         key.objectid = 0;
10647         key.offset = 0;
10648         key.type = 0;
10649
10650         ret = btrfs_search_slot(NULL, cur_root, &key, path, 0, 0);
10651         if (ret < 0)
10652                 goto out;
10653         /* Iterate all regular file extents and fill its csum */
10654         while (1) {
10655                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
10656
10657                 if (key.type != BTRFS_EXTENT_DATA_KEY)
10658                         goto next;
10659                 node = path->nodes[0];
10660                 slot = path->slots[0];
10661                 fi = btrfs_item_ptr(node, slot, struct btrfs_file_extent_item);
10662                 if (btrfs_file_extent_type(node, fi) != BTRFS_FILE_EXTENT_REG)
10663                         goto next;
10664                 start = btrfs_file_extent_disk_bytenr(node, fi);
10665                 len = btrfs_file_extent_disk_num_bytes(node, fi);
10666
10667                 ret = populate_csum(trans, csum_root, buf, start, len);
10668                 if (ret == -EEXIST)
10669                         ret = 0;
10670                 if (ret < 0)
10671                         goto out;
10672 next:
10673                 /*
10674                  * TODO: if next leaf is corrupted, jump to nearest next valid
10675                  * leaf.
10676                  */
10677                 ret = btrfs_next_item(cur_root, path);
10678                 if (ret < 0)
10679                         goto out;
10680                 if (ret > 0) {
10681                         ret = 0;
10682                         goto out;
10683                 }
10684         }
10685
10686 out:
10687         btrfs_free_path(path);
10688         free(buf);
10689         return ret;
10690 }
10691
10692 static int fill_csum_tree_from_fs(struct btrfs_trans_handle *trans,
10693                                   struct btrfs_root *csum_root)
10694 {
10695         struct btrfs_fs_info *fs_info = csum_root->fs_info;
10696         struct btrfs_path *path;
10697         struct btrfs_root *tree_root = fs_info->tree_root;
10698         struct btrfs_root *cur_root;
10699         struct extent_buffer *node;
10700         struct btrfs_key key;
10701         int slot = 0;
10702         int ret = 0;
10703
10704         path = btrfs_alloc_path();
10705         if (!path)
10706                 return -ENOMEM;
10707
10708         key.objectid = BTRFS_FS_TREE_OBJECTID;
10709         key.offset = 0;
10710         key.type = BTRFS_ROOT_ITEM_KEY;
10711
10712         ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
10713         if (ret < 0)
10714                 goto out;
10715         if (ret > 0) {
10716                 ret = -ENOENT;
10717                 goto out;
10718         }
10719
10720         while (1) {
10721                 node = path->nodes[0];
10722                 slot = path->slots[0];
10723                 btrfs_item_key_to_cpu(node, &key, slot);
10724                 if (key.objectid > BTRFS_LAST_FREE_OBJECTID)
10725                         goto out;
10726                 if (key.type != BTRFS_ROOT_ITEM_KEY)
10727                         goto next;
10728                 if (!is_fstree(key.objectid))
10729                         goto next;
10730                 key.offset = (u64)-1;
10731
10732                 cur_root = btrfs_read_fs_root(fs_info, &key);
10733                 if (IS_ERR(cur_root) || !cur_root) {
10734                         fprintf(stderr, "Fail to read fs/subvol tree: %lld\n",
10735                                 key.objectid);
10736                         goto out;
10737                 }
10738                 ret = fill_csum_tree_from_one_fs_root(trans, csum_root,
10739                                 cur_root);
10740                 if (ret < 0)
10741                         goto out;
10742 next:
10743                 ret = btrfs_next_item(tree_root, path);
10744                 if (ret > 0) {
10745                         ret = 0;
10746                         goto out;
10747                 }
10748                 if (ret < 0)
10749                         goto out;
10750         }
10751
10752 out:
10753         btrfs_free_path(path);
10754         return ret;
10755 }
10756
10757 static int fill_csum_tree_from_extent(struct btrfs_trans_handle *trans,
10758                                       struct btrfs_root *csum_root)
10759 {
10760         struct btrfs_root *extent_root = csum_root->fs_info->extent_root;
10761         struct btrfs_path *path;
10762         struct btrfs_extent_item *ei;
10763         struct extent_buffer *leaf;
10764         char *buf;
10765         struct btrfs_key key;
10766         int ret;
10767
10768         path = btrfs_alloc_path();
10769         if (!path)
10770                 return -ENOMEM;
10771
10772         key.objectid = 0;
10773         key.type = BTRFS_EXTENT_ITEM_KEY;
10774         key.offset = 0;
10775
10776         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
10777         if (ret < 0) {
10778                 btrfs_free_path(path);
10779                 return ret;
10780         }
10781
10782         buf = malloc(csum_root->sectorsize);
10783         if (!buf) {
10784                 btrfs_free_path(path);
10785                 return -ENOMEM;
10786         }
10787
10788         while (1) {
10789                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
10790                         ret = btrfs_next_leaf(extent_root, path);
10791                         if (ret < 0)
10792                                 break;
10793                         if (ret) {
10794                                 ret = 0;
10795                                 break;
10796                         }
10797                 }
10798                 leaf = path->nodes[0];
10799
10800                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
10801                 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
10802                         path->slots[0]++;
10803                         continue;
10804                 }
10805
10806                 ei = btrfs_item_ptr(leaf, path->slots[0],
10807                                     struct btrfs_extent_item);
10808                 if (!(btrfs_extent_flags(leaf, ei) &
10809                       BTRFS_EXTENT_FLAG_DATA)) {
10810                         path->slots[0]++;
10811                         continue;
10812                 }
10813
10814                 ret = populate_csum(trans, csum_root, buf, key.objectid,
10815                                     key.offset);
10816                 if (ret)
10817                         break;
10818                 path->slots[0]++;
10819         }
10820
10821         btrfs_free_path(path);
10822         free(buf);
10823         return ret;
10824 }
10825
10826 /*
10827  * Recalculate the csum and put it into the csum tree.
10828  *
10829  * Extent tree init will wipe out all the extent info, so in that case, we
10830  * can't depend on extent tree, but use fs tree.  If search_fs_tree is set, we
10831  * will use fs/subvol trees to init the csum tree.
10832  */
10833 static int fill_csum_tree(struct btrfs_trans_handle *trans,
10834                           struct btrfs_root *csum_root,
10835                           int search_fs_tree)
10836 {
10837         if (search_fs_tree)
10838                 return fill_csum_tree_from_fs(trans, csum_root);
10839         else
10840                 return fill_csum_tree_from_extent(trans, csum_root);
10841 }
10842
10843 static void free_roots_info_cache(void)
10844 {
10845         if (!roots_info_cache)
10846                 return;
10847
10848         while (!cache_tree_empty(roots_info_cache)) {
10849                 struct cache_extent *entry;
10850                 struct root_item_info *rii;
10851
10852                 entry = first_cache_extent(roots_info_cache);
10853                 if (!entry)
10854                         break;
10855                 remove_cache_extent(roots_info_cache, entry);
10856                 rii = container_of(entry, struct root_item_info, cache_extent);
10857                 free(rii);
10858         }
10859
10860         free(roots_info_cache);
10861         roots_info_cache = NULL;
10862 }
10863
10864 static int build_roots_info_cache(struct btrfs_fs_info *info)
10865 {
10866         int ret = 0;
10867         struct btrfs_key key;
10868         struct extent_buffer *leaf;
10869         struct btrfs_path *path;
10870
10871         if (!roots_info_cache) {
10872                 roots_info_cache = malloc(sizeof(*roots_info_cache));
10873                 if (!roots_info_cache)
10874                         return -ENOMEM;
10875                 cache_tree_init(roots_info_cache);
10876         }
10877
10878         path = btrfs_alloc_path();
10879         if (!path)
10880                 return -ENOMEM;
10881
10882         key.objectid = 0;
10883         key.type = BTRFS_EXTENT_ITEM_KEY;
10884         key.offset = 0;
10885
10886         ret = btrfs_search_slot(NULL, info->extent_root, &key, path, 0, 0);
10887         if (ret < 0)
10888                 goto out;
10889         leaf = path->nodes[0];
10890
10891         while (1) {
10892                 struct btrfs_key found_key;
10893                 struct btrfs_extent_item *ei;
10894                 struct btrfs_extent_inline_ref *iref;
10895                 int slot = path->slots[0];
10896                 int type;
10897                 u64 flags;
10898                 u64 root_id;
10899                 u8 level;
10900                 struct cache_extent *entry;
10901                 struct root_item_info *rii;
10902
10903                 if (slot >= btrfs_header_nritems(leaf)) {
10904                         ret = btrfs_next_leaf(info->extent_root, path);
10905                         if (ret < 0) {
10906                                 break;
10907                         } else if (ret) {
10908                                 ret = 0;
10909                                 break;
10910                         }
10911                         leaf = path->nodes[0];
10912                         slot = path->slots[0];
10913                 }
10914
10915                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10916
10917                 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
10918                     found_key.type != BTRFS_METADATA_ITEM_KEY)
10919                         goto next;
10920
10921                 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
10922                 flags = btrfs_extent_flags(leaf, ei);
10923
10924                 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
10925                     !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
10926                         goto next;
10927
10928                 if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
10929                         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
10930                         level = found_key.offset;
10931                 } else {
10932                         struct btrfs_tree_block_info *binfo;
10933
10934                         binfo = (struct btrfs_tree_block_info *)(ei + 1);
10935                         iref = (struct btrfs_extent_inline_ref *)(binfo + 1);
10936                         level = btrfs_tree_block_level(leaf, binfo);
10937                 }
10938
10939                 /*
10940                  * For a root extent, it must be of the following type and the
10941                  * first (and only one) iref in the item.
10942                  */
10943                 type = btrfs_extent_inline_ref_type(leaf, iref);
10944                 if (type != BTRFS_TREE_BLOCK_REF_KEY)
10945                         goto next;
10946
10947                 root_id = btrfs_extent_inline_ref_offset(leaf, iref);
10948                 entry = lookup_cache_extent(roots_info_cache, root_id, 1);
10949                 if (!entry) {
10950                         rii = malloc(sizeof(struct root_item_info));
10951                         if (!rii) {
10952                                 ret = -ENOMEM;
10953                                 goto out;
10954                         }
10955                         rii->cache_extent.start = root_id;
10956                         rii->cache_extent.size = 1;
10957                         rii->level = (u8)-1;
10958                         entry = &rii->cache_extent;
10959                         ret = insert_cache_extent(roots_info_cache, entry);
10960                         ASSERT(ret == 0);
10961                 } else {
10962                         rii = container_of(entry, struct root_item_info,
10963                                            cache_extent);
10964                 }
10965
10966                 ASSERT(rii->cache_extent.start == root_id);
10967                 ASSERT(rii->cache_extent.size == 1);
10968
10969                 if (level > rii->level || rii->level == (u8)-1) {
10970                         rii->level = level;
10971                         rii->bytenr = found_key.objectid;
10972                         rii->gen = btrfs_extent_generation(leaf, ei);
10973                         rii->node_count = 1;
10974                 } else if (level == rii->level) {
10975                         rii->node_count++;
10976                 }
10977 next:
10978                 path->slots[0]++;
10979         }
10980
10981 out:
10982         btrfs_free_path(path);
10983
10984         return ret;
10985 }
10986
10987 static int maybe_repair_root_item(struct btrfs_fs_info *info,
10988                                   struct btrfs_path *path,
10989                                   const struct btrfs_key *root_key,
10990                                   const int read_only_mode)
10991 {
10992         const u64 root_id = root_key->objectid;
10993         struct cache_extent *entry;
10994         struct root_item_info *rii;
10995         struct btrfs_root_item ri;
10996         unsigned long offset;
10997
10998         entry = lookup_cache_extent(roots_info_cache, root_id, 1);
10999         if (!entry) {
11000                 fprintf(stderr,
11001                         "Error: could not find extent items for root %llu\n",
11002                         root_key->objectid);
11003                 return -ENOENT;
11004         }
11005
11006         rii = container_of(entry, struct root_item_info, cache_extent);
11007         ASSERT(rii->cache_extent.start == root_id);
11008         ASSERT(rii->cache_extent.size == 1);
11009
11010         if (rii->node_count != 1) {
11011                 fprintf(stderr,
11012                         "Error: could not find btree root extent for root %llu\n",
11013                         root_id);
11014                 return -ENOENT;
11015         }
11016
11017         offset = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
11018         read_extent_buffer(path->nodes[0], &ri, offset, sizeof(ri));
11019
11020         if (btrfs_root_bytenr(&ri) != rii->bytenr ||
11021             btrfs_root_level(&ri) != rii->level ||
11022             btrfs_root_generation(&ri) != rii->gen) {
11023
11024                 /*
11025                  * If we're in repair mode but our caller told us to not update
11026                  * the root item, i.e. just check if it needs to be updated, don't
11027                  * print this message, since the caller will call us again shortly
11028                  * for the same root item without read only mode (the caller will
11029                  * open a transaction first).
11030                  */
11031                 if (!(read_only_mode && repair))
11032                         fprintf(stderr,
11033                                 "%sroot item for root %llu,"
11034                                 " current bytenr %llu, current gen %llu, current level %u,"
11035                                 " new bytenr %llu, new gen %llu, new level %u\n",
11036                                 (read_only_mode ? "" : "fixing "),
11037                                 root_id,
11038                                 btrfs_root_bytenr(&ri), btrfs_root_generation(&ri),
11039                                 btrfs_root_level(&ri),
11040                                 rii->bytenr, rii->gen, rii->level);
11041
11042                 if (btrfs_root_generation(&ri) > rii->gen) {
11043                         fprintf(stderr,
11044                                 "root %llu has a root item with a more recent gen (%llu) compared to the found root node (%llu)\n",
11045                                 root_id, btrfs_root_generation(&ri), rii->gen);
11046                         return -EINVAL;
11047                 }
11048
11049                 if (!read_only_mode) {
11050                         btrfs_set_root_bytenr(&ri, rii->bytenr);
11051                         btrfs_set_root_level(&ri, rii->level);
11052                         btrfs_set_root_generation(&ri, rii->gen);
11053                         write_extent_buffer(path->nodes[0], &ri,
11054                                             offset, sizeof(ri));
11055                 }
11056
11057                 return 1;
11058         }
11059
11060         return 0;
11061 }
11062
11063 /*
11064  * A regression introduced in the 3.17 kernel (more specifically in 3.17-rc2),
11065  * caused read-only snapshots to be corrupted if they were created at a moment
11066  * when the source subvolume/snapshot had orphan items. The issue was that the
11067  * on-disk root items became incorrect, referring to the pre orphan cleanup root
11068  * node instead of the post orphan cleanup root node.
11069  * So this function, and its callees, just detects and fixes those cases. Even
11070  * though the regression was for read-only snapshots, this function applies to
11071  * any snapshot/subvolume root.
11072  * This must be run before any other repair code - not doing it so, makes other
11073  * repair code delete or modify backrefs in the extent tree for example, which
11074  * will result in an inconsistent fs after repairing the root items.
11075  */
11076 static int repair_root_items(struct btrfs_fs_info *info)
11077 {
11078         struct btrfs_path *path = NULL;
11079         struct btrfs_key key;
11080         struct extent_buffer *leaf;
11081         struct btrfs_trans_handle *trans = NULL;
11082         int ret = 0;
11083         int bad_roots = 0;
11084         int need_trans = 0;
11085
11086         ret = build_roots_info_cache(info);
11087         if (ret)
11088                 goto out;
11089
11090         path = btrfs_alloc_path();
11091         if (!path) {
11092                 ret = -ENOMEM;
11093                 goto out;
11094         }
11095
11096         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
11097         key.type = BTRFS_ROOT_ITEM_KEY;
11098         key.offset = 0;
11099
11100 again:
11101         /*
11102          * Avoid opening and committing transactions if a leaf doesn't have
11103          * any root items that need to be fixed, so that we avoid rotating
11104          * backup roots unnecessarily.
11105          */
11106         if (need_trans) {
11107                 trans = btrfs_start_transaction(info->tree_root, 1);
11108                 if (IS_ERR(trans)) {
11109                         ret = PTR_ERR(trans);
11110                         goto out;
11111                 }
11112         }
11113
11114         ret = btrfs_search_slot(trans, info->tree_root, &key, path,
11115                                 0, trans ? 1 : 0);
11116         if (ret < 0)
11117                 goto out;
11118         leaf = path->nodes[0];
11119
11120         while (1) {
11121                 struct btrfs_key found_key;
11122
11123                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
11124                         int no_more_keys = find_next_key(path, &key);
11125
11126                         btrfs_release_path(path);
11127                         if (trans) {
11128                                 ret = btrfs_commit_transaction(trans,
11129                                                                info->tree_root);
11130                                 trans = NULL;
11131                                 if (ret < 0)
11132                                         goto out;
11133                         }
11134                         need_trans = 0;
11135                         if (no_more_keys)
11136                                 break;
11137                         goto again;
11138                 }
11139
11140                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
11141
11142                 if (found_key.type != BTRFS_ROOT_ITEM_KEY)
11143                         goto next;
11144                 if (found_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
11145                         goto next;
11146
11147                 ret = maybe_repair_root_item(info, path, &found_key,
11148                                              trans ? 0 : 1);
11149                 if (ret < 0)
11150                         goto out;
11151                 if (ret) {
11152                         if (!trans && repair) {
11153                                 need_trans = 1;
11154                                 key = found_key;
11155                                 btrfs_release_path(path);
11156                                 goto again;
11157                         }
11158                         bad_roots++;
11159                 }
11160 next:
11161                 path->slots[0]++;
11162         }
11163         ret = 0;
11164 out:
11165         free_roots_info_cache();
11166         btrfs_free_path(path);
11167         if (trans)
11168                 btrfs_commit_transaction(trans, info->tree_root);
11169         if (ret < 0)
11170                 return ret;
11171
11172         return bad_roots;
11173 }
11174
11175 static int clear_free_space_cache(struct btrfs_fs_info *fs_info)
11176 {
11177         struct btrfs_trans_handle *trans;
11178         struct btrfs_block_group_cache *bg_cache;
11179         u64 current = 0;
11180         int ret = 0;
11181
11182         /* Clear all free space cache inodes and its extent data */
11183         while (1) {
11184                 bg_cache = btrfs_lookup_first_block_group(fs_info, current);
11185                 if (!bg_cache)
11186                         break;
11187                 ret = btrfs_clear_free_space_cache(fs_info, bg_cache);
11188                 if (ret < 0)
11189                         return ret;
11190                 current = bg_cache->key.objectid + bg_cache->key.offset;
11191         }
11192
11193         /* Don't forget to set cache_generation to -1 */
11194         trans = btrfs_start_transaction(fs_info->tree_root, 0);
11195         if (IS_ERR(trans)) {
11196                 error("failed to update super block cache generation");
11197                 return PTR_ERR(trans);
11198         }
11199         btrfs_set_super_cache_generation(fs_info->super_copy, (u64)-1);
11200         btrfs_commit_transaction(trans, fs_info->tree_root);
11201
11202         return ret;
11203 }
11204
11205 const char * const cmd_check_usage[] = {
11206         "btrfs check [options] <device>",
11207         "Check structural integrity of a filesystem (unmounted).",
11208         "Check structural integrity of an unmounted filesystem. Verify internal",
11209         "trees' consistency and item connectivity. In the repair mode try to",
11210         "fix the problems found. ",
11211         "WARNING: the repair mode is considered dangerous",
11212         "",
11213         "-s|--super <superblock>     use this superblock copy",
11214         "-b|--backup                 use the first valid backup root copy",
11215         "--repair                    try to repair the filesystem",
11216         "--readonly                  run in read-only mode (default)",
11217         "--init-csum-tree            create a new CRC tree",
11218         "--init-extent-tree          create a new extent tree",
11219         "--mode <MODE>               allows choice of memory/IO trade-offs",
11220         "                            where MODE is one of:",
11221         "                            original - read inodes and extents to memory (requires",
11222         "                                       more memory, does less IO)",
11223         "                            lowmem   - try to use less memory but read blocks again",
11224         "                                       when needed",
11225         "--check-data-csum           verify checksums of data blocks",
11226         "-Q|--qgroup-report          print a report on qgroup consistency",
11227         "-E|--subvol-extents <subvolid>",
11228         "                            print subvolume extents and sharing state",
11229         "-r|--tree-root <bytenr>     use the given bytenr for the tree root",
11230         "--chunk-root <bytenr>       use the given bytenr for the chunk tree root",
11231         "-p|--progress               indicate progress",
11232         "--clear-space-cache v1|v2   clear space cache for v1 or v2",
11233         "                            NOTE: v1 support implemented",
11234         NULL
11235 };
11236
11237 int cmd_check(int argc, char **argv)
11238 {
11239         struct cache_tree root_cache;
11240         struct btrfs_root *root;
11241         struct btrfs_fs_info *info;
11242         u64 bytenr = 0;
11243         u64 subvolid = 0;
11244         u64 tree_root_bytenr = 0;
11245         u64 chunk_root_bytenr = 0;
11246         char uuidbuf[BTRFS_UUID_UNPARSED_SIZE];
11247         int ret;
11248         u64 num;
11249         int init_csum_tree = 0;
11250         int readonly = 0;
11251         int clear_space_cache = 0;
11252         int qgroup_report = 0;
11253         int qgroups_repaired = 0;
11254         unsigned ctree_flags = OPEN_CTREE_EXCLUSIVE;
11255
11256         while(1) {
11257                 int c;
11258                 enum { GETOPT_VAL_REPAIR = 257, GETOPT_VAL_INIT_CSUM,
11259                         GETOPT_VAL_INIT_EXTENT, GETOPT_VAL_CHECK_CSUM,
11260                         GETOPT_VAL_READONLY, GETOPT_VAL_CHUNK_TREE,
11261                         GETOPT_VAL_MODE, GETOPT_VAL_CLEAR_SPACE_CACHE };
11262                 static const struct option long_options[] = {
11263                         { "super", required_argument, NULL, 's' },
11264                         { "repair", no_argument, NULL, GETOPT_VAL_REPAIR },
11265                         { "readonly", no_argument, NULL, GETOPT_VAL_READONLY },
11266                         { "init-csum-tree", no_argument, NULL,
11267                                 GETOPT_VAL_INIT_CSUM },
11268                         { "init-extent-tree", no_argument, NULL,
11269                                 GETOPT_VAL_INIT_EXTENT },
11270                         { "check-data-csum", no_argument, NULL,
11271                                 GETOPT_VAL_CHECK_CSUM },
11272                         { "backup", no_argument, NULL, 'b' },
11273                         { "subvol-extents", required_argument, NULL, 'E' },
11274                         { "qgroup-report", no_argument, NULL, 'Q' },
11275                         { "tree-root", required_argument, NULL, 'r' },
11276                         { "chunk-root", required_argument, NULL,
11277                                 GETOPT_VAL_CHUNK_TREE },
11278                         { "progress", no_argument, NULL, 'p' },
11279                         { "mode", required_argument, NULL,
11280                                 GETOPT_VAL_MODE },
11281                         { "clear-space-cache", required_argument, NULL,
11282                                 GETOPT_VAL_CLEAR_SPACE_CACHE},
11283                         { NULL, 0, NULL, 0}
11284                 };
11285
11286                 c = getopt_long(argc, argv, "as:br:p", long_options, NULL);
11287                 if (c < 0)
11288                         break;
11289                 switch(c) {
11290                         case 'a': /* ignored */ break;
11291                         case 'b':
11292                                 ctree_flags |= OPEN_CTREE_BACKUP_ROOT;
11293                                 break;
11294                         case 's':
11295                                 num = arg_strtou64(optarg);
11296                                 if (num >= BTRFS_SUPER_MIRROR_MAX) {
11297                                         error(
11298                                         "super mirror should be less than %d",
11299                                                 BTRFS_SUPER_MIRROR_MAX);
11300                                         exit(1);
11301                                 }
11302                                 bytenr = btrfs_sb_offset(((int)num));
11303                                 printf("using SB copy %llu, bytenr %llu\n", num,
11304                                        (unsigned long long)bytenr);
11305                                 break;
11306                         case 'Q':
11307                                 qgroup_report = 1;
11308                                 break;
11309                         case 'E':
11310                                 subvolid = arg_strtou64(optarg);
11311                                 break;
11312                         case 'r':
11313                                 tree_root_bytenr = arg_strtou64(optarg);
11314                                 break;
11315                         case GETOPT_VAL_CHUNK_TREE:
11316                                 chunk_root_bytenr = arg_strtou64(optarg);
11317                                 break;
11318                         case 'p':
11319                                 ctx.progress_enabled = true;
11320                                 break;
11321                         case '?':
11322                         case 'h':
11323                                 usage(cmd_check_usage);
11324                         case GETOPT_VAL_REPAIR:
11325                                 printf("enabling repair mode\n");
11326                                 repair = 1;
11327                                 ctree_flags |= OPEN_CTREE_WRITES;
11328                                 break;
11329                         case GETOPT_VAL_READONLY:
11330                                 readonly = 1;
11331                                 break;
11332                         case GETOPT_VAL_INIT_CSUM:
11333                                 printf("Creating a new CRC tree\n");
11334                                 init_csum_tree = 1;
11335                                 repair = 1;
11336                                 ctree_flags |= OPEN_CTREE_WRITES;
11337                                 break;
11338                         case GETOPT_VAL_INIT_EXTENT:
11339                                 init_extent_tree = 1;
11340                                 ctree_flags |= (OPEN_CTREE_WRITES |
11341                                                 OPEN_CTREE_NO_BLOCK_GROUPS);
11342                                 repair = 1;
11343                                 break;
11344                         case GETOPT_VAL_CHECK_CSUM:
11345                                 check_data_csum = 1;
11346                                 break;
11347                         case GETOPT_VAL_MODE:
11348                                 check_mode = parse_check_mode(optarg);
11349                                 if (check_mode == CHECK_MODE_UNKNOWN) {
11350                                         error("unknown mode: %s", optarg);
11351                                         exit(1);
11352                                 }
11353                                 break;
11354                         case GETOPT_VAL_CLEAR_SPACE_CACHE:
11355                                 if (strcmp(optarg, "v1") != 0) {
11356                                         error(
11357                         "only v1 support implmented, unrecognized value %s",
11358                         optarg);
11359                                         exit(1);
11360                                 }
11361                                 clear_space_cache = 1;
11362                                 ctree_flags |= OPEN_CTREE_WRITES;
11363                                 break;
11364                 }
11365         }
11366
11367         if (check_argc_exact(argc - optind, 1))
11368                 usage(cmd_check_usage);
11369
11370         if (ctx.progress_enabled) {
11371                 ctx.tp = TASK_NOTHING;
11372                 ctx.info = task_init(print_status_check, print_status_return, &ctx);
11373         }
11374
11375         /* This check is the only reason for --readonly to exist */
11376         if (readonly && repair) {
11377                 error("repair options are not compatible with --readonly");
11378                 exit(1);
11379         }
11380
11381         /*
11382          * Not supported yet
11383          */
11384         if (repair && check_mode == CHECK_MODE_LOWMEM) {
11385                 error("low memory mode doesn't support repair yet");
11386                 exit(1);
11387         }
11388
11389         radix_tree_init();
11390         cache_tree_init(&root_cache);
11391
11392         if((ret = check_mounted(argv[optind])) < 0) {
11393                 error("could not check mount status: %s", strerror(-ret));
11394                 goto err_out;
11395         } else if(ret) {
11396                 error("%s is currently mounted, aborting", argv[optind]);
11397                 ret = -EBUSY;
11398                 goto err_out;
11399         }
11400
11401         /* only allow partial opening under repair mode */
11402         if (repair)
11403                 ctree_flags |= OPEN_CTREE_PARTIAL;
11404
11405         info = open_ctree_fs_info(argv[optind], bytenr, tree_root_bytenr,
11406                                   chunk_root_bytenr, ctree_flags);
11407         if (!info) {
11408                 error("cannot open file system");
11409                 ret = -EIO;
11410                 goto err_out;
11411         }
11412
11413         global_info = info;
11414         root = info->fs_root;
11415         if (clear_space_cache) {
11416                 if (btrfs_fs_compat_ro(info,
11417                                 BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE)) {
11418                         error(
11419                         "free space cache v2 detected, clearing not implemented");
11420                         ret = 1;
11421                         goto close_out;
11422                 }
11423                 printf("Clearing free space cache\n");
11424                 ret = clear_free_space_cache(info);
11425                 if (ret) {
11426                         error("failed to clear free space cache");
11427                         ret = 1;
11428                 } else {
11429                         printf("Free space cache cleared\n");
11430                 }
11431                 goto close_out;
11432         }
11433
11434         /*
11435          * repair mode will force us to commit transaction which
11436          * will make us fail to load log tree when mounting.
11437          */
11438         if (repair && btrfs_super_log_root(info->super_copy)) {
11439                 ret = ask_user("repair mode will force to clear out log tree, are you sure?");
11440                 if (!ret) {
11441                         ret = 1;
11442                         goto close_out;
11443                 }
11444                 ret = zero_log_tree(root);
11445                 if (ret) {
11446                         error("failed to zero log tree: %d", ret);
11447                         goto close_out;
11448                 }
11449         }
11450
11451         uuid_unparse(info->super_copy->fsid, uuidbuf);
11452         if (qgroup_report) {
11453                 printf("Print quota groups for %s\nUUID: %s\n", argv[optind],
11454                        uuidbuf);
11455                 ret = qgroup_verify_all(info);
11456                 if (ret == 0)
11457                         report_qgroups(1);
11458                 goto close_out;
11459         }
11460         if (subvolid) {
11461                 printf("Print extent state for subvolume %llu on %s\nUUID: %s\n",
11462                        subvolid, argv[optind], uuidbuf);
11463                 ret = print_extent_state(info, subvolid);
11464                 goto close_out;
11465         }
11466         printf("Checking filesystem on %s\nUUID: %s\n", argv[optind], uuidbuf);
11467
11468         if (!extent_buffer_uptodate(info->tree_root->node) ||
11469             !extent_buffer_uptodate(info->dev_root->node) ||
11470             !extent_buffer_uptodate(info->chunk_root->node)) {
11471                 error("critical roots corrupted, unable to check the filesystem");
11472                 ret = -EIO;
11473                 goto close_out;
11474         }
11475
11476         if (init_extent_tree || init_csum_tree) {
11477                 struct btrfs_trans_handle *trans;
11478
11479                 trans = btrfs_start_transaction(info->extent_root, 0);
11480                 if (IS_ERR(trans)) {
11481                         error("error starting transaction");
11482                         ret = PTR_ERR(trans);
11483                         goto close_out;
11484                 }
11485
11486                 if (init_extent_tree) {
11487                         printf("Creating a new extent tree\n");
11488                         ret = reinit_extent_tree(trans, info);
11489                         if (ret)
11490                                 goto close_out;
11491                 }
11492
11493                 if (init_csum_tree) {
11494                         printf("Reinitialize checksum tree\n");
11495                         ret = btrfs_fsck_reinit_root(trans, info->csum_root, 0);
11496                         if (ret) {
11497                                 error("checksum tree initialization failed: %d",
11498                                                 ret);
11499                                 ret = -EIO;
11500                                 goto close_out;
11501                         }
11502
11503                         ret = fill_csum_tree(trans, info->csum_root,
11504                                              init_extent_tree);
11505                         if (ret) {
11506                                 error("checksum tree refilling failed: %d", ret);
11507                                 return -EIO;
11508                         }
11509                 }
11510                 /*
11511                  * Ok now we commit and run the normal fsck, which will add
11512                  * extent entries for all of the items it finds.
11513                  */
11514                 ret = btrfs_commit_transaction(trans, info->extent_root);
11515                 if (ret)
11516                         goto close_out;
11517         }
11518         if (!extent_buffer_uptodate(info->extent_root->node)) {
11519                 error("critical: extent_root, unable to check the filesystem");
11520                 ret = -EIO;
11521                 goto close_out;
11522         }
11523         if (!extent_buffer_uptodate(info->csum_root->node)) {
11524                 error("critical: csum_root, unable to check the filesystem");
11525                 ret = -EIO;
11526                 goto close_out;
11527         }
11528
11529         if (!ctx.progress_enabled)
11530                 printf("checking extents");
11531         if (check_mode == CHECK_MODE_LOWMEM)
11532                 ret = check_chunks_and_extents_v2(root);
11533         else
11534                 ret = check_chunks_and_extents(root);
11535         if (ret)
11536                 printf("Errors found in extent allocation tree or chunk allocation");
11537
11538         ret = repair_root_items(info);
11539         if (ret < 0)
11540                 goto close_out;
11541         if (repair) {
11542                 fprintf(stderr, "Fixed %d roots.\n", ret);
11543                 ret = 0;
11544         } else if (ret > 0) {
11545                 fprintf(stderr,
11546                        "Found %d roots with an outdated root item.\n",
11547                        ret);
11548                 fprintf(stderr,
11549                         "Please run a filesystem check with the option --repair to fix them.\n");
11550                 ret = 1;
11551                 goto close_out;
11552         }
11553
11554         if (!ctx.progress_enabled) {
11555                 if (btrfs_fs_compat_ro(info, BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE))
11556                         fprintf(stderr, "checking free space tree\n");
11557                 else
11558                         fprintf(stderr, "checking free space cache\n");
11559         }
11560         ret = check_space_cache(root);
11561         if (ret)
11562                 goto out;
11563
11564         /*
11565          * We used to have to have these hole extents in between our real
11566          * extents so if we don't have this flag set we need to make sure there
11567          * are no gaps in the file extents for inodes, otherwise we can just
11568          * ignore it when this happens.
11569          */
11570         no_holes = btrfs_fs_incompat(root->fs_info,
11571                                      BTRFS_FEATURE_INCOMPAT_NO_HOLES);
11572         if (!ctx.progress_enabled)
11573                 fprintf(stderr, "checking fs roots\n");
11574         ret = check_fs_roots(root, &root_cache);
11575         if (ret)
11576                 goto out;
11577
11578         fprintf(stderr, "checking csums\n");
11579         ret = check_csums(root);
11580         if (ret)
11581                 goto out;
11582
11583         fprintf(stderr, "checking root refs\n");
11584         ret = check_root_refs(root, &root_cache);
11585         if (ret)
11586                 goto out;
11587
11588         while (repair && !list_empty(&root->fs_info->recow_ebs)) {
11589                 struct extent_buffer *eb;
11590
11591                 eb = list_first_entry(&root->fs_info->recow_ebs,
11592                                       struct extent_buffer, recow);
11593                 list_del_init(&eb->recow);
11594                 ret = recow_extent_buffer(root, eb);
11595                 if (ret)
11596                         break;
11597         }
11598
11599         while (!list_empty(&delete_items)) {
11600                 struct bad_item *bad;
11601
11602                 bad = list_first_entry(&delete_items, struct bad_item, list);
11603                 list_del_init(&bad->list);
11604                 if (repair)
11605                         ret = delete_bad_item(root, bad);
11606                 free(bad);
11607         }
11608
11609         if (info->quota_enabled) {
11610                 int err;
11611                 fprintf(stderr, "checking quota groups\n");
11612                 err = qgroup_verify_all(info);
11613                 if (err)
11614                         goto out;
11615                 report_qgroups(0);
11616                 err = repair_qgroups(info, &qgroups_repaired);
11617                 if (err)
11618                         goto out;
11619         }
11620
11621         if (!list_empty(&root->fs_info->recow_ebs)) {
11622                 error("transid errors in file system");
11623                 ret = 1;
11624         }
11625 out:
11626         /* Don't override original ret */
11627         if (!ret && qgroups_repaired)
11628                 ret = qgroups_repaired;
11629
11630         if (found_old_backref) { /*
11631                  * there was a disk format change when mixed
11632                  * backref was in testing tree. The old format
11633                  * existed about one week.
11634                  */
11635                 printf("\n * Found old mixed backref format. "
11636                        "The old format is not supported! *"
11637                        "\n * Please mount the FS in readonly mode, "
11638                        "backup data and re-format the FS. *\n\n");
11639                 ret = 1;
11640         }
11641         printf("found %llu bytes used err is %d\n",
11642                (unsigned long long)bytes_used, ret);
11643         printf("total csum bytes: %llu\n",(unsigned long long)total_csum_bytes);
11644         printf("total tree bytes: %llu\n",
11645                (unsigned long long)total_btree_bytes);
11646         printf("total fs tree bytes: %llu\n",
11647                (unsigned long long)total_fs_tree_bytes);
11648         printf("total extent tree bytes: %llu\n",
11649                (unsigned long long)total_extent_tree_bytes);
11650         printf("btree space waste bytes: %llu\n",
11651                (unsigned long long)btree_space_waste);
11652         printf("file data blocks allocated: %llu\n referenced %llu\n",
11653                 (unsigned long long)data_bytes_allocated,
11654                 (unsigned long long)data_bytes_referenced);
11655
11656         free_qgroup_counts();
11657         free_root_recs_tree(&root_cache);
11658 close_out:
11659         close_ctree(root);
11660 err_out:
11661         if (ctx.progress_enabled)
11662                 task_deinit(ctx.info);
11663
11664         return ret;
11665 }