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         path = btrfs_alloc_path();
5770         if (!path) {
5771                 fprintf(stderr, "Error allocating path\n");
5772                 return -ENOMEM;
5773         }
5774
5775         key.objectid = bytenr;
5776         key.type = BTRFS_EXTENT_ITEM_KEY;
5777         key.offset = (u64)-1;
5778
5779 again:
5780         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
5781                                 0, 0);
5782         if (ret < 0) {
5783                 fprintf(stderr, "Error looking up extent record %d\n", ret);
5784                 btrfs_free_path(path);
5785                 return ret;
5786         } else if (ret) {
5787                 if (path->slots[0] > 0) {
5788                         path->slots[0]--;
5789                 } else {
5790                         ret = btrfs_prev_leaf(root, path);
5791                         if (ret < 0) {
5792                                 goto out;
5793                         } else if (ret > 0) {
5794                                 ret = 0;
5795                                 goto out;
5796                         }
5797                 }
5798         }
5799
5800         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
5801
5802         /*
5803          * Block group items come before extent items if they have the same
5804          * bytenr, so walk back one more just in case.  Dear future traveller,
5805          * first congrats on mastering time travel.  Now if it's not too much
5806          * trouble could you go back to 2006 and tell Chris to make the
5807          * BLOCK_GROUP_ITEM_KEY (and BTRFS_*_REF_KEY) lower than the
5808          * EXTENT_ITEM_KEY please?
5809          */
5810         while (key.type > BTRFS_EXTENT_ITEM_KEY) {
5811                 if (path->slots[0] > 0) {
5812                         path->slots[0]--;
5813                 } else {
5814                         ret = btrfs_prev_leaf(root, path);
5815                         if (ret < 0) {
5816                                 goto out;
5817                         } else if (ret > 0) {
5818                                 ret = 0;
5819                                 goto out;
5820                         }
5821                 }
5822                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
5823         }
5824
5825         while (num_bytes) {
5826                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5827                         ret = btrfs_next_leaf(root, path);
5828                         if (ret < 0) {
5829                                 fprintf(stderr, "Error going to next leaf "
5830                                         "%d\n", ret);
5831                                 btrfs_free_path(path);
5832                                 return ret;
5833                         } else if (ret) {
5834                                 break;
5835                         }
5836                 }
5837                 leaf = path->nodes[0];
5838                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5839                 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
5840                         path->slots[0]++;
5841                         continue;
5842                 }
5843                 if (key.objectid + key.offset < bytenr) {
5844                         path->slots[0]++;
5845                         continue;
5846                 }
5847                 if (key.objectid > bytenr + num_bytes)
5848                         break;
5849
5850                 if (key.objectid == bytenr) {
5851                         if (key.offset >= num_bytes) {
5852                                 num_bytes = 0;
5853                                 break;
5854                         }
5855                         num_bytes -= key.offset;
5856                         bytenr += key.offset;
5857                 } else if (key.objectid < bytenr) {
5858                         if (key.objectid + key.offset >= bytenr + num_bytes) {
5859                                 num_bytes = 0;
5860                                 break;
5861                         }
5862                         num_bytes = (bytenr + num_bytes) -
5863                                 (key.objectid + key.offset);
5864                         bytenr = key.objectid + key.offset;
5865                 } else {
5866                         if (key.objectid + key.offset < bytenr + num_bytes) {
5867                                 u64 new_start = key.objectid + key.offset;
5868                                 u64 new_bytes = bytenr + num_bytes - new_start;
5869
5870                                 /*
5871                                  * Weird case, the extent is in the middle of
5872                                  * our range, we'll have to search one side
5873                                  * and then the other.  Not sure if this happens
5874                                  * in real life, but no harm in coding it up
5875                                  * anyway just in case.
5876                                  */
5877                                 btrfs_release_path(path);
5878                                 ret = check_extent_exists(root, new_start,
5879                                                           new_bytes);
5880                                 if (ret) {
5881                                         fprintf(stderr, "Right section didn't "
5882                                                 "have a record\n");
5883                                         break;
5884                                 }
5885                                 num_bytes = key.objectid - bytenr;
5886                                 goto again;
5887                         }
5888                         num_bytes = key.objectid - bytenr;
5889                 }
5890                 path->slots[0]++;
5891         }
5892         ret = 0;
5893
5894 out:
5895         if (num_bytes && !ret) {
5896                 fprintf(stderr, "There are no extents for csum range "
5897                         "%Lu-%Lu\n", bytenr, bytenr+num_bytes);
5898                 ret = 1;
5899         }
5900
5901         btrfs_free_path(path);
5902         return ret;
5903 }
5904
5905 static int check_csums(struct btrfs_root *root)
5906 {
5907         struct btrfs_path *path;
5908         struct extent_buffer *leaf;
5909         struct btrfs_key key;
5910         u64 offset = 0, num_bytes = 0;
5911         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
5912         int errors = 0;
5913         int ret;
5914         u64 data_len;
5915         unsigned long leaf_offset;
5916
5917         root = root->fs_info->csum_root;
5918         if (!extent_buffer_uptodate(root->node)) {
5919                 fprintf(stderr, "No valid csum tree found\n");
5920                 return -ENOENT;
5921         }
5922
5923         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
5924         key.type = BTRFS_EXTENT_CSUM_KEY;
5925         key.offset = 0;
5926
5927         path = btrfs_alloc_path();
5928         if (!path)
5929                 return -ENOMEM;
5930
5931         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5932         if (ret < 0) {
5933                 fprintf(stderr, "Error searching csum tree %d\n", ret);
5934                 btrfs_free_path(path);
5935                 return ret;
5936         }
5937
5938         if (ret > 0 && path->slots[0])
5939                 path->slots[0]--;
5940         ret = 0;
5941
5942         while (1) {
5943                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5944                         ret = btrfs_next_leaf(root, path);
5945                         if (ret < 0) {
5946                                 fprintf(stderr, "Error going to next leaf "
5947                                         "%d\n", ret);
5948                                 break;
5949                         }
5950                         if (ret)
5951                                 break;
5952                 }
5953                 leaf = path->nodes[0];
5954
5955                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5956                 if (key.type != BTRFS_EXTENT_CSUM_KEY) {
5957                         path->slots[0]++;
5958                         continue;
5959                 }
5960
5961                 data_len = (btrfs_item_size_nr(leaf, path->slots[0]) /
5962                               csum_size) * root->sectorsize;
5963                 if (!check_data_csum)
5964                         goto skip_csum_check;
5965                 leaf_offset = btrfs_item_ptr_offset(leaf, path->slots[0]);
5966                 ret = check_extent_csums(root, key.offset, data_len,
5967                                          leaf_offset, leaf);
5968                 if (ret)
5969                         break;
5970 skip_csum_check:
5971                 if (!num_bytes) {
5972                         offset = key.offset;
5973                 } else if (key.offset != offset + num_bytes) {
5974                         ret = check_extent_exists(root, offset, num_bytes);
5975                         if (ret) {
5976                                 fprintf(stderr, "Csum exists for %Lu-%Lu but "
5977                                         "there is no extent record\n",
5978                                         offset, offset+num_bytes);
5979                                 errors++;
5980                         }
5981                         offset = key.offset;
5982                         num_bytes = 0;
5983                 }
5984                 num_bytes += data_len;
5985                 path->slots[0]++;
5986         }
5987
5988         btrfs_free_path(path);
5989         return errors;
5990 }
5991
5992 static int is_dropped_key(struct btrfs_key *key,
5993                           struct btrfs_key *drop_key) {
5994         if (key->objectid < drop_key->objectid)
5995                 return 1;
5996         else if (key->objectid == drop_key->objectid) {
5997                 if (key->type < drop_key->type)
5998                         return 1;
5999                 else if (key->type == drop_key->type) {
6000                         if (key->offset < drop_key->offset)
6001                                 return 1;
6002                 }
6003         }
6004         return 0;
6005 }
6006
6007 /*
6008  * Here are the rules for FULL_BACKREF.
6009  *
6010  * 1) If BTRFS_HEADER_FLAG_RELOC is set then we have FULL_BACKREF set.
6011  * 2) If btrfs_header_owner(buf) no longer points to buf then we have
6012  *      FULL_BACKREF set.
6013  * 3) We cowed the block walking down a reloc tree.  This is impossible to tell
6014  *    if it happened after the relocation occurred since we'll have dropped the
6015  *    reloc root, so it's entirely possible to have FULL_BACKREF set on buf and
6016  *    have no real way to know for sure.
6017  *
6018  * We process the blocks one root at a time, and we start from the lowest root
6019  * objectid and go to the highest.  So we can just lookup the owner backref for
6020  * the record and if we don't find it then we know it doesn't exist and we have
6021  * a FULL BACKREF.
6022  *
6023  * FIXME: if we ever start reclaiming root objectid's then we need to fix this
6024  * assumption and simply indicate that we _think_ that the FULL BACKREF needs to
6025  * be set or not and then we can check later once we've gathered all the refs.
6026  */
6027 static int calc_extent_flag(struct btrfs_root *root,
6028                            struct cache_tree *extent_cache,
6029                            struct extent_buffer *buf,
6030                            struct root_item_record *ri,
6031                            u64 *flags)
6032 {
6033         struct extent_record *rec;
6034         struct cache_extent *cache;
6035         struct tree_backref *tback;
6036         u64 owner = 0;
6037
6038         cache = lookup_cache_extent(extent_cache, buf->start, 1);
6039         /* we have added this extent before */
6040         if (!cache)
6041                 return -ENOENT;
6042
6043         rec = container_of(cache, struct extent_record, cache);
6044
6045         /*
6046          * Except file/reloc tree, we can not have
6047          * FULL BACKREF MODE
6048          */
6049         if (ri->objectid < BTRFS_FIRST_FREE_OBJECTID)
6050                 goto normal;
6051         /*
6052          * root node
6053          */
6054         if (buf->start == ri->bytenr)
6055                 goto normal;
6056
6057         if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
6058                 goto full_backref;
6059
6060         owner = btrfs_header_owner(buf);
6061         if (owner == ri->objectid)
6062                 goto normal;
6063
6064         tback = find_tree_backref(rec, 0, owner);
6065         if (!tback)
6066                 goto full_backref;
6067 normal:
6068         *flags = 0;
6069         if (rec->flag_block_full_backref != FLAG_UNSET &&
6070             rec->flag_block_full_backref != 0)
6071                 rec->bad_full_backref = 1;
6072         return 0;
6073 full_backref:
6074         *flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6075         if (rec->flag_block_full_backref != FLAG_UNSET &&
6076             rec->flag_block_full_backref != 1)
6077                 rec->bad_full_backref = 1;
6078         return 0;
6079 }
6080
6081 static void report_mismatch_key_root(u8 key_type, u64 rootid)
6082 {
6083         fprintf(stderr, "Invalid key type(");
6084         print_key_type(stderr, 0, key_type);
6085         fprintf(stderr, ") found in root(");
6086         print_objectid(stderr, rootid, 0);
6087         fprintf(stderr, ")\n");
6088 }
6089
6090 /*
6091  * Check if the key is valid with its extent buffer.
6092  *
6093  * This is a early check in case invalid key exists in a extent buffer
6094  * This is not comprehensive yet, but should prevent wrong key/item passed
6095  * further
6096  */
6097 static int check_type_with_root(u64 rootid, u8 key_type)
6098 {
6099         switch (key_type) {
6100         /* Only valid in chunk tree */
6101         case BTRFS_DEV_ITEM_KEY:
6102         case BTRFS_CHUNK_ITEM_KEY:
6103                 if (rootid != BTRFS_CHUNK_TREE_OBJECTID)
6104                         goto err;
6105                 break;
6106         /* valid in csum and log tree */
6107         case BTRFS_CSUM_TREE_OBJECTID:
6108                 if (!(rootid == BTRFS_TREE_LOG_OBJECTID ||
6109                       is_fstree(rootid)))
6110                         goto err;
6111                 break;
6112         case BTRFS_EXTENT_ITEM_KEY:
6113         case BTRFS_METADATA_ITEM_KEY:
6114         case BTRFS_BLOCK_GROUP_ITEM_KEY:
6115                 if (rootid != BTRFS_EXTENT_TREE_OBJECTID)
6116                         goto err;
6117                 break;
6118         case BTRFS_ROOT_ITEM_KEY:
6119                 if (rootid != BTRFS_ROOT_TREE_OBJECTID)
6120                         goto err;
6121                 break;
6122         case BTRFS_DEV_EXTENT_KEY:
6123                 if (rootid != BTRFS_DEV_TREE_OBJECTID)
6124                         goto err;
6125                 break;
6126         }
6127         return 0;
6128 err:
6129         report_mismatch_key_root(key_type, rootid);
6130         return -EINVAL;
6131 }
6132
6133 static int run_next_block(struct btrfs_root *root,
6134                           struct block_info *bits,
6135                           int bits_nr,
6136                           u64 *last,
6137                           struct cache_tree *pending,
6138                           struct cache_tree *seen,
6139                           struct cache_tree *reada,
6140                           struct cache_tree *nodes,
6141                           struct cache_tree *extent_cache,
6142                           struct cache_tree *chunk_cache,
6143                           struct rb_root *dev_cache,
6144                           struct block_group_tree *block_group_cache,
6145                           struct device_extent_tree *dev_extent_cache,
6146                           struct root_item_record *ri)
6147 {
6148         struct extent_buffer *buf;
6149         struct extent_record *rec = NULL;
6150         u64 bytenr;
6151         u32 size;
6152         u64 parent;
6153         u64 owner;
6154         u64 flags;
6155         u64 ptr;
6156         u64 gen = 0;
6157         int ret = 0;
6158         int i;
6159         int nritems;
6160         struct btrfs_key key;
6161         struct cache_extent *cache;
6162         int reada_bits;
6163
6164         nritems = pick_next_pending(pending, reada, nodes, *last, bits,
6165                                     bits_nr, &reada_bits);
6166         if (nritems == 0)
6167                 return 1;
6168
6169         if (!reada_bits) {
6170                 for(i = 0; i < nritems; i++) {
6171                         ret = add_cache_extent(reada, bits[i].start,
6172                                                bits[i].size);
6173                         if (ret == -EEXIST)
6174                                 continue;
6175
6176                         /* fixme, get the parent transid */
6177                         readahead_tree_block(root, bits[i].start,
6178                                              bits[i].size, 0);
6179                 }
6180         }
6181         *last = bits[0].start;
6182         bytenr = bits[0].start;
6183         size = bits[0].size;
6184
6185         cache = lookup_cache_extent(pending, bytenr, size);
6186         if (cache) {
6187                 remove_cache_extent(pending, cache);
6188                 free(cache);
6189         }
6190         cache = lookup_cache_extent(reada, bytenr, size);
6191         if (cache) {
6192                 remove_cache_extent(reada, cache);
6193                 free(cache);
6194         }
6195         cache = lookup_cache_extent(nodes, bytenr, size);
6196         if (cache) {
6197                 remove_cache_extent(nodes, cache);
6198                 free(cache);
6199         }
6200         cache = lookup_cache_extent(extent_cache, bytenr, size);
6201         if (cache) {
6202                 rec = container_of(cache, struct extent_record, cache);
6203                 gen = rec->parent_generation;
6204         }
6205
6206         /* fixme, get the real parent transid */
6207         buf = read_tree_block(root, bytenr, size, gen);
6208         if (!extent_buffer_uptodate(buf)) {
6209                 record_bad_block_io(root->fs_info,
6210                                     extent_cache, bytenr, size);
6211                 goto out;
6212         }
6213
6214         nritems = btrfs_header_nritems(buf);
6215
6216         flags = 0;
6217         if (!init_extent_tree) {
6218                 ret = btrfs_lookup_extent_info(NULL, root, bytenr,
6219                                        btrfs_header_level(buf), 1, NULL,
6220                                        &flags);
6221                 if (ret < 0) {
6222                         ret = calc_extent_flag(root, extent_cache, buf, ri, &flags);
6223                         if (ret < 0) {
6224                                 fprintf(stderr, "Couldn't calc extent flags\n");
6225                                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6226                         }
6227                 }
6228         } else {
6229                 flags = 0;
6230                 ret = calc_extent_flag(root, extent_cache, buf, ri, &flags);
6231                 if (ret < 0) {
6232                         fprintf(stderr, "Couldn't calc extent flags\n");
6233                         flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6234                 }
6235         }
6236
6237         if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6238                 if (ri != NULL &&
6239                     ri->objectid != BTRFS_TREE_RELOC_OBJECTID &&
6240                     ri->objectid == btrfs_header_owner(buf)) {
6241                         /*
6242                          * Ok we got to this block from it's original owner and
6243                          * we have FULL_BACKREF set.  Relocation can leave
6244                          * converted blocks over so this is altogether possible,
6245                          * however it's not possible if the generation > the
6246                          * last snapshot, so check for this case.
6247                          */
6248                         if (!btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC) &&
6249                             btrfs_header_generation(buf) > ri->last_snapshot) {
6250                                 flags &= ~BTRFS_BLOCK_FLAG_FULL_BACKREF;
6251                                 rec->bad_full_backref = 1;
6252                         }
6253                 }
6254         } else {
6255                 if (ri != NULL &&
6256                     (ri->objectid == BTRFS_TREE_RELOC_OBJECTID ||
6257                      btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) {
6258                         flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6259                         rec->bad_full_backref = 1;
6260                 }
6261         }
6262
6263         if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6264                 rec->flag_block_full_backref = 1;
6265                 parent = bytenr;
6266                 owner = 0;
6267         } else {
6268                 rec->flag_block_full_backref = 0;
6269                 parent = 0;
6270                 owner = btrfs_header_owner(buf);
6271         }
6272
6273         ret = check_block(root, extent_cache, buf, flags);
6274         if (ret)
6275                 goto out;
6276
6277         if (btrfs_is_leaf(buf)) {
6278                 btree_space_waste += btrfs_leaf_free_space(root, buf);
6279                 for (i = 0; i < nritems; i++) {
6280                         struct btrfs_file_extent_item *fi;
6281                         btrfs_item_key_to_cpu(buf, &key, i);
6282                         /*
6283                          * Check key type against the leaf owner.
6284                          * Could filter quite a lot of early error if
6285                          * owner is correct
6286                          */
6287                         if (check_type_with_root(btrfs_header_owner(buf),
6288                                                  key.type)) {
6289                                 fprintf(stderr, "ignoring invalid key\n");
6290                                 continue;
6291                         }
6292                         if (key.type == BTRFS_EXTENT_ITEM_KEY) {
6293                                 process_extent_item(root, extent_cache, buf,
6294                                                     i);
6295                                 continue;
6296                         }
6297                         if (key.type == BTRFS_METADATA_ITEM_KEY) {
6298                                 process_extent_item(root, extent_cache, buf,
6299                                                     i);
6300                                 continue;
6301                         }
6302                         if (key.type == BTRFS_EXTENT_CSUM_KEY) {
6303                                 total_csum_bytes +=
6304                                         btrfs_item_size_nr(buf, i);
6305                                 continue;
6306                         }
6307                         if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6308                                 process_chunk_item(chunk_cache, &key, buf, i);
6309                                 continue;
6310                         }
6311                         if (key.type == BTRFS_DEV_ITEM_KEY) {
6312                                 process_device_item(dev_cache, &key, buf, i);
6313                                 continue;
6314                         }
6315                         if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
6316                                 process_block_group_item(block_group_cache,
6317                                         &key, buf, i);
6318                                 continue;
6319                         }
6320                         if (key.type == BTRFS_DEV_EXTENT_KEY) {
6321                                 process_device_extent_item(dev_extent_cache,
6322                                         &key, buf, i);
6323                                 continue;
6324
6325                         }
6326                         if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
6327 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6328                                 process_extent_ref_v0(extent_cache, buf, i);
6329 #else
6330                                 BUG();
6331 #endif
6332                                 continue;
6333                         }
6334
6335                         if (key.type == BTRFS_TREE_BLOCK_REF_KEY) {
6336                                 ret = add_tree_backref(extent_cache,
6337                                                 key.objectid, 0, key.offset, 0);
6338                                 if (ret < 0)
6339                                         error("add_tree_backref failed: %s",
6340                                               strerror(-ret));
6341                                 continue;
6342                         }
6343                         if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
6344                                 ret = add_tree_backref(extent_cache,
6345                                                 key.objectid, key.offset, 0, 0);
6346                                 if (ret < 0)
6347                                         error("add_tree_backref failed: %s",
6348                                               strerror(-ret));
6349                                 continue;
6350                         }
6351                         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
6352                                 struct btrfs_extent_data_ref *ref;
6353                                 ref = btrfs_item_ptr(buf, i,
6354                                                 struct btrfs_extent_data_ref);
6355                                 add_data_backref(extent_cache,
6356                                         key.objectid, 0,
6357                                         btrfs_extent_data_ref_root(buf, ref),
6358                                         btrfs_extent_data_ref_objectid(buf,
6359                                                                        ref),
6360                                         btrfs_extent_data_ref_offset(buf, ref),
6361                                         btrfs_extent_data_ref_count(buf, ref),
6362                                         0, root->sectorsize);
6363                                 continue;
6364                         }
6365                         if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
6366                                 struct btrfs_shared_data_ref *ref;
6367                                 ref = btrfs_item_ptr(buf, i,
6368                                                 struct btrfs_shared_data_ref);
6369                                 add_data_backref(extent_cache,
6370                                         key.objectid, key.offset, 0, 0, 0,
6371                                         btrfs_shared_data_ref_count(buf, ref),
6372                                         0, root->sectorsize);
6373                                 continue;
6374                         }
6375                         if (key.type == BTRFS_ORPHAN_ITEM_KEY) {
6376                                 struct bad_item *bad;
6377
6378                                 if (key.objectid == BTRFS_ORPHAN_OBJECTID)
6379                                         continue;
6380                                 if (!owner)
6381                                         continue;
6382                                 bad = malloc(sizeof(struct bad_item));
6383                                 if (!bad)
6384                                         continue;
6385                                 INIT_LIST_HEAD(&bad->list);
6386                                 memcpy(&bad->key, &key,
6387                                        sizeof(struct btrfs_key));
6388                                 bad->root_id = owner;
6389                                 list_add_tail(&bad->list, &delete_items);
6390                                 continue;
6391                         }
6392                         if (key.type != BTRFS_EXTENT_DATA_KEY)
6393                                 continue;
6394                         fi = btrfs_item_ptr(buf, i,
6395                                             struct btrfs_file_extent_item);
6396                         if (btrfs_file_extent_type(buf, fi) ==
6397                             BTRFS_FILE_EXTENT_INLINE)
6398                                 continue;
6399                         if (btrfs_file_extent_disk_bytenr(buf, fi) == 0)
6400                                 continue;
6401
6402                         data_bytes_allocated +=
6403                                 btrfs_file_extent_disk_num_bytes(buf, fi);
6404                         if (data_bytes_allocated < root->sectorsize) {
6405                                 abort();
6406                         }
6407                         data_bytes_referenced +=
6408                                 btrfs_file_extent_num_bytes(buf, fi);
6409                         add_data_backref(extent_cache,
6410                                 btrfs_file_extent_disk_bytenr(buf, fi),
6411                                 parent, owner, key.objectid, key.offset -
6412                                 btrfs_file_extent_offset(buf, fi), 1, 1,
6413                                 btrfs_file_extent_disk_num_bytes(buf, fi));
6414                 }
6415         } else {
6416                 int level;
6417                 struct btrfs_key first_key;
6418
6419                 first_key.objectid = 0;
6420
6421                 if (nritems > 0)
6422                         btrfs_item_key_to_cpu(buf, &first_key, 0);
6423                 level = btrfs_header_level(buf);
6424                 for (i = 0; i < nritems; i++) {
6425                         struct extent_record tmpl;
6426
6427                         ptr = btrfs_node_blockptr(buf, i);
6428                         size = root->nodesize;
6429                         btrfs_node_key_to_cpu(buf, &key, i);
6430                         if (ri != NULL) {
6431                                 if ((level == ri->drop_level)
6432                                     && is_dropped_key(&key, &ri->drop_key)) {
6433                                         continue;
6434                                 }
6435                         }
6436
6437                         memset(&tmpl, 0, sizeof(tmpl));
6438                         btrfs_cpu_key_to_disk(&tmpl.parent_key, &key);
6439                         tmpl.parent_generation = btrfs_node_ptr_generation(buf, i);
6440                         tmpl.start = ptr;
6441                         tmpl.nr = size;
6442                         tmpl.refs = 1;
6443                         tmpl.metadata = 1;
6444                         tmpl.max_size = size;
6445                         ret = add_extent_rec(extent_cache, &tmpl);
6446                         if (ret < 0)
6447                                 goto out;
6448
6449                         ret = add_tree_backref(extent_cache, ptr, parent,
6450                                         owner, 1);
6451                         if (ret < 0) {
6452                                 error("add_tree_backref failed: %s",
6453                                       strerror(-ret));
6454                                 continue;
6455                         }
6456
6457                         if (level > 1) {
6458                                 add_pending(nodes, seen, ptr, size);
6459                         } else {
6460                                 add_pending(pending, seen, ptr, size);
6461                         }
6462                 }
6463                 btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) -
6464                                       nritems) * sizeof(struct btrfs_key_ptr);
6465         }
6466         total_btree_bytes += buf->len;
6467         if (fs_root_objectid(btrfs_header_owner(buf)))
6468                 total_fs_tree_bytes += buf->len;
6469         if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID)
6470                 total_extent_tree_bytes += buf->len;
6471         if (!found_old_backref &&
6472             btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID &&
6473             btrfs_header_backref_rev(buf) == BTRFS_MIXED_BACKREF_REV &&
6474             !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
6475                 found_old_backref = 1;
6476 out:
6477         free_extent_buffer(buf);
6478         return ret;
6479 }
6480
6481 static int add_root_to_pending(struct extent_buffer *buf,
6482                                struct cache_tree *extent_cache,
6483                                struct cache_tree *pending,
6484                                struct cache_tree *seen,
6485                                struct cache_tree *nodes,
6486                                u64 objectid)
6487 {
6488         struct extent_record tmpl;
6489         int ret;
6490
6491         if (btrfs_header_level(buf) > 0)
6492                 add_pending(nodes, seen, buf->start, buf->len);
6493         else
6494                 add_pending(pending, seen, buf->start, buf->len);
6495
6496         memset(&tmpl, 0, sizeof(tmpl));
6497         tmpl.start = buf->start;
6498         tmpl.nr = buf->len;
6499         tmpl.is_root = 1;
6500         tmpl.refs = 1;
6501         tmpl.metadata = 1;
6502         tmpl.max_size = buf->len;
6503         add_extent_rec(extent_cache, &tmpl);
6504
6505         if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
6506             btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
6507                 ret = add_tree_backref(extent_cache, buf->start, buf->start,
6508                                 0, 1);
6509         else
6510                 ret = add_tree_backref(extent_cache, buf->start, 0, objectid,
6511                                 1);
6512         return ret;
6513 }
6514
6515 /* as we fix the tree, we might be deleting blocks that
6516  * we're tracking for repair.  This hook makes sure we
6517  * remove any backrefs for blocks as we are fixing them.
6518  */
6519 static int free_extent_hook(struct btrfs_trans_handle *trans,
6520                             struct btrfs_root *root,
6521                             u64 bytenr, u64 num_bytes, u64 parent,
6522                             u64 root_objectid, u64 owner, u64 offset,
6523                             int refs_to_drop)
6524 {
6525         struct extent_record *rec;
6526         struct cache_extent *cache;
6527         int is_data;
6528         struct cache_tree *extent_cache = root->fs_info->fsck_extent_cache;
6529
6530         is_data = owner >= BTRFS_FIRST_FREE_OBJECTID;
6531         cache = lookup_cache_extent(extent_cache, bytenr, num_bytes);
6532         if (!cache)
6533                 return 0;
6534
6535         rec = container_of(cache, struct extent_record, cache);
6536         if (is_data) {
6537                 struct data_backref *back;
6538                 back = find_data_backref(rec, parent, root_objectid, owner,
6539                                          offset, 1, bytenr, num_bytes);
6540                 if (!back)
6541                         goto out;
6542                 if (back->node.found_ref) {
6543                         back->found_ref -= refs_to_drop;
6544                         if (rec->refs)
6545                                 rec->refs -= refs_to_drop;
6546                 }
6547                 if (back->node.found_extent_tree) {
6548                         back->num_refs -= refs_to_drop;
6549                         if (rec->extent_item_refs)
6550                                 rec->extent_item_refs -= refs_to_drop;
6551                 }
6552                 if (back->found_ref == 0)
6553                         back->node.found_ref = 0;
6554                 if (back->num_refs == 0)
6555                         back->node.found_extent_tree = 0;
6556
6557                 if (!back->node.found_extent_tree && back->node.found_ref) {
6558                         list_del(&back->node.list);
6559                         free(back);
6560                 }
6561         } else {
6562                 struct tree_backref *back;
6563                 back = find_tree_backref(rec, parent, root_objectid);
6564                 if (!back)
6565                         goto out;
6566                 if (back->node.found_ref) {
6567                         if (rec->refs)
6568                                 rec->refs--;
6569                         back->node.found_ref = 0;
6570                 }
6571                 if (back->node.found_extent_tree) {
6572                         if (rec->extent_item_refs)
6573                                 rec->extent_item_refs--;
6574                         back->node.found_extent_tree = 0;
6575                 }
6576                 if (!back->node.found_extent_tree && back->node.found_ref) {
6577                         list_del(&back->node.list);
6578                         free(back);
6579                 }
6580         }
6581         maybe_free_extent_rec(extent_cache, rec);
6582 out:
6583         return 0;
6584 }
6585
6586 static int delete_extent_records(struct btrfs_trans_handle *trans,
6587                                  struct btrfs_root *root,
6588                                  struct btrfs_path *path,
6589                                  u64 bytenr, u64 new_len)
6590 {
6591         struct btrfs_key key;
6592         struct btrfs_key found_key;
6593         struct extent_buffer *leaf;
6594         int ret;
6595         int slot;
6596
6597
6598         key.objectid = bytenr;
6599         key.type = (u8)-1;
6600         key.offset = (u64)-1;
6601
6602         while(1) {
6603                 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
6604                                         &key, path, 0, 1);
6605                 if (ret < 0)
6606                         break;
6607
6608                 if (ret > 0) {
6609                         ret = 0;
6610                         if (path->slots[0] == 0)
6611                                 break;
6612                         path->slots[0]--;
6613                 }
6614                 ret = 0;
6615
6616                 leaf = path->nodes[0];
6617                 slot = path->slots[0];
6618
6619                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6620                 if (found_key.objectid != bytenr)
6621                         break;
6622
6623                 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
6624                     found_key.type != BTRFS_METADATA_ITEM_KEY &&
6625                     found_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
6626                     found_key.type != BTRFS_EXTENT_DATA_REF_KEY &&
6627                     found_key.type != BTRFS_EXTENT_REF_V0_KEY &&
6628                     found_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
6629                     found_key.type != BTRFS_SHARED_DATA_REF_KEY) {
6630                         btrfs_release_path(path);
6631                         if (found_key.type == 0) {
6632                                 if (found_key.offset == 0)
6633                                         break;
6634                                 key.offset = found_key.offset - 1;
6635                                 key.type = found_key.type;
6636                         }
6637                         key.type = found_key.type - 1;
6638                         key.offset = (u64)-1;
6639                         continue;
6640                 }
6641
6642                 fprintf(stderr, "repair deleting extent record: key %Lu %u %Lu\n",
6643                         found_key.objectid, found_key.type, found_key.offset);
6644
6645                 ret = btrfs_del_item(trans, root->fs_info->extent_root, path);
6646                 if (ret)
6647                         break;
6648                 btrfs_release_path(path);
6649
6650                 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
6651                     found_key.type == BTRFS_METADATA_ITEM_KEY) {
6652                         u64 bytes = (found_key.type == BTRFS_EXTENT_ITEM_KEY) ?
6653                                 found_key.offset : root->nodesize;
6654
6655                         ret = btrfs_update_block_group(trans, root, bytenr,
6656                                                        bytes, 0, 0);
6657                         if (ret)
6658                                 break;
6659                 }
6660         }
6661
6662         btrfs_release_path(path);
6663         return ret;
6664 }
6665
6666 /*
6667  * for a single backref, this will allocate a new extent
6668  * and add the backref to it.
6669  */
6670 static int record_extent(struct btrfs_trans_handle *trans,
6671                          struct btrfs_fs_info *info,
6672                          struct btrfs_path *path,
6673                          struct extent_record *rec,
6674                          struct extent_backref *back,
6675                          int allocated, u64 flags)
6676 {
6677         int ret;
6678         struct btrfs_root *extent_root = info->extent_root;
6679         struct extent_buffer *leaf;
6680         struct btrfs_key ins_key;
6681         struct btrfs_extent_item *ei;
6682         struct tree_backref *tback;
6683         struct data_backref *dback;
6684         struct btrfs_tree_block_info *bi;
6685
6686         if (!back->is_data)
6687                 rec->max_size = max_t(u64, rec->max_size,
6688                                     info->extent_root->nodesize);
6689
6690         if (!allocated) {
6691                 u32 item_size = sizeof(*ei);
6692
6693                 if (!back->is_data)
6694                         item_size += sizeof(*bi);
6695
6696                 ins_key.objectid = rec->start;
6697                 ins_key.offset = rec->max_size;
6698                 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
6699
6700                 ret = btrfs_insert_empty_item(trans, extent_root, path,
6701                                         &ins_key, item_size);
6702                 if (ret)
6703                         goto fail;
6704
6705                 leaf = path->nodes[0];
6706                 ei = btrfs_item_ptr(leaf, path->slots[0],
6707                                     struct btrfs_extent_item);
6708
6709                 btrfs_set_extent_refs(leaf, ei, 0);
6710                 btrfs_set_extent_generation(leaf, ei, rec->generation);
6711
6712                 if (back->is_data) {
6713                         btrfs_set_extent_flags(leaf, ei,
6714                                                BTRFS_EXTENT_FLAG_DATA);
6715                 } else {
6716                         struct btrfs_disk_key copy_key;;
6717
6718                         tback = to_tree_backref(back);
6719                         bi = (struct btrfs_tree_block_info *)(ei + 1);
6720                         memset_extent_buffer(leaf, 0, (unsigned long)bi,
6721                                              sizeof(*bi));
6722
6723                         btrfs_set_disk_key_objectid(&copy_key,
6724                                                     rec->info_objectid);
6725                         btrfs_set_disk_key_type(&copy_key, 0);
6726                         btrfs_set_disk_key_offset(&copy_key, 0);
6727
6728                         btrfs_set_tree_block_level(leaf, bi, rec->info_level);
6729                         btrfs_set_tree_block_key(leaf, bi, &copy_key);
6730
6731                         btrfs_set_extent_flags(leaf, ei,
6732                                                BTRFS_EXTENT_FLAG_TREE_BLOCK | flags);
6733                 }
6734
6735                 btrfs_mark_buffer_dirty(leaf);
6736                 ret = btrfs_update_block_group(trans, extent_root, rec->start,
6737                                                rec->max_size, 1, 0);
6738                 if (ret)
6739                         goto fail;
6740                 btrfs_release_path(path);
6741         }
6742
6743         if (back->is_data) {
6744                 u64 parent;
6745                 int i;
6746
6747                 dback = to_data_backref(back);
6748                 if (back->full_backref)
6749                         parent = dback->parent;
6750                 else
6751                         parent = 0;
6752
6753                 for (i = 0; i < dback->found_ref; i++) {
6754                         /* if parent != 0, we're doing a full backref
6755                          * passing BTRFS_FIRST_FREE_OBJECTID as the owner
6756                          * just makes the backref allocator create a data
6757                          * backref
6758                          */
6759                         ret = btrfs_inc_extent_ref(trans, info->extent_root,
6760                                                    rec->start, rec->max_size,
6761                                                    parent,
6762                                                    dback->root,
6763                                                    parent ?
6764                                                    BTRFS_FIRST_FREE_OBJECTID :
6765                                                    dback->owner,
6766                                                    dback->offset);
6767                         if (ret)
6768                                 break;
6769                 }
6770                 fprintf(stderr, "adding new data backref"
6771                                 " on %llu %s %llu owner %llu"
6772                                 " offset %llu found %d\n",
6773                                 (unsigned long long)rec->start,
6774                                 back->full_backref ?
6775                                 "parent" : "root",
6776                                 back->full_backref ?
6777                                 (unsigned long long)parent :
6778                                 (unsigned long long)dback->root,
6779                                 (unsigned long long)dback->owner,
6780                                 (unsigned long long)dback->offset,
6781                                 dback->found_ref);
6782         } else {
6783                 u64 parent;
6784
6785                 tback = to_tree_backref(back);
6786                 if (back->full_backref)
6787                         parent = tback->parent;
6788                 else
6789                         parent = 0;
6790
6791                 ret = btrfs_inc_extent_ref(trans, info->extent_root,
6792                                            rec->start, rec->max_size,
6793                                            parent, tback->root, 0, 0);
6794                 fprintf(stderr, "adding new tree backref on "
6795                         "start %llu len %llu parent %llu root %llu\n",
6796                         rec->start, rec->max_size, parent, tback->root);
6797         }
6798 fail:
6799         btrfs_release_path(path);
6800         return ret;
6801 }
6802
6803 static struct extent_entry *find_entry(struct list_head *entries,
6804                                        u64 bytenr, u64 bytes)
6805 {
6806         struct extent_entry *entry = NULL;
6807
6808         list_for_each_entry(entry, entries, list) {
6809                 if (entry->bytenr == bytenr && entry->bytes == bytes)
6810                         return entry;
6811         }
6812
6813         return NULL;
6814 }
6815
6816 static struct extent_entry *find_most_right_entry(struct list_head *entries)
6817 {
6818         struct extent_entry *entry, *best = NULL, *prev = NULL;
6819
6820         list_for_each_entry(entry, entries, list) {
6821                 if (!prev) {
6822                         prev = entry;
6823                         continue;
6824                 }
6825
6826                 /*
6827                  * If there are as many broken entries as entries then we know
6828                  * not to trust this particular entry.
6829                  */
6830                 if (entry->broken == entry->count)
6831                         continue;
6832
6833                 /*
6834                  * If our current entry == best then we can't be sure our best
6835                  * is really the best, so we need to keep searching.
6836                  */
6837                 if (best && best->count == entry->count) {
6838                         prev = entry;
6839                         best = NULL;
6840                         continue;
6841                 }
6842
6843                 /* Prev == entry, not good enough, have to keep searching */
6844                 if (!prev->broken && prev->count == entry->count)
6845                         continue;
6846
6847                 if (!best)
6848                         best = (prev->count > entry->count) ? prev : entry;
6849                 else if (best->count < entry->count)
6850                         best = entry;
6851                 prev = entry;
6852         }
6853
6854         return best;
6855 }
6856
6857 static int repair_ref(struct btrfs_fs_info *info, struct btrfs_path *path,
6858                       struct data_backref *dback, struct extent_entry *entry)
6859 {
6860         struct btrfs_trans_handle *trans;
6861         struct btrfs_root *root;
6862         struct btrfs_file_extent_item *fi;
6863         struct extent_buffer *leaf;
6864         struct btrfs_key key;
6865         u64 bytenr, bytes;
6866         int ret, err;
6867
6868         key.objectid = dback->root;
6869         key.type = BTRFS_ROOT_ITEM_KEY;
6870         key.offset = (u64)-1;
6871         root = btrfs_read_fs_root(info, &key);
6872         if (IS_ERR(root)) {
6873                 fprintf(stderr, "Couldn't find root for our ref\n");
6874                 return -EINVAL;
6875         }
6876
6877         /*
6878          * The backref points to the original offset of the extent if it was
6879          * split, so we need to search down to the offset we have and then walk
6880          * forward until we find the backref we're looking for.
6881          */
6882         key.objectid = dback->owner;
6883         key.type = BTRFS_EXTENT_DATA_KEY;
6884         key.offset = dback->offset;
6885         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6886         if (ret < 0) {
6887                 fprintf(stderr, "Error looking up ref %d\n", ret);
6888                 return ret;
6889         }
6890
6891         while (1) {
6892                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
6893                         ret = btrfs_next_leaf(root, path);
6894                         if (ret) {
6895                                 fprintf(stderr, "Couldn't find our ref, next\n");
6896                                 return -EINVAL;
6897                         }
6898                 }
6899                 leaf = path->nodes[0];
6900                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6901                 if (key.objectid != dback->owner ||
6902                     key.type != BTRFS_EXTENT_DATA_KEY) {
6903                         fprintf(stderr, "Couldn't find our ref, search\n");
6904                         return -EINVAL;
6905                 }
6906                 fi = btrfs_item_ptr(leaf, path->slots[0],
6907                                     struct btrfs_file_extent_item);
6908                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6909                 bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
6910
6911                 if (bytenr == dback->disk_bytenr && bytes == dback->bytes)
6912                         break;
6913                 path->slots[0]++;
6914         }
6915
6916         btrfs_release_path(path);
6917
6918         trans = btrfs_start_transaction(root, 1);
6919         if (IS_ERR(trans))
6920                 return PTR_ERR(trans);
6921
6922         /*
6923          * Ok we have the key of the file extent we want to fix, now we can cow
6924          * down to the thing and fix it.
6925          */
6926         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6927         if (ret < 0) {
6928                 fprintf(stderr, "Error cowing down to ref [%Lu, %u, %Lu]: %d\n",
6929                         key.objectid, key.type, key.offset, ret);
6930                 goto out;
6931         }
6932         if (ret > 0) {
6933                 fprintf(stderr, "Well that's odd, we just found this key "
6934                         "[%Lu, %u, %Lu]\n", key.objectid, key.type,
6935                         key.offset);
6936                 ret = -EINVAL;
6937                 goto out;
6938         }
6939         leaf = path->nodes[0];
6940         fi = btrfs_item_ptr(leaf, path->slots[0],
6941                             struct btrfs_file_extent_item);
6942
6943         if (btrfs_file_extent_compression(leaf, fi) &&
6944             dback->disk_bytenr != entry->bytenr) {
6945                 fprintf(stderr, "Ref doesn't match the record start and is "
6946                         "compressed, please take a btrfs-image of this file "
6947                         "system and send it to a btrfs developer so they can "
6948                         "complete this functionality for bytenr %Lu\n",
6949                         dback->disk_bytenr);
6950                 ret = -EINVAL;
6951                 goto out;
6952         }
6953
6954         if (dback->node.broken && dback->disk_bytenr != entry->bytenr) {
6955                 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
6956         } else if (dback->disk_bytenr > entry->bytenr) {
6957                 u64 off_diff, offset;
6958
6959                 off_diff = dback->disk_bytenr - entry->bytenr;
6960                 offset = btrfs_file_extent_offset(leaf, fi);
6961                 if (dback->disk_bytenr + offset +
6962                     btrfs_file_extent_num_bytes(leaf, fi) >
6963                     entry->bytenr + entry->bytes) {
6964                         fprintf(stderr, "Ref is past the entry end, please "
6965                                 "take a btrfs-image of this file system and "
6966                                 "send it to a btrfs developer, ref %Lu\n",
6967                                 dback->disk_bytenr);
6968                         ret = -EINVAL;
6969                         goto out;
6970                 }
6971                 offset += off_diff;
6972                 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
6973                 btrfs_set_file_extent_offset(leaf, fi, offset);
6974         } else if (dback->disk_bytenr < entry->bytenr) {
6975                 u64 offset;
6976
6977                 offset = btrfs_file_extent_offset(leaf, fi);
6978                 if (dback->disk_bytenr + offset < entry->bytenr) {
6979                         fprintf(stderr, "Ref is before the entry start, please"
6980                                 " take a btrfs-image of this file system and "
6981                                 "send it to a btrfs developer, ref %Lu\n",
6982                                 dback->disk_bytenr);
6983                         ret = -EINVAL;
6984                         goto out;
6985                 }
6986
6987                 offset += dback->disk_bytenr;
6988                 offset -= entry->bytenr;
6989                 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
6990                 btrfs_set_file_extent_offset(leaf, fi, offset);
6991         }
6992
6993         btrfs_set_file_extent_disk_num_bytes(leaf, fi, entry->bytes);
6994
6995         /*
6996          * Chances are if disk_num_bytes were wrong then so is ram_bytes, but
6997          * only do this if we aren't using compression, otherwise it's a
6998          * trickier case.
6999          */
7000         if (!btrfs_file_extent_compression(leaf, fi))
7001                 btrfs_set_file_extent_ram_bytes(leaf, fi, entry->bytes);
7002         else
7003                 printf("ram bytes may be wrong?\n");
7004         btrfs_mark_buffer_dirty(leaf);
7005 out:
7006         err = btrfs_commit_transaction(trans, root);
7007         btrfs_release_path(path);
7008         return ret ? ret : err;
7009 }
7010
7011 static int verify_backrefs(struct btrfs_fs_info *info, struct btrfs_path *path,
7012                            struct extent_record *rec)
7013 {
7014         struct extent_backref *back;
7015         struct data_backref *dback;
7016         struct extent_entry *entry, *best = NULL;
7017         LIST_HEAD(entries);
7018         int nr_entries = 0;
7019         int broken_entries = 0;
7020         int ret = 0;
7021         short mismatch = 0;
7022
7023         /*
7024          * Metadata is easy and the backrefs should always agree on bytenr and
7025          * size, if not we've got bigger issues.
7026          */
7027         if (rec->metadata)
7028                 return 0;
7029
7030         list_for_each_entry(back, &rec->backrefs, list) {
7031                 if (back->full_backref || !back->is_data)
7032                         continue;
7033
7034                 dback = to_data_backref(back);
7035
7036                 /*
7037                  * We only pay attention to backrefs that we found a real
7038                  * backref for.
7039                  */
7040                 if (dback->found_ref == 0)
7041                         continue;
7042
7043                 /*
7044                  * For now we only catch when the bytes don't match, not the
7045                  * bytenr.  We can easily do this at the same time, but I want
7046                  * to have a fs image to test on before we just add repair
7047                  * functionality willy-nilly so we know we won't screw up the
7048                  * repair.
7049                  */
7050
7051                 entry = find_entry(&entries, dback->disk_bytenr,
7052                                    dback->bytes);
7053                 if (!entry) {
7054                         entry = malloc(sizeof(struct extent_entry));
7055                         if (!entry) {
7056                                 ret = -ENOMEM;
7057                                 goto out;
7058                         }
7059                         memset(entry, 0, sizeof(*entry));
7060                         entry->bytenr = dback->disk_bytenr;
7061                         entry->bytes = dback->bytes;
7062                         list_add_tail(&entry->list, &entries);
7063                         nr_entries++;
7064                 }
7065
7066                 /*
7067                  * If we only have on entry we may think the entries agree when
7068                  * in reality they don't so we have to do some extra checking.
7069                  */
7070                 if (dback->disk_bytenr != rec->start ||
7071                     dback->bytes != rec->nr || back->broken)
7072                         mismatch = 1;
7073
7074                 if (back->broken) {
7075                         entry->broken++;
7076                         broken_entries++;
7077                 }
7078
7079                 entry->count++;
7080         }
7081
7082         /* Yay all the backrefs agree, carry on good sir */
7083         if (nr_entries <= 1 && !mismatch)
7084                 goto out;
7085
7086         fprintf(stderr, "attempting to repair backref discrepency for bytenr "
7087                 "%Lu\n", rec->start);
7088
7089         /*
7090          * First we want to see if the backrefs can agree amongst themselves who
7091          * is right, so figure out which one of the entries has the highest
7092          * count.
7093          */
7094         best = find_most_right_entry(&entries);
7095
7096         /*
7097          * Ok so we may have an even split between what the backrefs think, so
7098          * this is where we use the extent ref to see what it thinks.
7099          */
7100         if (!best) {
7101                 entry = find_entry(&entries, rec->start, rec->nr);
7102                 if (!entry && (!broken_entries || !rec->found_rec)) {
7103                         fprintf(stderr, "Backrefs don't agree with each other "
7104                                 "and extent record doesn't agree with anybody,"
7105                                 " so we can't fix bytenr %Lu bytes %Lu\n",
7106                                 rec->start, rec->nr);
7107                         ret = -EINVAL;
7108                         goto out;
7109                 } else if (!entry) {
7110                         /*
7111                          * Ok our backrefs were broken, we'll assume this is the
7112                          * correct value and add an entry for this range.
7113                          */
7114                         entry = malloc(sizeof(struct extent_entry));
7115                         if (!entry) {
7116                                 ret = -ENOMEM;
7117                                 goto out;
7118                         }
7119                         memset(entry, 0, sizeof(*entry));
7120                         entry->bytenr = rec->start;
7121                         entry->bytes = rec->nr;
7122                         list_add_tail(&entry->list, &entries);
7123                         nr_entries++;
7124                 }
7125                 entry->count++;
7126                 best = find_most_right_entry(&entries);
7127                 if (!best) {
7128                         fprintf(stderr, "Backrefs and extent record evenly "
7129                                 "split on who is right, this is going to "
7130                                 "require user input to fix bytenr %Lu bytes "
7131                                 "%Lu\n", rec->start, rec->nr);
7132                         ret = -EINVAL;
7133                         goto out;
7134                 }
7135         }
7136
7137         /*
7138          * I don't think this can happen currently as we'll abort() if we catch
7139          * this case higher up, but in case somebody removes that we still can't
7140          * deal with it properly here yet, so just bail out of that's the case.
7141          */
7142         if (best->bytenr != rec->start) {
7143                 fprintf(stderr, "Extent start and backref starts don't match, "
7144                         "please use btrfs-image on this file system and send "
7145                         "it to a btrfs developer so they can make fsck fix "
7146                         "this particular case.  bytenr is %Lu, bytes is %Lu\n",
7147                         rec->start, rec->nr);
7148                 ret = -EINVAL;
7149                 goto out;
7150         }
7151
7152         /*
7153          * Ok great we all agreed on an extent record, let's go find the real
7154          * references and fix up the ones that don't match.
7155          */
7156         list_for_each_entry(back, &rec->backrefs, list) {
7157                 if (back->full_backref || !back->is_data)
7158                         continue;
7159
7160                 dback = to_data_backref(back);
7161
7162                 /*
7163                  * Still ignoring backrefs that don't have a real ref attached
7164                  * to them.
7165                  */
7166                 if (dback->found_ref == 0)
7167                         continue;
7168
7169                 if (dback->bytes == best->bytes &&
7170                     dback->disk_bytenr == best->bytenr)
7171                         continue;
7172
7173                 ret = repair_ref(info, path, dback, best);
7174                 if (ret)
7175                         goto out;
7176         }
7177
7178         /*
7179          * Ok we messed with the actual refs, which means we need to drop our
7180          * entire cache and go back and rescan.  I know this is a huge pain and
7181          * adds a lot of extra work, but it's the only way to be safe.  Once all
7182          * the backrefs agree we may not need to do anything to the extent
7183          * record itself.
7184          */
7185         ret = -EAGAIN;
7186 out:
7187         while (!list_empty(&entries)) {
7188                 entry = list_entry(entries.next, struct extent_entry, list);
7189                 list_del_init(&entry->list);
7190                 free(entry);
7191         }
7192         return ret;
7193 }
7194
7195 static int process_duplicates(struct btrfs_root *root,
7196                               struct cache_tree *extent_cache,
7197                               struct extent_record *rec)
7198 {
7199         struct extent_record *good, *tmp;
7200         struct cache_extent *cache;
7201         int ret;
7202
7203         /*
7204          * If we found a extent record for this extent then return, or if we
7205          * have more than one duplicate we are likely going to need to delete
7206          * something.
7207          */
7208         if (rec->found_rec || rec->num_duplicates > 1)
7209                 return 0;
7210
7211         /* Shouldn't happen but just in case */
7212         BUG_ON(!rec->num_duplicates);
7213
7214         /*
7215          * So this happens if we end up with a backref that doesn't match the
7216          * actual extent entry.  So either the backref is bad or the extent
7217          * entry is bad.  Either way we want to have the extent_record actually
7218          * reflect what we found in the extent_tree, so we need to take the
7219          * duplicate out and use that as the extent_record since the only way we
7220          * get a duplicate is if we find a real life BTRFS_EXTENT_ITEM_KEY.
7221          */
7222         remove_cache_extent(extent_cache, &rec->cache);
7223
7224         good = to_extent_record(rec->dups.next);
7225         list_del_init(&good->list);
7226         INIT_LIST_HEAD(&good->backrefs);
7227         INIT_LIST_HEAD(&good->dups);
7228         good->cache.start = good->start;
7229         good->cache.size = good->nr;
7230         good->content_checked = 0;
7231         good->owner_ref_checked = 0;
7232         good->num_duplicates = 0;
7233         good->refs = rec->refs;
7234         list_splice_init(&rec->backrefs, &good->backrefs);
7235         while (1) {
7236                 cache = lookup_cache_extent(extent_cache, good->start,
7237                                             good->nr);
7238                 if (!cache)
7239                         break;
7240                 tmp = container_of(cache, struct extent_record, cache);
7241
7242                 /*
7243                  * If we find another overlapping extent and it's found_rec is
7244                  * set then it's a duplicate and we need to try and delete
7245                  * something.
7246                  */
7247                 if (tmp->found_rec || tmp->num_duplicates > 0) {
7248                         if (list_empty(&good->list))
7249                                 list_add_tail(&good->list,
7250                                               &duplicate_extents);
7251                         good->num_duplicates += tmp->num_duplicates + 1;
7252                         list_splice_init(&tmp->dups, &good->dups);
7253                         list_del_init(&tmp->list);
7254                         list_add_tail(&tmp->list, &good->dups);
7255                         remove_cache_extent(extent_cache, &tmp->cache);
7256                         continue;
7257                 }
7258
7259                 /*
7260                  * Ok we have another non extent item backed extent rec, so lets
7261                  * just add it to this extent and carry on like we did above.
7262                  */
7263                 good->refs += tmp->refs;
7264                 list_splice_init(&tmp->backrefs, &good->backrefs);
7265                 remove_cache_extent(extent_cache, &tmp->cache);
7266                 free(tmp);
7267         }
7268         ret = insert_cache_extent(extent_cache, &good->cache);
7269         BUG_ON(ret);
7270         free(rec);
7271         return good->num_duplicates ? 0 : 1;
7272 }
7273
7274 static int delete_duplicate_records(struct btrfs_root *root,
7275                                     struct extent_record *rec)
7276 {
7277         struct btrfs_trans_handle *trans;
7278         LIST_HEAD(delete_list);
7279         struct btrfs_path *path;
7280         struct extent_record *tmp, *good, *n;
7281         int nr_del = 0;
7282         int ret = 0, err;
7283         struct btrfs_key key;
7284
7285         path = btrfs_alloc_path();
7286         if (!path) {
7287                 ret = -ENOMEM;
7288                 goto out;
7289         }
7290
7291         good = rec;
7292         /* Find the record that covers all of the duplicates. */
7293         list_for_each_entry(tmp, &rec->dups, list) {
7294                 if (good->start < tmp->start)
7295                         continue;
7296                 if (good->nr > tmp->nr)
7297                         continue;
7298
7299                 if (tmp->start + tmp->nr < good->start + good->nr) {
7300                         fprintf(stderr, "Ok we have overlapping extents that "
7301                                 "aren't completely covered by each other, this "
7302                                 "is going to require more careful thought.  "
7303                                 "The extents are [%Lu-%Lu] and [%Lu-%Lu]\n",
7304                                 tmp->start, tmp->nr, good->start, good->nr);
7305                         abort();
7306                 }
7307                 good = tmp;
7308         }
7309
7310         if (good != rec)
7311                 list_add_tail(&rec->list, &delete_list);
7312
7313         list_for_each_entry_safe(tmp, n, &rec->dups, list) {
7314                 if (tmp == good)
7315                         continue;
7316                 list_move_tail(&tmp->list, &delete_list);
7317         }
7318
7319         root = root->fs_info->extent_root;
7320         trans = btrfs_start_transaction(root, 1);
7321         if (IS_ERR(trans)) {
7322                 ret = PTR_ERR(trans);
7323                 goto out;
7324         }
7325
7326         list_for_each_entry(tmp, &delete_list, list) {
7327                 if (tmp->found_rec == 0)
7328                         continue;
7329                 key.objectid = tmp->start;
7330                 key.type = BTRFS_EXTENT_ITEM_KEY;
7331                 key.offset = tmp->nr;
7332
7333                 /* Shouldn't happen but just in case */
7334                 if (tmp->metadata) {
7335                         fprintf(stderr, "Well this shouldn't happen, extent "
7336                                 "record overlaps but is metadata? "
7337                                 "[%Lu, %Lu]\n", tmp->start, tmp->nr);
7338                         abort();
7339                 }
7340
7341                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7342                 if (ret) {
7343                         if (ret > 0)
7344                                 ret = -EINVAL;
7345                         break;
7346                 }
7347                 ret = btrfs_del_item(trans, root, path);
7348                 if (ret)
7349                         break;
7350                 btrfs_release_path(path);
7351                 nr_del++;
7352         }
7353         err = btrfs_commit_transaction(trans, root);
7354         if (err && !ret)
7355                 ret = err;
7356 out:
7357         while (!list_empty(&delete_list)) {
7358                 tmp = to_extent_record(delete_list.next);
7359                 list_del_init(&tmp->list);
7360                 if (tmp == rec)
7361                         continue;
7362                 free(tmp);
7363         }
7364
7365         while (!list_empty(&rec->dups)) {
7366                 tmp = to_extent_record(rec->dups.next);
7367                 list_del_init(&tmp->list);
7368                 free(tmp);
7369         }
7370
7371         btrfs_free_path(path);
7372
7373         if (!ret && !nr_del)
7374                 rec->num_duplicates = 0;
7375
7376         return ret ? ret : nr_del;
7377 }
7378
7379 static int find_possible_backrefs(struct btrfs_fs_info *info,
7380                                   struct btrfs_path *path,
7381                                   struct cache_tree *extent_cache,
7382                                   struct extent_record *rec)
7383 {
7384         struct btrfs_root *root;
7385         struct extent_backref *back;
7386         struct data_backref *dback;
7387         struct cache_extent *cache;
7388         struct btrfs_file_extent_item *fi;
7389         struct btrfs_key key;
7390         u64 bytenr, bytes;
7391         int ret;
7392
7393         list_for_each_entry(back, &rec->backrefs, list) {
7394                 /* Don't care about full backrefs (poor unloved backrefs) */
7395                 if (back->full_backref || !back->is_data)
7396                         continue;
7397
7398                 dback = to_data_backref(back);
7399
7400                 /* We found this one, we don't need to do a lookup */
7401                 if (dback->found_ref)
7402                         continue;
7403
7404                 key.objectid = dback->root;
7405                 key.type = BTRFS_ROOT_ITEM_KEY;
7406                 key.offset = (u64)-1;
7407
7408                 root = btrfs_read_fs_root(info, &key);
7409
7410                 /* No root, definitely a bad ref, skip */
7411                 if (IS_ERR(root) && PTR_ERR(root) == -ENOENT)
7412                         continue;
7413                 /* Other err, exit */
7414                 if (IS_ERR(root))
7415                         return PTR_ERR(root);
7416
7417                 key.objectid = dback->owner;
7418                 key.type = BTRFS_EXTENT_DATA_KEY;
7419                 key.offset = dback->offset;
7420                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7421                 if (ret) {
7422                         btrfs_release_path(path);
7423                         if (ret < 0)
7424                                 return ret;
7425                         /* Didn't find it, we can carry on */
7426                         ret = 0;
7427                         continue;
7428                 }
7429
7430                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
7431                                     struct btrfs_file_extent_item);
7432                 bytenr = btrfs_file_extent_disk_bytenr(path->nodes[0], fi);
7433                 bytes = btrfs_file_extent_disk_num_bytes(path->nodes[0], fi);
7434                 btrfs_release_path(path);
7435                 cache = lookup_cache_extent(extent_cache, bytenr, 1);
7436                 if (cache) {
7437                         struct extent_record *tmp;
7438                         tmp = container_of(cache, struct extent_record, cache);
7439
7440                         /*
7441                          * If we found an extent record for the bytenr for this
7442                          * particular backref then we can't add it to our
7443                          * current extent record.  We only want to add backrefs
7444                          * that don't have a corresponding extent item in the
7445                          * extent tree since they likely belong to this record
7446                          * and we need to fix it if it doesn't match bytenrs.
7447                          */
7448                         if  (tmp->found_rec)
7449                                 continue;
7450                 }
7451
7452                 dback->found_ref += 1;
7453                 dback->disk_bytenr = bytenr;
7454                 dback->bytes = bytes;
7455
7456                 /*
7457                  * Set this so the verify backref code knows not to trust the
7458                  * values in this backref.
7459                  */
7460                 back->broken = 1;
7461         }
7462
7463         return 0;
7464 }
7465
7466 /*
7467  * Record orphan data ref into corresponding root.
7468  *
7469  * Return 0 if the extent item contains data ref and recorded.
7470  * Return 1 if the extent item contains no useful data ref
7471  *   On that case, it may contains only shared_dataref or metadata backref
7472  *   or the file extent exists(this should be handled by the extent bytenr
7473  *   recovery routine)
7474  * Return <0 if something goes wrong.
7475  */
7476 static int record_orphan_data_extents(struct btrfs_fs_info *fs_info,
7477                                       struct extent_record *rec)
7478 {
7479         struct btrfs_key key;
7480         struct btrfs_root *dest_root;
7481         struct extent_backref *back;
7482         struct data_backref *dback;
7483         struct orphan_data_extent *orphan;
7484         struct btrfs_path *path;
7485         int recorded_data_ref = 0;
7486         int ret = 0;
7487
7488         if (rec->metadata)
7489                 return 1;
7490         path = btrfs_alloc_path();
7491         if (!path)
7492                 return -ENOMEM;
7493         list_for_each_entry(back, &rec->backrefs, list) {
7494                 if (back->full_backref || !back->is_data ||
7495                     !back->found_extent_tree)
7496                         continue;
7497                 dback = to_data_backref(back);
7498                 if (dback->found_ref)
7499                         continue;
7500                 key.objectid = dback->root;
7501                 key.type = BTRFS_ROOT_ITEM_KEY;
7502                 key.offset = (u64)-1;
7503
7504                 dest_root = btrfs_read_fs_root(fs_info, &key);
7505
7506                 /* For non-exist root we just skip it */
7507                 if (IS_ERR(dest_root) || !dest_root)
7508                         continue;
7509
7510                 key.objectid = dback->owner;
7511                 key.type = BTRFS_EXTENT_DATA_KEY;
7512                 key.offset = dback->offset;
7513
7514                 ret = btrfs_search_slot(NULL, dest_root, &key, path, 0, 0);
7515                 btrfs_release_path(path);
7516                 /*
7517                  * For ret < 0, it's OK since the fs-tree may be corrupted,
7518                  * we need to record it for inode/file extent rebuild.
7519                  * For ret > 0, we record it only for file extent rebuild.
7520                  * For ret == 0, the file extent exists but only bytenr
7521                  * mismatch, let the original bytenr fix routine to handle,
7522                  * don't record it.
7523                  */
7524                 if (ret == 0)
7525                         continue;
7526                 ret = 0;
7527                 orphan = malloc(sizeof(*orphan));
7528                 if (!orphan) {
7529                         ret = -ENOMEM;
7530                         goto out;
7531                 }
7532                 INIT_LIST_HEAD(&orphan->list);
7533                 orphan->root = dback->root;
7534                 orphan->objectid = dback->owner;
7535                 orphan->offset = dback->offset;
7536                 orphan->disk_bytenr = rec->cache.start;
7537                 orphan->disk_len = rec->cache.size;
7538                 list_add(&dest_root->orphan_data_extents, &orphan->list);
7539                 recorded_data_ref = 1;
7540         }
7541 out:
7542         btrfs_free_path(path);
7543         if (!ret)
7544                 return !recorded_data_ref;
7545         else
7546                 return ret;
7547 }
7548
7549 /*
7550  * when an incorrect extent item is found, this will delete
7551  * all of the existing entries for it and recreate them
7552  * based on what the tree scan found.
7553  */
7554 static int fixup_extent_refs(struct btrfs_fs_info *info,
7555                              struct cache_tree *extent_cache,
7556                              struct extent_record *rec)
7557 {
7558         struct btrfs_trans_handle *trans = NULL;
7559         int ret;
7560         struct btrfs_path *path;
7561         struct list_head *cur = rec->backrefs.next;
7562         struct cache_extent *cache;
7563         struct extent_backref *back;
7564         int allocated = 0;
7565         u64 flags = 0;
7566
7567         if (rec->flag_block_full_backref)
7568                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7569
7570         path = btrfs_alloc_path();
7571         if (!path)
7572                 return -ENOMEM;
7573
7574         if (rec->refs != rec->extent_item_refs && !rec->metadata) {
7575                 /*
7576                  * Sometimes the backrefs themselves are so broken they don't
7577                  * get attached to any meaningful rec, so first go back and
7578                  * check any of our backrefs that we couldn't find and throw
7579                  * them into the list if we find the backref so that
7580                  * verify_backrefs can figure out what to do.
7581                  */
7582                 ret = find_possible_backrefs(info, path, extent_cache, rec);
7583                 if (ret < 0)
7584                         goto out;
7585         }
7586
7587         /* step one, make sure all of the backrefs agree */
7588         ret = verify_backrefs(info, path, rec);
7589         if (ret < 0)
7590                 goto out;
7591
7592         trans = btrfs_start_transaction(info->extent_root, 1);
7593         if (IS_ERR(trans)) {
7594                 ret = PTR_ERR(trans);
7595                 goto out;
7596         }
7597
7598         /* step two, delete all the existing records */
7599         ret = delete_extent_records(trans, info->extent_root, path,
7600                                     rec->start, rec->max_size);
7601
7602         if (ret < 0)
7603                 goto out;
7604
7605         /* was this block corrupt?  If so, don't add references to it */
7606         cache = lookup_cache_extent(info->corrupt_blocks,
7607                                     rec->start, rec->max_size);
7608         if (cache) {
7609                 ret = 0;
7610                 goto out;
7611         }
7612
7613         /* step three, recreate all the refs we did find */
7614         while(cur != &rec->backrefs) {
7615                 back = to_extent_backref(cur);
7616                 cur = cur->next;
7617
7618                 /*
7619                  * if we didn't find any references, don't create a
7620                  * new extent record
7621                  */
7622                 if (!back->found_ref)
7623                         continue;
7624
7625                 rec->bad_full_backref = 0;
7626                 ret = record_extent(trans, info, path, rec, back, allocated, flags);
7627                 allocated = 1;
7628
7629                 if (ret)
7630                         goto out;
7631         }
7632 out:
7633         if (trans) {
7634                 int err = btrfs_commit_transaction(trans, info->extent_root);
7635                 if (!ret)
7636                         ret = err;
7637         }
7638
7639         btrfs_free_path(path);
7640         return ret;
7641 }
7642
7643 static int fixup_extent_flags(struct btrfs_fs_info *fs_info,
7644                               struct extent_record *rec)
7645 {
7646         struct btrfs_trans_handle *trans;
7647         struct btrfs_root *root = fs_info->extent_root;
7648         struct btrfs_path *path;
7649         struct btrfs_extent_item *ei;
7650         struct btrfs_key key;
7651         u64 flags;
7652         int ret = 0;
7653
7654         key.objectid = rec->start;
7655         if (rec->metadata) {
7656                 key.type = BTRFS_METADATA_ITEM_KEY;
7657                 key.offset = rec->info_level;
7658         } else {
7659                 key.type = BTRFS_EXTENT_ITEM_KEY;
7660                 key.offset = rec->max_size;
7661         }
7662
7663         path = btrfs_alloc_path();
7664         if (!path)
7665                 return -ENOMEM;
7666
7667         trans = btrfs_start_transaction(root, 0);
7668         if (IS_ERR(trans)) {
7669                 btrfs_free_path(path);
7670                 return PTR_ERR(trans);
7671         }
7672
7673         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
7674         if (ret < 0) {
7675                 btrfs_free_path(path);
7676                 btrfs_commit_transaction(trans, root);
7677                 return ret;
7678         } else if (ret) {
7679                 fprintf(stderr, "Didn't find extent for %llu\n",
7680                         (unsigned long long)rec->start);
7681                 btrfs_free_path(path);
7682                 btrfs_commit_transaction(trans, root);
7683                 return -ENOENT;
7684         }
7685
7686         ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
7687                             struct btrfs_extent_item);
7688         flags = btrfs_extent_flags(path->nodes[0], ei);
7689         if (rec->flag_block_full_backref) {
7690                 fprintf(stderr, "setting full backref on %llu\n",
7691                         (unsigned long long)key.objectid);
7692                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7693         } else {
7694                 fprintf(stderr, "clearing full backref on %llu\n",
7695                         (unsigned long long)key.objectid);
7696                 flags &= ~BTRFS_BLOCK_FLAG_FULL_BACKREF;
7697         }
7698         btrfs_set_extent_flags(path->nodes[0], ei, flags);
7699         btrfs_mark_buffer_dirty(path->nodes[0]);
7700         btrfs_free_path(path);
7701         return btrfs_commit_transaction(trans, root);
7702 }
7703
7704 /* right now we only prune from the extent allocation tree */
7705 static int prune_one_block(struct btrfs_trans_handle *trans,
7706                            struct btrfs_fs_info *info,
7707                            struct btrfs_corrupt_block *corrupt)
7708 {
7709         int ret;
7710         struct btrfs_path path;
7711         struct extent_buffer *eb;
7712         u64 found;
7713         int slot;
7714         int nritems;
7715         int level = corrupt->level + 1;
7716
7717         btrfs_init_path(&path);
7718 again:
7719         /* we want to stop at the parent to our busted block */
7720         path.lowest_level = level;
7721
7722         ret = btrfs_search_slot(trans, info->extent_root,
7723                                 &corrupt->key, &path, -1, 1);
7724
7725         if (ret < 0)
7726                 goto out;
7727
7728         eb = path.nodes[level];
7729         if (!eb) {
7730                 ret = -ENOENT;
7731                 goto out;
7732         }
7733
7734         /*
7735          * hopefully the search gave us the block we want to prune,
7736          * lets try that first
7737          */
7738         slot = path.slots[level];
7739         found =  btrfs_node_blockptr(eb, slot);
7740         if (found == corrupt->cache.start)
7741                 goto del_ptr;
7742
7743         nritems = btrfs_header_nritems(eb);
7744
7745         /* the search failed, lets scan this node and hope we find it */
7746         for (slot = 0; slot < nritems; slot++) {
7747                 found =  btrfs_node_blockptr(eb, slot);
7748                 if (found == corrupt->cache.start)
7749                         goto del_ptr;
7750         }
7751         /*
7752          * we couldn't find the bad block.  TODO, search all the nodes for pointers
7753          * to this block
7754          */
7755         if (eb == info->extent_root->node) {
7756                 ret = -ENOENT;
7757                 goto out;
7758         } else {
7759                 level++;
7760                 btrfs_release_path(&path);
7761                 goto again;
7762         }
7763
7764 del_ptr:
7765         printk("deleting pointer to block %Lu\n", corrupt->cache.start);
7766         ret = btrfs_del_ptr(trans, info->extent_root, &path, level, slot);
7767
7768 out:
7769         btrfs_release_path(&path);
7770         return ret;
7771 }
7772
7773 static int prune_corrupt_blocks(struct btrfs_fs_info *info)
7774 {
7775         struct btrfs_trans_handle *trans = NULL;
7776         struct cache_extent *cache;
7777         struct btrfs_corrupt_block *corrupt;
7778
7779         while (1) {
7780                 cache = search_cache_extent(info->corrupt_blocks, 0);
7781                 if (!cache)
7782                         break;
7783                 if (!trans) {
7784                         trans = btrfs_start_transaction(info->extent_root, 1);
7785                         if (IS_ERR(trans))
7786                                 return PTR_ERR(trans);
7787                 }
7788                 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
7789                 prune_one_block(trans, info, corrupt);
7790                 remove_cache_extent(info->corrupt_blocks, cache);
7791         }
7792         if (trans)
7793                 return btrfs_commit_transaction(trans, info->extent_root);
7794         return 0;
7795 }
7796
7797 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info)
7798 {
7799         struct btrfs_block_group_cache *cache;
7800         u64 start, end;
7801         int ret;
7802
7803         while (1) {
7804                 ret = find_first_extent_bit(&fs_info->free_space_cache, 0,
7805                                             &start, &end, EXTENT_DIRTY);
7806                 if (ret)
7807                         break;
7808                 clear_extent_dirty(&fs_info->free_space_cache, start, end,
7809                                    GFP_NOFS);
7810         }
7811
7812         start = 0;
7813         while (1) {
7814                 cache = btrfs_lookup_first_block_group(fs_info, start);
7815                 if (!cache)
7816                         break;
7817                 if (cache->cached)
7818                         cache->cached = 0;
7819                 start = cache->key.objectid + cache->key.offset;
7820         }
7821 }
7822
7823 static int check_extent_refs(struct btrfs_root *root,
7824                              struct cache_tree *extent_cache)
7825 {
7826         struct extent_record *rec;
7827         struct cache_extent *cache;
7828         int err = 0;
7829         int ret = 0;
7830         int fixed = 0;
7831         int had_dups = 0;
7832         int recorded = 0;
7833
7834         if (repair) {
7835                 /*
7836                  * if we're doing a repair, we have to make sure
7837                  * we don't allocate from the problem extents.
7838                  * In the worst case, this will be all the
7839                  * extents in the FS
7840                  */
7841                 cache = search_cache_extent(extent_cache, 0);
7842                 while(cache) {
7843                         rec = container_of(cache, struct extent_record, cache);
7844                         set_extent_dirty(root->fs_info->excluded_extents,
7845                                          rec->start,
7846                                          rec->start + rec->max_size - 1,
7847                                          GFP_NOFS);
7848                         cache = next_cache_extent(cache);
7849                 }
7850
7851                 /* pin down all the corrupted blocks too */
7852                 cache = search_cache_extent(root->fs_info->corrupt_blocks, 0);
7853                 while(cache) {
7854                         set_extent_dirty(root->fs_info->excluded_extents,
7855                                          cache->start,
7856                                          cache->start + cache->size - 1,
7857                                          GFP_NOFS);
7858                         cache = next_cache_extent(cache);
7859                 }
7860                 prune_corrupt_blocks(root->fs_info);
7861                 reset_cached_block_groups(root->fs_info);
7862         }
7863
7864         reset_cached_block_groups(root->fs_info);
7865
7866         /*
7867          * We need to delete any duplicate entries we find first otherwise we
7868          * could mess up the extent tree when we have backrefs that actually
7869          * belong to a different extent item and not the weird duplicate one.
7870          */
7871         while (repair && !list_empty(&duplicate_extents)) {
7872                 rec = to_extent_record(duplicate_extents.next);
7873                 list_del_init(&rec->list);
7874
7875                 /* Sometimes we can find a backref before we find an actual
7876                  * extent, so we need to process it a little bit to see if there
7877                  * truly are multiple EXTENT_ITEM_KEY's for the same range, or
7878                  * if this is a backref screwup.  If we need to delete stuff
7879                  * process_duplicates() will return 0, otherwise it will return
7880                  * 1 and we
7881                  */
7882                 if (process_duplicates(root, extent_cache, rec))
7883                         continue;
7884                 ret = delete_duplicate_records(root, rec);
7885                 if (ret < 0)
7886                         return ret;
7887                 /*
7888                  * delete_duplicate_records will return the number of entries
7889                  * deleted, so if it's greater than 0 then we know we actually
7890                  * did something and we need to remove.
7891                  */
7892                 if (ret)
7893                         had_dups = 1;
7894         }
7895
7896         if (had_dups)
7897                 return -EAGAIN;
7898
7899         while(1) {
7900                 int cur_err = 0;
7901
7902                 fixed = 0;
7903                 recorded = 0;
7904                 cache = search_cache_extent(extent_cache, 0);
7905                 if (!cache)
7906                         break;
7907                 rec = container_of(cache, struct extent_record, cache);
7908                 if (rec->num_duplicates) {
7909                         fprintf(stderr, "extent item %llu has multiple extent "
7910                                 "items\n", (unsigned long long)rec->start);
7911                         err = 1;
7912                         cur_err = 1;
7913                 }
7914
7915                 if (rec->refs != rec->extent_item_refs) {
7916                         fprintf(stderr, "ref mismatch on [%llu %llu] ",
7917                                 (unsigned long long)rec->start,
7918                                 (unsigned long long)rec->nr);
7919                         fprintf(stderr, "extent item %llu, found %llu\n",
7920                                 (unsigned long long)rec->extent_item_refs,
7921                                 (unsigned long long)rec->refs);
7922                         ret = record_orphan_data_extents(root->fs_info, rec);
7923                         if (ret < 0)
7924                                 goto repair_abort;
7925                         if (ret == 0) {
7926                                 recorded = 1;
7927                         } else {
7928                                 /*
7929                                  * we can't use the extent to repair file
7930                                  * extent, let the fallback method handle it.
7931                                  */
7932                                 if (!fixed && repair) {
7933                                         ret = fixup_extent_refs(
7934                                                         root->fs_info,
7935                                                         extent_cache, rec);
7936                                         if (ret)
7937                                                 goto repair_abort;
7938                                         fixed = 1;
7939                                 }
7940                         }
7941                         err = 1;
7942                         cur_err = 1;
7943                 }
7944                 if (all_backpointers_checked(rec, 1)) {
7945                         fprintf(stderr, "backpointer mismatch on [%llu %llu]\n",
7946                                 (unsigned long long)rec->start,
7947                                 (unsigned long long)rec->nr);
7948
7949                         if (!fixed && !recorded && repair) {
7950                                 ret = fixup_extent_refs(root->fs_info,
7951                                                         extent_cache, rec);
7952                                 if (ret)
7953                                         goto repair_abort;
7954                                 fixed = 1;
7955                         }
7956                         cur_err = 1;
7957                         err = 1;
7958                 }
7959                 if (!rec->owner_ref_checked) {
7960                         fprintf(stderr, "owner ref check failed [%llu %llu]\n",
7961                                 (unsigned long long)rec->start,
7962                                 (unsigned long long)rec->nr);
7963                         if (!fixed && !recorded && repair) {
7964                                 ret = fixup_extent_refs(root->fs_info,
7965                                                         extent_cache, rec);
7966                                 if (ret)
7967                                         goto repair_abort;
7968                                 fixed = 1;
7969                         }
7970                         err = 1;
7971                         cur_err = 1;
7972                 }
7973                 if (rec->bad_full_backref) {
7974                         fprintf(stderr, "bad full backref, on [%llu]\n",
7975                                 (unsigned long long)rec->start);
7976                         if (repair) {
7977                                 ret = fixup_extent_flags(root->fs_info, rec);
7978                                 if (ret)
7979                                         goto repair_abort;
7980                                 fixed = 1;
7981                         }
7982                         err = 1;
7983                         cur_err = 1;
7984                 }
7985                 /*
7986                  * Although it's not a extent ref's problem, we reuse this
7987                  * routine for error reporting.
7988                  * No repair function yet.
7989                  */
7990                 if (rec->crossing_stripes) {
7991                         fprintf(stderr,
7992                                 "bad metadata [%llu, %llu) crossing stripe boundary\n",
7993                                 rec->start, rec->start + rec->max_size);
7994                         err = 1;
7995                         cur_err = 1;
7996                 }
7997
7998                 if (rec->wrong_chunk_type) {
7999                         fprintf(stderr,
8000                                 "bad extent [%llu, %llu), type mismatch with chunk\n",
8001                                 rec->start, rec->start + rec->max_size);
8002                         err = 1;
8003                         cur_err = 1;
8004                 }
8005
8006                 remove_cache_extent(extent_cache, cache);
8007                 free_all_extent_backrefs(rec);
8008                 if (!init_extent_tree && repair && (!cur_err || fixed))
8009                         clear_extent_dirty(root->fs_info->excluded_extents,
8010                                            rec->start,
8011                                            rec->start + rec->max_size - 1,
8012                                            GFP_NOFS);
8013                 free(rec);
8014         }
8015 repair_abort:
8016         if (repair) {
8017                 if (ret && ret != -EAGAIN) {
8018                         fprintf(stderr, "failed to repair damaged filesystem, aborting\n");
8019                         exit(1);
8020                 } else if (!ret) {
8021                         struct btrfs_trans_handle *trans;
8022
8023                         root = root->fs_info->extent_root;
8024                         trans = btrfs_start_transaction(root, 1);
8025                         if (IS_ERR(trans)) {
8026                                 ret = PTR_ERR(trans);
8027                                 goto repair_abort;
8028                         }
8029
8030                         btrfs_fix_block_accounting(trans, root);
8031                         ret = btrfs_commit_transaction(trans, root);
8032                         if (ret)
8033                                 goto repair_abort;
8034                 }
8035                 if (err)
8036                         fprintf(stderr, "repaired damaged extent references\n");
8037                 return ret;
8038         }
8039         return err;
8040 }
8041
8042 u64 calc_stripe_length(u64 type, u64 length, int num_stripes)
8043 {
8044         u64 stripe_size;
8045
8046         if (type & BTRFS_BLOCK_GROUP_RAID0) {
8047                 stripe_size = length;
8048                 stripe_size /= num_stripes;
8049         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
8050                 stripe_size = length * 2;
8051                 stripe_size /= num_stripes;
8052         } else if (type & BTRFS_BLOCK_GROUP_RAID5) {
8053                 stripe_size = length;
8054                 stripe_size /= (num_stripes - 1);
8055         } else if (type & BTRFS_BLOCK_GROUP_RAID6) {
8056                 stripe_size = length;
8057                 stripe_size /= (num_stripes - 2);
8058         } else {
8059                 stripe_size = length;
8060         }
8061         return stripe_size;
8062 }
8063
8064 /*
8065  * Check the chunk with its block group/dev list ref:
8066  * Return 0 if all refs seems valid.
8067  * Return 1 if part of refs seems valid, need later check for rebuild ref
8068  * like missing block group and needs to search extent tree to rebuild them.
8069  * Return -1 if essential refs are missing and unable to rebuild.
8070  */
8071 static int check_chunk_refs(struct chunk_record *chunk_rec,
8072                             struct block_group_tree *block_group_cache,
8073                             struct device_extent_tree *dev_extent_cache,
8074                             int silent)
8075 {
8076         struct cache_extent *block_group_item;
8077         struct block_group_record *block_group_rec;
8078         struct cache_extent *dev_extent_item;
8079         struct device_extent_record *dev_extent_rec;
8080         u64 devid;
8081         u64 offset;
8082         u64 length;
8083         int metadump_v2 = 0;
8084         int i;
8085         int ret = 0;
8086
8087         block_group_item = lookup_cache_extent(&block_group_cache->tree,
8088                                                chunk_rec->offset,
8089                                                chunk_rec->length);
8090         if (block_group_item) {
8091                 block_group_rec = container_of(block_group_item,
8092                                                struct block_group_record,
8093                                                cache);
8094                 if (chunk_rec->length != block_group_rec->offset ||
8095                     chunk_rec->offset != block_group_rec->objectid ||
8096                     (!metadump_v2 &&
8097                      chunk_rec->type_flags != block_group_rec->flags)) {
8098                         if (!silent)
8099                                 fprintf(stderr,
8100                                         "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) mismatch with block group[%llu, %u, %llu]: offset(%llu), objectid(%llu), flags(%llu)\n",
8101                                         chunk_rec->objectid,
8102                                         chunk_rec->type,
8103                                         chunk_rec->offset,
8104                                         chunk_rec->length,
8105                                         chunk_rec->offset,
8106                                         chunk_rec->type_flags,
8107                                         block_group_rec->objectid,
8108                                         block_group_rec->type,
8109                                         block_group_rec->offset,
8110                                         block_group_rec->offset,
8111                                         block_group_rec->objectid,
8112                                         block_group_rec->flags);
8113                         ret = -1;
8114                 } else {
8115                         list_del_init(&block_group_rec->list);
8116                         chunk_rec->bg_rec = block_group_rec;
8117                 }
8118         } else {
8119                 if (!silent)
8120                         fprintf(stderr,
8121                                 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) is not found in block group\n",
8122                                 chunk_rec->objectid,
8123                                 chunk_rec->type,
8124                                 chunk_rec->offset,
8125                                 chunk_rec->length,
8126                                 chunk_rec->offset,
8127                                 chunk_rec->type_flags);
8128                 ret = 1;
8129         }
8130
8131         if (metadump_v2)
8132                 return ret;
8133
8134         length = calc_stripe_length(chunk_rec->type_flags, chunk_rec->length,
8135                                     chunk_rec->num_stripes);
8136         for (i = 0; i < chunk_rec->num_stripes; ++i) {
8137                 devid = chunk_rec->stripes[i].devid;
8138                 offset = chunk_rec->stripes[i].offset;
8139                 dev_extent_item = lookup_cache_extent2(&dev_extent_cache->tree,
8140                                                        devid, offset, length);
8141                 if (dev_extent_item) {
8142                         dev_extent_rec = container_of(dev_extent_item,
8143                                                 struct device_extent_record,
8144                                                 cache);
8145                         if (dev_extent_rec->objectid != devid ||
8146                             dev_extent_rec->offset != offset ||
8147                             dev_extent_rec->chunk_offset != chunk_rec->offset ||
8148                             dev_extent_rec->length != length) {
8149                                 if (!silent)
8150                                         fprintf(stderr,
8151                                                 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] dismatch dev extent[%llu, %llu, %llu]\n",
8152                                                 chunk_rec->objectid,
8153                                                 chunk_rec->type,
8154                                                 chunk_rec->offset,
8155                                                 chunk_rec->stripes[i].devid,
8156                                                 chunk_rec->stripes[i].offset,
8157                                                 dev_extent_rec->objectid,
8158                                                 dev_extent_rec->offset,
8159                                                 dev_extent_rec->length);
8160                                 ret = -1;
8161                         } else {
8162                                 list_move(&dev_extent_rec->chunk_list,
8163                                           &chunk_rec->dextents);
8164                         }
8165                 } else {
8166                         if (!silent)
8167                                 fprintf(stderr,
8168                                         "Chunk[%llu, %u, %llu] stripe[%llu, %llu] is not found in dev extent\n",
8169                                         chunk_rec->objectid,
8170                                         chunk_rec->type,
8171                                         chunk_rec->offset,
8172                                         chunk_rec->stripes[i].devid,
8173                                         chunk_rec->stripes[i].offset);
8174                         ret = -1;
8175                 }
8176         }
8177         return ret;
8178 }
8179
8180 /* check btrfs_chunk -> btrfs_dev_extent / btrfs_block_group_item */
8181 int check_chunks(struct cache_tree *chunk_cache,
8182                  struct block_group_tree *block_group_cache,
8183                  struct device_extent_tree *dev_extent_cache,
8184                  struct list_head *good, struct list_head *bad,
8185                  struct list_head *rebuild, int silent)
8186 {
8187         struct cache_extent *chunk_item;
8188         struct chunk_record *chunk_rec;
8189         struct block_group_record *bg_rec;
8190         struct device_extent_record *dext_rec;
8191         int err;
8192         int ret = 0;
8193
8194         chunk_item = first_cache_extent(chunk_cache);
8195         while (chunk_item) {
8196                 chunk_rec = container_of(chunk_item, struct chunk_record,
8197                                          cache);
8198                 err = check_chunk_refs(chunk_rec, block_group_cache,
8199                                        dev_extent_cache, silent);
8200                 if (err < 0)
8201                         ret = err;
8202                 if (err == 0 && good)
8203                         list_add_tail(&chunk_rec->list, good);
8204                 if (err > 0 && rebuild)
8205                         list_add_tail(&chunk_rec->list, rebuild);
8206                 if (err < 0 && bad)
8207                         list_add_tail(&chunk_rec->list, bad);
8208                 chunk_item = next_cache_extent(chunk_item);
8209         }
8210
8211         list_for_each_entry(bg_rec, &block_group_cache->block_groups, list) {
8212                 if (!silent)
8213                         fprintf(stderr,
8214                                 "Block group[%llu, %llu] (flags = %llu) didn't find the relative chunk.\n",
8215                                 bg_rec->objectid,
8216                                 bg_rec->offset,
8217                                 bg_rec->flags);
8218                 if (!ret)
8219                         ret = 1;
8220         }
8221
8222         list_for_each_entry(dext_rec, &dev_extent_cache->no_chunk_orphans,
8223                             chunk_list) {
8224                 if (!silent)
8225                         fprintf(stderr,
8226                                 "Device extent[%llu, %llu, %llu] didn't find the relative chunk.\n",
8227                                 dext_rec->objectid,
8228                                 dext_rec->offset,
8229                                 dext_rec->length);
8230                 if (!ret)
8231                         ret = 1;
8232         }
8233         return ret;
8234 }
8235
8236
8237 static int check_device_used(struct device_record *dev_rec,
8238                              struct device_extent_tree *dext_cache)
8239 {
8240         struct cache_extent *cache;
8241         struct device_extent_record *dev_extent_rec;
8242         u64 total_byte = 0;
8243
8244         cache = search_cache_extent2(&dext_cache->tree, dev_rec->devid, 0);
8245         while (cache) {
8246                 dev_extent_rec = container_of(cache,
8247                                               struct device_extent_record,
8248                                               cache);
8249                 if (dev_extent_rec->objectid != dev_rec->devid)
8250                         break;
8251
8252                 list_del_init(&dev_extent_rec->device_list);
8253                 total_byte += dev_extent_rec->length;
8254                 cache = next_cache_extent(cache);
8255         }
8256
8257         if (total_byte != dev_rec->byte_used) {
8258                 fprintf(stderr,
8259                         "Dev extent's total-byte(%llu) is not equal to byte-used(%llu) in dev[%llu, %u, %llu]\n",
8260                         total_byte, dev_rec->byte_used, dev_rec->objectid,
8261                         dev_rec->type, dev_rec->offset);
8262                 return -1;
8263         } else {
8264                 return 0;
8265         }
8266 }
8267
8268 /* check btrfs_dev_item -> btrfs_dev_extent */
8269 static int check_devices(struct rb_root *dev_cache,
8270                          struct device_extent_tree *dev_extent_cache)
8271 {
8272         struct rb_node *dev_node;
8273         struct device_record *dev_rec;
8274         struct device_extent_record *dext_rec;
8275         int err;
8276         int ret = 0;
8277
8278         dev_node = rb_first(dev_cache);
8279         while (dev_node) {
8280                 dev_rec = container_of(dev_node, struct device_record, node);
8281                 err = check_device_used(dev_rec, dev_extent_cache);
8282                 if (err)
8283                         ret = err;
8284
8285                 dev_node = rb_next(dev_node);
8286         }
8287         list_for_each_entry(dext_rec, &dev_extent_cache->no_device_orphans,
8288                             device_list) {
8289                 fprintf(stderr,
8290                         "Device extent[%llu, %llu, %llu] didn't find its device.\n",
8291                         dext_rec->objectid, dext_rec->offset, dext_rec->length);
8292                 if (!ret)
8293                         ret = 1;
8294         }
8295         return ret;
8296 }
8297
8298 static int add_root_item_to_list(struct list_head *head,
8299                                   u64 objectid, u64 bytenr, u64 last_snapshot,
8300                                   u8 level, u8 drop_level,
8301                                   int level_size, struct btrfs_key *drop_key)
8302 {
8303
8304         struct root_item_record *ri_rec;
8305         ri_rec = malloc(sizeof(*ri_rec));
8306         if (!ri_rec)
8307                 return -ENOMEM;
8308         ri_rec->bytenr = bytenr;
8309         ri_rec->objectid = objectid;
8310         ri_rec->level = level;
8311         ri_rec->level_size = level_size;
8312         ri_rec->drop_level = drop_level;
8313         ri_rec->last_snapshot = last_snapshot;
8314         if (drop_key)
8315                 memcpy(&ri_rec->drop_key, drop_key, sizeof(*drop_key));
8316         list_add_tail(&ri_rec->list, head);
8317
8318         return 0;
8319 }
8320
8321 static void free_root_item_list(struct list_head *list)
8322 {
8323         struct root_item_record *ri_rec;
8324
8325         while (!list_empty(list)) {
8326                 ri_rec = list_first_entry(list, struct root_item_record,
8327                                           list);
8328                 list_del_init(&ri_rec->list);
8329                 free(ri_rec);
8330         }
8331 }
8332
8333 static int deal_root_from_list(struct list_head *list,
8334                                struct btrfs_root *root,
8335                                struct block_info *bits,
8336                                int bits_nr,
8337                                struct cache_tree *pending,
8338                                struct cache_tree *seen,
8339                                struct cache_tree *reada,
8340                                struct cache_tree *nodes,
8341                                struct cache_tree *extent_cache,
8342                                struct cache_tree *chunk_cache,
8343                                struct rb_root *dev_cache,
8344                                struct block_group_tree *block_group_cache,
8345                                struct device_extent_tree *dev_extent_cache)
8346 {
8347         int ret = 0;
8348         u64 last;
8349
8350         while (!list_empty(list)) {
8351                 struct root_item_record *rec;
8352                 struct extent_buffer *buf;
8353                 rec = list_entry(list->next,
8354                                  struct root_item_record, list);
8355                 last = 0;
8356                 buf = read_tree_block(root->fs_info->tree_root,
8357                                       rec->bytenr, rec->level_size, 0);
8358                 if (!extent_buffer_uptodate(buf)) {
8359                         free_extent_buffer(buf);
8360                         ret = -EIO;
8361                         break;
8362                 }
8363                 ret = add_root_to_pending(buf, extent_cache, pending,
8364                                     seen, nodes, rec->objectid);
8365                 if (ret < 0)
8366                         break;
8367                 /*
8368                  * To rebuild extent tree, we need deal with snapshot
8369                  * one by one, otherwise we deal with node firstly which
8370                  * can maximize readahead.
8371                  */
8372                 while (1) {
8373                         ret = run_next_block(root, bits, bits_nr, &last,
8374                                              pending, seen, reada, nodes,
8375                                              extent_cache, chunk_cache,
8376                                              dev_cache, block_group_cache,
8377                                              dev_extent_cache, rec);
8378                         if (ret != 0)
8379                                 break;
8380                 }
8381                 free_extent_buffer(buf);
8382                 list_del(&rec->list);
8383                 free(rec);
8384                 if (ret < 0)
8385                         break;
8386         }
8387         while (ret >= 0) {
8388                 ret = run_next_block(root, bits, bits_nr, &last, pending, seen,
8389                                      reada, nodes, extent_cache, chunk_cache,
8390                                      dev_cache, block_group_cache,
8391                                      dev_extent_cache, NULL);
8392                 if (ret != 0) {
8393                         if (ret > 0)
8394                                 ret = 0;
8395                         break;
8396                 }
8397         }
8398         return ret;
8399 }
8400
8401 static int check_chunks_and_extents(struct btrfs_root *root)
8402 {
8403         struct rb_root dev_cache;
8404         struct cache_tree chunk_cache;
8405         struct block_group_tree block_group_cache;
8406         struct device_extent_tree dev_extent_cache;
8407         struct cache_tree extent_cache;
8408         struct cache_tree seen;
8409         struct cache_tree pending;
8410         struct cache_tree reada;
8411         struct cache_tree nodes;
8412         struct extent_io_tree excluded_extents;
8413         struct cache_tree corrupt_blocks;
8414         struct btrfs_path path;
8415         struct btrfs_key key;
8416         struct btrfs_key found_key;
8417         int ret, err = 0;
8418         struct block_info *bits;
8419         int bits_nr;
8420         struct extent_buffer *leaf;
8421         int slot;
8422         struct btrfs_root_item ri;
8423         struct list_head dropping_trees;
8424         struct list_head normal_trees;
8425         struct btrfs_root *root1;
8426         u64 objectid;
8427         u32 level_size;
8428         u8 level;
8429
8430         dev_cache = RB_ROOT;
8431         cache_tree_init(&chunk_cache);
8432         block_group_tree_init(&block_group_cache);
8433         device_extent_tree_init(&dev_extent_cache);
8434
8435         cache_tree_init(&extent_cache);
8436         cache_tree_init(&seen);
8437         cache_tree_init(&pending);
8438         cache_tree_init(&nodes);
8439         cache_tree_init(&reada);
8440         cache_tree_init(&corrupt_blocks);
8441         extent_io_tree_init(&excluded_extents);
8442         INIT_LIST_HEAD(&dropping_trees);
8443         INIT_LIST_HEAD(&normal_trees);
8444
8445         if (repair) {
8446                 root->fs_info->excluded_extents = &excluded_extents;
8447                 root->fs_info->fsck_extent_cache = &extent_cache;
8448                 root->fs_info->free_extent_hook = free_extent_hook;
8449                 root->fs_info->corrupt_blocks = &corrupt_blocks;
8450         }
8451
8452         bits_nr = 1024;
8453         bits = malloc(bits_nr * sizeof(struct block_info));
8454         if (!bits) {
8455                 perror("malloc");
8456                 exit(1);
8457         }
8458
8459         if (ctx.progress_enabled) {
8460                 ctx.tp = TASK_EXTENTS;
8461                 task_start(ctx.info);
8462         }
8463
8464 again:
8465         root1 = root->fs_info->tree_root;
8466         level = btrfs_header_level(root1->node);
8467         ret = add_root_item_to_list(&normal_trees, root1->root_key.objectid,
8468                                     root1->node->start, 0, level, 0,
8469                                     root1->nodesize, NULL);
8470         if (ret < 0)
8471                 goto out;
8472         root1 = root->fs_info->chunk_root;
8473         level = btrfs_header_level(root1->node);
8474         ret = add_root_item_to_list(&normal_trees, root1->root_key.objectid,
8475                                     root1->node->start, 0, level, 0,
8476                                     root1->nodesize, NULL);
8477         if (ret < 0)
8478                 goto out;
8479         btrfs_init_path(&path);
8480         key.offset = 0;
8481         key.objectid = 0;
8482         key.type = BTRFS_ROOT_ITEM_KEY;
8483         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
8484                                         &key, &path, 0, 0);
8485         if (ret < 0)
8486                 goto out;
8487         while(1) {
8488                 leaf = path.nodes[0];
8489                 slot = path.slots[0];
8490                 if (slot >= btrfs_header_nritems(path.nodes[0])) {
8491                         ret = btrfs_next_leaf(root, &path);
8492                         if (ret != 0)
8493                                 break;
8494                         leaf = path.nodes[0];
8495                         slot = path.slots[0];
8496                 }
8497                 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
8498                 if (found_key.type == BTRFS_ROOT_ITEM_KEY) {
8499                         unsigned long offset;
8500                         u64 last_snapshot;
8501
8502                         offset = btrfs_item_ptr_offset(leaf, path.slots[0]);
8503                         read_extent_buffer(leaf, &ri, offset, sizeof(ri));
8504                         last_snapshot = btrfs_root_last_snapshot(&ri);
8505                         if (btrfs_disk_key_objectid(&ri.drop_progress) == 0) {
8506                                 level = btrfs_root_level(&ri);
8507                                 level_size = root->nodesize;
8508                                 ret = add_root_item_to_list(&normal_trees,
8509                                                 found_key.objectid,
8510                                                 btrfs_root_bytenr(&ri),
8511                                                 last_snapshot, level,
8512                                                 0, level_size, NULL);
8513                                 if (ret < 0)
8514                                         goto out;
8515                         } else {
8516                                 level = btrfs_root_level(&ri);
8517                                 level_size = root->nodesize;
8518                                 objectid = found_key.objectid;
8519                                 btrfs_disk_key_to_cpu(&found_key,
8520                                                       &ri.drop_progress);
8521                                 ret = add_root_item_to_list(&dropping_trees,
8522                                                 objectid,
8523                                                 btrfs_root_bytenr(&ri),
8524                                                 last_snapshot, level,
8525                                                 ri.drop_level,
8526                                                 level_size, &found_key);
8527                                 if (ret < 0)
8528                                         goto out;
8529                         }
8530                 }
8531                 path.slots[0]++;
8532         }
8533         btrfs_release_path(&path);
8534
8535         /*
8536          * check_block can return -EAGAIN if it fixes something, please keep
8537          * this in mind when dealing with return values from these functions, if
8538          * we get -EAGAIN we want to fall through and restart the loop.
8539          */
8540         ret = deal_root_from_list(&normal_trees, root, bits, bits_nr, &pending,
8541                                   &seen, &reada, &nodes, &extent_cache,
8542                                   &chunk_cache, &dev_cache, &block_group_cache,
8543                                   &dev_extent_cache);
8544         if (ret < 0) {
8545                 if (ret == -EAGAIN)
8546                         goto loop;
8547                 goto out;
8548         }
8549         ret = deal_root_from_list(&dropping_trees, root, bits, bits_nr,
8550                                   &pending, &seen, &reada, &nodes,
8551                                   &extent_cache, &chunk_cache, &dev_cache,
8552                                   &block_group_cache, &dev_extent_cache);
8553         if (ret < 0) {
8554                 if (ret == -EAGAIN)
8555                         goto loop;
8556                 goto out;
8557         }
8558
8559         ret = check_chunks(&chunk_cache, &block_group_cache,
8560                            &dev_extent_cache, NULL, NULL, NULL, 0);
8561         if (ret) {
8562                 if (ret == -EAGAIN)
8563                         goto loop;
8564                 err = ret;
8565         }
8566
8567         ret = check_extent_refs(root, &extent_cache);
8568         if (ret < 0) {
8569                 if (ret == -EAGAIN)
8570                         goto loop;
8571                 goto out;
8572         }
8573
8574         ret = check_devices(&dev_cache, &dev_extent_cache);
8575         if (ret && err)
8576                 ret = err;
8577
8578 out:
8579         task_stop(ctx.info);
8580         if (repair) {
8581                 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
8582                 extent_io_tree_cleanup(&excluded_extents);
8583                 root->fs_info->fsck_extent_cache = NULL;
8584                 root->fs_info->free_extent_hook = NULL;
8585                 root->fs_info->corrupt_blocks = NULL;
8586                 root->fs_info->excluded_extents = NULL;
8587         }
8588         free(bits);
8589         free_chunk_cache_tree(&chunk_cache);
8590         free_device_cache_tree(&dev_cache);
8591         free_block_group_tree(&block_group_cache);
8592         free_device_extent_tree(&dev_extent_cache);
8593         free_extent_cache_tree(&seen);
8594         free_extent_cache_tree(&pending);
8595         free_extent_cache_tree(&reada);
8596         free_extent_cache_tree(&nodes);
8597         return ret;
8598 loop:
8599         free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
8600         free_extent_cache_tree(&seen);
8601         free_extent_cache_tree(&pending);
8602         free_extent_cache_tree(&reada);
8603         free_extent_cache_tree(&nodes);
8604         free_chunk_cache_tree(&chunk_cache);
8605         free_block_group_tree(&block_group_cache);
8606         free_device_cache_tree(&dev_cache);
8607         free_device_extent_tree(&dev_extent_cache);
8608         free_extent_record_cache(root->fs_info, &extent_cache);
8609         free_root_item_list(&normal_trees);
8610         free_root_item_list(&dropping_trees);
8611         extent_io_tree_cleanup(&excluded_extents);
8612         goto again;
8613 }
8614
8615 /*
8616  * Check backrefs of a tree block given by @bytenr or @eb.
8617  *
8618  * @root:       the root containing the @bytenr or @eb
8619  * @eb:         tree block extent buffer, can be NULL
8620  * @bytenr:     bytenr of the tree block to search
8621  * @level:      tree level of the tree block
8622  * @owner:      owner of the tree block
8623  *
8624  * Return >0 for any error found and output error message
8625  * Return 0 for no error found
8626  */
8627 static int check_tree_block_ref(struct btrfs_root *root,
8628                                 struct extent_buffer *eb, u64 bytenr,
8629                                 int level, u64 owner)
8630 {
8631         struct btrfs_key key;
8632         struct btrfs_root *extent_root = root->fs_info->extent_root;
8633         struct btrfs_path path;
8634         struct btrfs_extent_item *ei;
8635         struct btrfs_extent_inline_ref *iref;
8636         struct extent_buffer *leaf;
8637         unsigned long end;
8638         unsigned long ptr;
8639         int slot;
8640         int skinny_level;
8641         int type;
8642         u32 nodesize = root->nodesize;
8643         u32 item_size;
8644         u64 offset;
8645         int found_ref = 0;
8646         int err = 0;
8647         int ret;
8648
8649         btrfs_init_path(&path);
8650         key.objectid = bytenr;
8651         if (btrfs_fs_incompat(root->fs_info,
8652                               BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA))
8653                 key.type = BTRFS_METADATA_ITEM_KEY;
8654         else
8655                 key.type = BTRFS_EXTENT_ITEM_KEY;
8656         key.offset = (u64)-1;
8657
8658         /* Search for the backref in extent tree */
8659         ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
8660         if (ret < 0) {
8661                 err |= BACKREF_MISSING;
8662                 goto out;
8663         }
8664         ret = btrfs_previous_extent_item(extent_root, &path, bytenr);
8665         if (ret) {
8666                 err |= BACKREF_MISSING;
8667                 goto out;
8668         }
8669
8670         leaf = path.nodes[0];
8671         slot = path.slots[0];
8672         btrfs_item_key_to_cpu(leaf, &key, slot);
8673
8674         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
8675
8676         if (key.type == BTRFS_METADATA_ITEM_KEY) {
8677                 skinny_level = (int)key.offset;
8678                 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
8679         } else {
8680                 struct btrfs_tree_block_info *info;
8681
8682                 info = (struct btrfs_tree_block_info *)(ei + 1);
8683                 skinny_level = btrfs_tree_block_level(leaf, info);
8684                 iref = (struct btrfs_extent_inline_ref *)(info + 1);
8685         }
8686
8687         if (eb) {
8688                 u64 header_gen;
8689                 u64 extent_gen;
8690
8691                 if (!(btrfs_extent_flags(leaf, ei) &
8692                       BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
8693                         error(
8694                 "extent[%llu %u] backref type mismatch, missing bit: %llx",
8695                                 key.objectid, nodesize,
8696                                 BTRFS_EXTENT_FLAG_TREE_BLOCK);
8697                         err = BACKREF_MISMATCH;
8698                 }
8699                 header_gen = btrfs_header_generation(eb);
8700                 extent_gen = btrfs_extent_generation(leaf, ei);
8701                 if (header_gen != extent_gen) {
8702                         error(
8703         "extent[%llu %u] backref generation mismatch, wanted: %llu, have: %llu",
8704                                 key.objectid, nodesize, header_gen,
8705                                 extent_gen);
8706                         err = BACKREF_MISMATCH;
8707                 }
8708                 if (level != skinny_level) {
8709                         error(
8710                         "extent[%llu %u] level mismatch, wanted: %u, have: %u",
8711                                 key.objectid, nodesize, level, skinny_level);
8712                         err = BACKREF_MISMATCH;
8713                 }
8714                 if (!is_fstree(owner) && btrfs_extent_refs(leaf, ei) != 1) {
8715                         error(
8716                         "extent[%llu %u] is referred by other roots than %llu",
8717                                 key.objectid, nodesize, root->objectid);
8718                         err = BACKREF_MISMATCH;
8719                 }
8720         }
8721
8722         /*
8723          * Iterate the extent/metadata item to find the exact backref
8724          */
8725         item_size = btrfs_item_size_nr(leaf, slot);
8726         ptr = (unsigned long)iref;
8727         end = (unsigned long)ei + item_size;
8728         while (ptr < end) {
8729                 iref = (struct btrfs_extent_inline_ref *)ptr;
8730                 type = btrfs_extent_inline_ref_type(leaf, iref);
8731                 offset = btrfs_extent_inline_ref_offset(leaf, iref);
8732
8733                 if (type == BTRFS_TREE_BLOCK_REF_KEY &&
8734                         (offset == root->objectid || offset == owner)) {
8735                         found_ref = 1;
8736                 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
8737                         /* Check if the backref points to valid referencer */
8738                         found_ref = !check_tree_block_ref(root, NULL, offset,
8739                                                           level + 1, owner);
8740                 }
8741
8742                 if (found_ref)
8743                         break;
8744                 ptr += btrfs_extent_inline_ref_size(type);
8745         }
8746
8747         /*
8748          * Inlined extent item doesn't have what we need, check
8749          * TREE_BLOCK_REF_KEY
8750          */
8751         if (!found_ref) {
8752                 btrfs_release_path(&path);
8753                 key.objectid = bytenr;
8754                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
8755                 key.offset = root->objectid;
8756
8757                 ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
8758                 if (!ret)
8759                         found_ref = 1;
8760         }
8761         if (!found_ref)
8762                 err |= BACKREF_MISSING;
8763 out:
8764         btrfs_release_path(&path);
8765         if (eb && (err & BACKREF_MISSING))
8766                 error("extent[%llu %u] backref lost (owner: %llu, level: %u)",
8767                         bytenr, nodesize, owner, level);
8768         return err;
8769 }
8770
8771 /*
8772  * Check EXTENT_DATA item, mainly for its dbackref in extent tree
8773  *
8774  * Return >0 any error found and output error message
8775  * Return 0 for no error found
8776  */
8777 static int check_extent_data_item(struct btrfs_root *root,
8778                                   struct extent_buffer *eb, int slot)
8779 {
8780         struct btrfs_file_extent_item *fi;
8781         struct btrfs_path path;
8782         struct btrfs_root *extent_root = root->fs_info->extent_root;
8783         struct btrfs_key fi_key;
8784         struct btrfs_key dbref_key;
8785         struct extent_buffer *leaf;
8786         struct btrfs_extent_item *ei;
8787         struct btrfs_extent_inline_ref *iref;
8788         struct btrfs_extent_data_ref *dref;
8789         u64 owner;
8790         u64 file_extent_gen;
8791         u64 disk_bytenr;
8792         u64 disk_num_bytes;
8793         u64 extent_num_bytes;
8794         u64 extent_flags;
8795         u64 extent_gen;
8796         u32 item_size;
8797         unsigned long end;
8798         unsigned long ptr;
8799         int type;
8800         u64 ref_root;
8801         int found_dbackref = 0;
8802         int err = 0;
8803         int ret;
8804
8805         btrfs_item_key_to_cpu(eb, &fi_key, slot);
8806         fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
8807         file_extent_gen = btrfs_file_extent_generation(eb, fi);
8808
8809         /* Nothing to check for hole and inline data extents */
8810         if (btrfs_file_extent_type(eb, fi) == BTRFS_FILE_EXTENT_INLINE ||
8811             btrfs_file_extent_disk_bytenr(eb, fi) == 0)
8812                 return 0;
8813
8814         disk_bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
8815         disk_num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
8816         extent_num_bytes = btrfs_file_extent_num_bytes(eb, fi);
8817
8818         /* Check unaligned disk_num_bytes and num_bytes */
8819         if (!IS_ALIGNED(disk_num_bytes, root->sectorsize)) {
8820                 error(
8821 "file extent [%llu, %llu] has unaligned disk num bytes: %llu, should be aligned to %u",
8822                         fi_key.objectid, fi_key.offset, disk_num_bytes,
8823                         root->sectorsize);
8824                 err |= BYTES_UNALIGNED;
8825         } else {
8826                 data_bytes_allocated += disk_num_bytes;
8827         }
8828         if (!IS_ALIGNED(extent_num_bytes, root->sectorsize)) {
8829                 error(
8830 "file extent [%llu, %llu] has unaligned num bytes: %llu, should be aligned to %u",
8831                         fi_key.objectid, fi_key.offset, extent_num_bytes,
8832                         root->sectorsize);
8833                 err |= BYTES_UNALIGNED;
8834         } else {
8835                 data_bytes_referenced += extent_num_bytes;
8836         }
8837         owner = btrfs_header_owner(eb);
8838
8839         /* Check the extent item of the file extent in extent tree */
8840         btrfs_init_path(&path);
8841         dbref_key.objectid = btrfs_file_extent_disk_bytenr(eb, fi);
8842         dbref_key.type = BTRFS_EXTENT_ITEM_KEY;
8843         dbref_key.offset = btrfs_file_extent_disk_num_bytes(eb, fi);
8844
8845         ret = btrfs_search_slot(NULL, extent_root, &dbref_key, &path, 0, 0);
8846         if (ret) {
8847                 err |= BACKREF_MISSING;
8848                 goto error;
8849         }
8850
8851         leaf = path.nodes[0];
8852         slot = path.slots[0];
8853         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
8854
8855         extent_flags = btrfs_extent_flags(leaf, ei);
8856         extent_gen = btrfs_extent_generation(leaf, ei);
8857
8858         if (!(extent_flags & BTRFS_EXTENT_FLAG_DATA)) {
8859                 error(
8860                     "extent[%llu %llu] backref type mismatch, wanted bit: %llx",
8861                     disk_bytenr, disk_num_bytes,
8862                     BTRFS_EXTENT_FLAG_DATA);
8863                 err |= BACKREF_MISMATCH;
8864         }
8865
8866         if (file_extent_gen < extent_gen) {
8867                 error(
8868 "extent[%llu %llu] backref generation mismatch, wanted: <=%llu, have: %llu",
8869                         disk_bytenr, disk_num_bytes, file_extent_gen,
8870                         extent_gen);
8871                 err |= BACKREF_MISMATCH;
8872         }
8873
8874         /* Check data backref inside that extent item */
8875         item_size = btrfs_item_size_nr(leaf, path.slots[0]);
8876         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
8877         ptr = (unsigned long)iref;
8878         end = (unsigned long)ei + item_size;
8879         while (ptr < end) {
8880                 iref = (struct btrfs_extent_inline_ref *)ptr;
8881                 type = btrfs_extent_inline_ref_type(leaf, iref);
8882                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
8883
8884                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
8885                         ref_root = btrfs_extent_data_ref_root(leaf, dref);
8886                         if (ref_root == owner || ref_root == root->objectid)
8887                                 found_dbackref = 1;
8888                 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
8889                         found_dbackref = !check_tree_block_ref(root, NULL,
8890                                 btrfs_extent_inline_ref_offset(leaf, iref),
8891                                 0, owner);
8892                 }
8893
8894                 if (found_dbackref)
8895                         break;
8896                 ptr += btrfs_extent_inline_ref_size(type);
8897         }
8898
8899         /* Didn't found inlined data backref, try EXTENT_DATA_REF_KEY */
8900         if (!found_dbackref) {
8901                 btrfs_release_path(&path);
8902
8903                 btrfs_init_path(&path);
8904                 dbref_key.objectid = btrfs_file_extent_disk_bytenr(eb, fi);
8905                 dbref_key.type = BTRFS_EXTENT_DATA_REF_KEY;
8906                 dbref_key.offset = hash_extent_data_ref(root->objectid,
8907                                 fi_key.objectid, fi_key.offset);
8908
8909                 ret = btrfs_search_slot(NULL, root->fs_info->extent_root,
8910                                         &dbref_key, &path, 0, 0);
8911                 if (!ret)
8912                         found_dbackref = 1;
8913         }
8914
8915         if (!found_dbackref)
8916                 err |= BACKREF_MISSING;
8917 error:
8918         btrfs_release_path(&path);
8919         if (err & BACKREF_MISSING) {
8920                 error("data extent[%llu %llu] backref lost",
8921                       disk_bytenr, disk_num_bytes);
8922         }
8923         return err;
8924 }
8925
8926 /*
8927  * Get real tree block level for the case like shared block
8928  * Return >= 0 as tree level
8929  * Return <0 for error
8930  */
8931 static int query_tree_block_level(struct btrfs_fs_info *fs_info, u64 bytenr)
8932 {
8933         struct extent_buffer *eb;
8934         struct btrfs_path path;
8935         struct btrfs_key key;
8936         struct btrfs_extent_item *ei;
8937         u64 flags;
8938         u64 transid;
8939         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
8940         u8 backref_level;
8941         u8 header_level;
8942         int ret;
8943
8944         /* Search extent tree for extent generation and level */
8945         key.objectid = bytenr;
8946         key.type = BTRFS_METADATA_ITEM_KEY;
8947         key.offset = (u64)-1;
8948
8949         btrfs_init_path(&path);
8950         ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, &path, 0, 0);
8951         if (ret < 0)
8952                 goto release_out;
8953         ret = btrfs_previous_extent_item(fs_info->extent_root, &path, bytenr);
8954         if (ret < 0)
8955                 goto release_out;
8956         if (ret > 0) {
8957                 ret = -ENOENT;
8958                 goto release_out;
8959         }
8960
8961         btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
8962         ei = btrfs_item_ptr(path.nodes[0], path.slots[0],
8963                             struct btrfs_extent_item);
8964         flags = btrfs_extent_flags(path.nodes[0], ei);
8965         if (!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
8966                 ret = -ENOENT;
8967                 goto release_out;
8968         }
8969
8970         /* Get transid for later read_tree_block() check */
8971         transid = btrfs_extent_generation(path.nodes[0], ei);
8972
8973         /* Get backref level as one source */
8974         if (key.type == BTRFS_METADATA_ITEM_KEY) {
8975                 backref_level = key.offset;
8976         } else {
8977                 struct btrfs_tree_block_info *info;
8978
8979                 info = (struct btrfs_tree_block_info *)(ei + 1);
8980                 backref_level = btrfs_tree_block_level(path.nodes[0], info);
8981         }
8982         btrfs_release_path(&path);
8983
8984         /* Get level from tree block as an alternative source */
8985         eb = read_tree_block_fs_info(fs_info, bytenr, nodesize, transid);
8986         if (!extent_buffer_uptodate(eb)) {
8987                 free_extent_buffer(eb);
8988                 return -EIO;
8989         }
8990         header_level = btrfs_header_level(eb);
8991         free_extent_buffer(eb);
8992
8993         if (header_level != backref_level)
8994                 return -EIO;
8995         return header_level;
8996
8997 release_out:
8998         btrfs_release_path(&path);
8999         return ret;
9000 }
9001
9002 /*
9003  * Check if a tree block backref is valid (points to a valid tree block)
9004  * if level == -1, level will be resolved
9005  * Return >0 for any error found and print error message
9006  */
9007 static int check_tree_block_backref(struct btrfs_fs_info *fs_info, u64 root_id,
9008                                     u64 bytenr, int level)
9009 {
9010         struct btrfs_root *root;
9011         struct btrfs_key key;
9012         struct btrfs_path path;
9013         struct extent_buffer *eb;
9014         struct extent_buffer *node;
9015         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
9016         int err = 0;
9017         int ret;
9018
9019         /* Query level for level == -1 special case */
9020         if (level == -1)
9021                 level = query_tree_block_level(fs_info, bytenr);
9022         if (level < 0) {
9023                 err |= REFERENCER_MISSING;
9024                 goto out;
9025         }
9026
9027         key.objectid = root_id;
9028         key.type = BTRFS_ROOT_ITEM_KEY;
9029         key.offset = (u64)-1;
9030
9031         root = btrfs_read_fs_root(fs_info, &key);
9032         if (IS_ERR(root)) {
9033                 err |= REFERENCER_MISSING;
9034                 goto out;
9035         }
9036
9037         /* Read out the tree block to get item/node key */
9038         eb = read_tree_block(root, bytenr, root->nodesize, 0);
9039         if (!extent_buffer_uptodate(eb)) {
9040                 err |= REFERENCER_MISSING;
9041                 free_extent_buffer(eb);
9042                 goto out;
9043         }
9044
9045         /* Empty tree, no need to check key */
9046         if (!btrfs_header_nritems(eb) && !level) {
9047                 free_extent_buffer(eb);
9048                 goto out;
9049         }
9050
9051         if (level)
9052                 btrfs_node_key_to_cpu(eb, &key, 0);
9053         else
9054                 btrfs_item_key_to_cpu(eb, &key, 0);
9055
9056         free_extent_buffer(eb);
9057
9058         btrfs_init_path(&path);
9059         path.lowest_level = level;
9060         /* Search with the first key, to ensure we can reach it */
9061         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
9062         if (ret < 0) {
9063                 err |= REFERENCER_MISSING;
9064                 goto release_out;
9065         }
9066
9067         node = path.nodes[level];
9068         if (btrfs_header_bytenr(node) != bytenr) {
9069                 error(
9070         "extent [%llu %d] referencer bytenr mismatch, wanted: %llu, have: %llu",
9071                         bytenr, nodesize, bytenr,
9072                         btrfs_header_bytenr(node));
9073                 err |= REFERENCER_MISMATCH;
9074         }
9075         if (btrfs_header_level(node) != level) {
9076                 error(
9077         "extent [%llu %d] referencer level mismatch, wanted: %d, have: %d",
9078                         bytenr, nodesize, level,
9079                         btrfs_header_level(node));
9080                 err |= REFERENCER_MISMATCH;
9081         }
9082
9083 release_out:
9084         btrfs_release_path(&path);
9085 out:
9086         if (err & REFERENCER_MISSING) {
9087                 if (level < 0)
9088                         error("extent [%llu %d] lost referencer (owner: %llu)",
9089                                 bytenr, nodesize, root_id);
9090                 else
9091                         error(
9092                 "extent [%llu %d] lost referencer (owner: %llu, level: %u)",
9093                                 bytenr, nodesize, root_id, level);
9094         }
9095
9096         return err;
9097 }
9098
9099 /*
9100  * Check referencer for shared block backref
9101  * If level == -1, this function will resolve the level.
9102  */
9103 static int check_shared_block_backref(struct btrfs_fs_info *fs_info,
9104                                      u64 parent, u64 bytenr, int level)
9105 {
9106         struct extent_buffer *eb;
9107         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
9108         u32 nr;
9109         int found_parent = 0;
9110         int i;
9111
9112         eb = read_tree_block_fs_info(fs_info, parent, nodesize, 0);
9113         if (!extent_buffer_uptodate(eb))
9114                 goto out;
9115
9116         if (level == -1)
9117                 level = query_tree_block_level(fs_info, bytenr);
9118         if (level < 0)
9119                 goto out;
9120
9121         if (level + 1 != btrfs_header_level(eb))
9122                 goto out;
9123
9124         nr = btrfs_header_nritems(eb);
9125         for (i = 0; i < nr; i++) {
9126                 if (bytenr == btrfs_node_blockptr(eb, i)) {
9127                         found_parent = 1;
9128                         break;
9129                 }
9130         }
9131 out:
9132         free_extent_buffer(eb);
9133         if (!found_parent) {
9134                 error(
9135         "shared extent[%llu %u] lost its parent (parent: %llu, level: %u)",
9136                         bytenr, nodesize, parent, level);
9137                 return REFERENCER_MISSING;
9138         }
9139         return 0;
9140 }
9141
9142 /*
9143  * Check referencer for normal (inlined) data ref
9144  * If len == 0, it will be resolved by searching in extent tree
9145  */
9146 static int check_extent_data_backref(struct btrfs_fs_info *fs_info,
9147                                      u64 root_id, u64 objectid, u64 offset,
9148                                      u64 bytenr, u64 len, u32 count)
9149 {
9150         struct btrfs_root *root;
9151         struct btrfs_root *extent_root = fs_info->extent_root;
9152         struct btrfs_key key;
9153         struct btrfs_path path;
9154         struct extent_buffer *leaf;
9155         struct btrfs_file_extent_item *fi;
9156         u32 found_count = 0;
9157         int slot;
9158         int ret = 0;
9159
9160         if (!len) {
9161                 key.objectid = bytenr;
9162                 key.type = BTRFS_EXTENT_ITEM_KEY;
9163                 key.offset = (u64)-1;
9164
9165                 btrfs_init_path(&path);
9166                 ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
9167                 if (ret < 0)
9168                         goto out;
9169                 ret = btrfs_previous_extent_item(extent_root, &path, bytenr);
9170                 if (ret)
9171                         goto out;
9172                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
9173                 if (key.objectid != bytenr ||
9174                     key.type != BTRFS_EXTENT_ITEM_KEY)
9175                         goto out;
9176                 len = key.offset;
9177                 btrfs_release_path(&path);
9178         }
9179         key.objectid = root_id;
9180         key.type = BTRFS_ROOT_ITEM_KEY;
9181         key.offset = (u64)-1;
9182         btrfs_init_path(&path);
9183
9184         root = btrfs_read_fs_root(fs_info, &key);
9185         if (IS_ERR(root))
9186                 goto out;
9187
9188         key.objectid = objectid;
9189         key.type = BTRFS_EXTENT_DATA_KEY;
9190         /*
9191          * It can be nasty as data backref offset is
9192          * file offset - file extent offset, which is smaller or
9193          * equal to original backref offset.  The only special case is
9194          * overflow.  So we need to special check and do further search.
9195          */
9196         key.offset = offset & (1ULL << 63) ? 0 : offset;
9197
9198         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
9199         if (ret < 0)
9200                 goto out;
9201
9202         /*
9203          * Search afterwards to get correct one
9204          * NOTE: As we must do a comprehensive check on the data backref to
9205          * make sure the dref count also matches, we must iterate all file
9206          * extents for that inode.
9207          */
9208         while (1) {
9209                 leaf = path.nodes[0];
9210                 slot = path.slots[0];
9211
9212                 btrfs_item_key_to_cpu(leaf, &key, slot);
9213                 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
9214                         break;
9215                 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
9216                 /*
9217                  * Except normal disk bytenr and disk num bytes, we still
9218                  * need to do extra check on dbackref offset as
9219                  * dbackref offset = file_offset - file_extent_offset
9220                  */
9221                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == bytenr &&
9222                     btrfs_file_extent_disk_num_bytes(leaf, fi) == len &&
9223                     (u64)(key.offset - btrfs_file_extent_offset(leaf, fi)) ==
9224                     offset)
9225                         found_count++;
9226
9227                 ret = btrfs_next_item(root, &path);
9228                 if (ret)
9229                         break;
9230         }
9231 out:
9232         btrfs_release_path(&path);
9233         if (found_count != count) {
9234                 error(
9235 "extent[%llu, %llu] referencer count mismatch (root: %llu, owner: %llu, offset: %llu) wanted: %u, have: %u",
9236                         bytenr, len, root_id, objectid, offset, count, found_count);
9237                 return REFERENCER_MISSING;
9238         }
9239         return 0;
9240 }
9241
9242 /*
9243  * Check if the referencer of a shared data backref exists
9244  */
9245 static int check_shared_data_backref(struct btrfs_fs_info *fs_info,
9246                                      u64 parent, u64 bytenr)
9247 {
9248         struct extent_buffer *eb;
9249         struct btrfs_key key;
9250         struct btrfs_file_extent_item *fi;
9251         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
9252         u32 nr;
9253         int found_parent = 0;
9254         int i;
9255
9256         eb = read_tree_block_fs_info(fs_info, parent, nodesize, 0);
9257         if (!extent_buffer_uptodate(eb))
9258                 goto out;
9259
9260         nr = btrfs_header_nritems(eb);
9261         for (i = 0; i < nr; i++) {
9262                 btrfs_item_key_to_cpu(eb, &key, i);
9263                 if (key.type != BTRFS_EXTENT_DATA_KEY)
9264                         continue;
9265
9266                 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
9267                 if (btrfs_file_extent_type(eb, fi) == BTRFS_FILE_EXTENT_INLINE)
9268                         continue;
9269
9270                 if (btrfs_file_extent_disk_bytenr(eb, fi) == bytenr) {
9271                         found_parent = 1;
9272                         break;
9273                 }
9274         }
9275
9276 out:
9277         free_extent_buffer(eb);
9278         if (!found_parent) {
9279                 error("shared extent %llu referencer lost (parent: %llu)",
9280                         bytenr, parent);
9281                 return REFERENCER_MISSING;
9282         }
9283         return 0;
9284 }
9285
9286 /*
9287  * This function will check a given extent item, including its backref and
9288  * itself (like crossing stripe boundary and type)
9289  *
9290  * Since we don't use extent_record anymore, introduce new error bit
9291  */
9292 static int check_extent_item(struct btrfs_fs_info *fs_info,
9293                              struct extent_buffer *eb, int slot)
9294 {
9295         struct btrfs_extent_item *ei;
9296         struct btrfs_extent_inline_ref *iref;
9297         struct btrfs_extent_data_ref *dref;
9298         unsigned long end;
9299         unsigned long ptr;
9300         int type;
9301         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
9302         u32 item_size = btrfs_item_size_nr(eb, slot);
9303         u64 flags;
9304         u64 offset;
9305         int metadata = 0;
9306         int level;
9307         struct btrfs_key key;
9308         int ret;
9309         int err = 0;
9310
9311         btrfs_item_key_to_cpu(eb, &key, slot);
9312         if (key.type == BTRFS_EXTENT_ITEM_KEY)
9313                 bytes_used += key.offset;
9314         else
9315                 bytes_used += nodesize;
9316
9317         if (item_size < sizeof(*ei)) {
9318                 /*
9319                  * COMPAT_EXTENT_TREE_V0 case, but it's already a super
9320                  * old thing when on disk format is still un-determined.
9321                  * No need to care about it anymore
9322                  */
9323                 error("unsupported COMPAT_EXTENT_TREE_V0 detected");
9324                 return -ENOTTY;
9325         }
9326
9327         ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
9328         flags = btrfs_extent_flags(eb, ei);
9329
9330         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
9331                 metadata = 1;
9332         if (metadata && check_crossing_stripes(global_info, key.objectid,
9333                                                eb->len)) {
9334                 error("bad metadata [%llu, %llu) crossing stripe boundary",
9335                       key.objectid, key.objectid + nodesize);
9336                 err |= CROSSING_STRIPE_BOUNDARY;
9337         }
9338
9339         ptr = (unsigned long)(ei + 1);
9340
9341         if (metadata && key.type == BTRFS_EXTENT_ITEM_KEY) {
9342                 /* Old EXTENT_ITEM metadata */
9343                 struct btrfs_tree_block_info *info;
9344
9345                 info = (struct btrfs_tree_block_info *)ptr;
9346                 level = btrfs_tree_block_level(eb, info);
9347                 ptr += sizeof(struct btrfs_tree_block_info);
9348         } else {
9349                 /* New METADATA_ITEM */
9350                 level = key.offset;
9351         }
9352         end = (unsigned long)ei + item_size;
9353
9354         if (ptr >= end) {
9355                 err |= ITEM_SIZE_MISMATCH;
9356                 goto out;
9357         }
9358
9359         /* Now check every backref in this extent item */
9360 next:
9361         iref = (struct btrfs_extent_inline_ref *)ptr;
9362         type = btrfs_extent_inline_ref_type(eb, iref);
9363         offset = btrfs_extent_inline_ref_offset(eb, iref);
9364         switch (type) {
9365         case BTRFS_TREE_BLOCK_REF_KEY:
9366                 ret = check_tree_block_backref(fs_info, offset, key.objectid,
9367                                                level);
9368                 err |= ret;
9369                 break;
9370         case BTRFS_SHARED_BLOCK_REF_KEY:
9371                 ret = check_shared_block_backref(fs_info, offset, key.objectid,
9372                                                  level);
9373                 err |= ret;
9374                 break;
9375         case BTRFS_EXTENT_DATA_REF_KEY:
9376                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
9377                 ret = check_extent_data_backref(fs_info,
9378                                 btrfs_extent_data_ref_root(eb, dref),
9379                                 btrfs_extent_data_ref_objectid(eb, dref),
9380                                 btrfs_extent_data_ref_offset(eb, dref),
9381                                 key.objectid, key.offset,
9382                                 btrfs_extent_data_ref_count(eb, dref));
9383                 err |= ret;
9384                 break;
9385         case BTRFS_SHARED_DATA_REF_KEY:
9386                 ret = check_shared_data_backref(fs_info, offset, key.objectid);
9387                 err |= ret;
9388                 break;
9389         default:
9390                 error("extent[%llu %d %llu] has unknown ref type: %d",
9391                         key.objectid, key.type, key.offset, type);
9392                 err |= UNKNOWN_TYPE;
9393                 goto out;
9394         }
9395
9396         ptr += btrfs_extent_inline_ref_size(type);
9397         if (ptr < end)
9398                 goto next;
9399
9400 out:
9401         return err;
9402 }
9403
9404 /*
9405  * Check if a dev extent item is referred correctly by its chunk
9406  */
9407 static int check_dev_extent_item(struct btrfs_fs_info *fs_info,
9408                                  struct extent_buffer *eb, int slot)
9409 {
9410         struct btrfs_root *chunk_root = fs_info->chunk_root;
9411         struct btrfs_dev_extent *ptr;
9412         struct btrfs_path path;
9413         struct btrfs_key chunk_key;
9414         struct btrfs_key devext_key;
9415         struct btrfs_chunk *chunk;
9416         struct extent_buffer *l;
9417         int num_stripes;
9418         u64 length;
9419         int i;
9420         int found_chunk = 0;
9421         int ret;
9422
9423         btrfs_item_key_to_cpu(eb, &devext_key, slot);
9424         ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_extent);
9425         length = btrfs_dev_extent_length(eb, ptr);
9426
9427         chunk_key.objectid = btrfs_dev_extent_chunk_objectid(eb, ptr);
9428         chunk_key.type = BTRFS_CHUNK_ITEM_KEY;
9429         chunk_key.offset = btrfs_dev_extent_chunk_offset(eb, ptr);
9430
9431         btrfs_init_path(&path);
9432         ret = btrfs_search_slot(NULL, chunk_root, &chunk_key, &path, 0, 0);
9433         if (ret)
9434                 goto out;
9435
9436         l = path.nodes[0];
9437         chunk = btrfs_item_ptr(l, path.slots[0], struct btrfs_chunk);
9438         if (btrfs_chunk_length(l, chunk) != length)
9439                 goto out;
9440
9441         num_stripes = btrfs_chunk_num_stripes(l, chunk);
9442         for (i = 0; i < num_stripes; i++) {
9443                 u64 devid = btrfs_stripe_devid_nr(l, chunk, i);
9444                 u64 offset = btrfs_stripe_offset_nr(l, chunk, i);
9445
9446                 if (devid == devext_key.objectid &&
9447                     offset == devext_key.offset) {
9448                         found_chunk = 1;
9449                         break;
9450                 }
9451         }
9452 out:
9453         btrfs_release_path(&path);
9454         if (!found_chunk) {
9455                 error(
9456                 "device extent[%llu, %llu, %llu] did not find the related chunk",
9457                         devext_key.objectid, devext_key.offset, length);
9458                 return REFERENCER_MISSING;
9459         }
9460         return 0;
9461 }
9462
9463 /*
9464  * Check if the used space is correct with the dev item
9465  */
9466 static int check_dev_item(struct btrfs_fs_info *fs_info,
9467                           struct extent_buffer *eb, int slot)
9468 {
9469         struct btrfs_root *dev_root = fs_info->dev_root;
9470         struct btrfs_dev_item *dev_item;
9471         struct btrfs_path path;
9472         struct btrfs_key key;
9473         struct btrfs_dev_extent *ptr;
9474         u64 dev_id;
9475         u64 used;
9476         u64 total = 0;
9477         int ret;
9478
9479         dev_item = btrfs_item_ptr(eb, slot, struct btrfs_dev_item);
9480         dev_id = btrfs_device_id(eb, dev_item);
9481         used = btrfs_device_bytes_used(eb, dev_item);
9482
9483         key.objectid = dev_id;
9484         key.type = BTRFS_DEV_EXTENT_KEY;
9485         key.offset = 0;
9486
9487         btrfs_init_path(&path);
9488         ret = btrfs_search_slot(NULL, dev_root, &key, &path, 0, 0);
9489         if (ret < 0) {
9490                 btrfs_item_key_to_cpu(eb, &key, slot);
9491                 error("cannot find any related dev extent for dev[%llu, %u, %llu]",
9492                         key.objectid, key.type, key.offset);
9493                 btrfs_release_path(&path);
9494                 return REFERENCER_MISSING;
9495         }
9496
9497         /* Iterate dev_extents to calculate the used space of a device */
9498         while (1) {
9499                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
9500
9501                 if (key.objectid > dev_id)
9502                         break;
9503                 if (key.type != BTRFS_DEV_EXTENT_KEY || key.objectid != dev_id)
9504                         goto next;
9505
9506                 ptr = btrfs_item_ptr(path.nodes[0], path.slots[0],
9507                                      struct btrfs_dev_extent);
9508                 total += btrfs_dev_extent_length(path.nodes[0], ptr);
9509 next:
9510                 ret = btrfs_next_item(dev_root, &path);
9511                 if (ret)
9512                         break;
9513         }
9514         btrfs_release_path(&path);
9515
9516         if (used != total) {
9517                 btrfs_item_key_to_cpu(eb, &key, slot);
9518                 error(
9519 "Dev extent's total-byte %llu is not equal to bytes-used %llu in dev[%llu, %u, %llu]",
9520                         total, used, BTRFS_ROOT_TREE_OBJECTID,
9521                         BTRFS_DEV_EXTENT_KEY, dev_id);
9522                 return ACCOUNTING_MISMATCH;
9523         }
9524         return 0;
9525 }
9526
9527 /*
9528  * Check a block group item with its referener (chunk) and its used space
9529  * with extent/metadata item
9530  */
9531 static int check_block_group_item(struct btrfs_fs_info *fs_info,
9532                                   struct extent_buffer *eb, int slot)
9533 {
9534         struct btrfs_root *extent_root = fs_info->extent_root;
9535         struct btrfs_root *chunk_root = fs_info->chunk_root;
9536         struct btrfs_block_group_item *bi;
9537         struct btrfs_block_group_item bg_item;
9538         struct btrfs_path path;
9539         struct btrfs_key bg_key;
9540         struct btrfs_key chunk_key;
9541         struct btrfs_key extent_key;
9542         struct btrfs_chunk *chunk;
9543         struct extent_buffer *leaf;
9544         struct btrfs_extent_item *ei;
9545         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
9546         u64 flags;
9547         u64 bg_flags;
9548         u64 used;
9549         u64 total = 0;
9550         int ret;
9551         int err = 0;
9552
9553         btrfs_item_key_to_cpu(eb, &bg_key, slot);
9554         bi = btrfs_item_ptr(eb, slot, struct btrfs_block_group_item);
9555         read_extent_buffer(eb, &bg_item, (unsigned long)bi, sizeof(bg_item));
9556         used = btrfs_block_group_used(&bg_item);
9557         bg_flags = btrfs_block_group_flags(&bg_item);
9558
9559         chunk_key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
9560         chunk_key.type = BTRFS_CHUNK_ITEM_KEY;
9561         chunk_key.offset = bg_key.objectid;
9562
9563         btrfs_init_path(&path);
9564         /* Search for the referencer chunk */
9565         ret = btrfs_search_slot(NULL, chunk_root, &chunk_key, &path, 0, 0);
9566         if (ret) {
9567                 error(
9568                 "block group[%llu %llu] did not find the related chunk item",
9569                         bg_key.objectid, bg_key.offset);
9570                 err |= REFERENCER_MISSING;
9571         } else {
9572                 chunk = btrfs_item_ptr(path.nodes[0], path.slots[0],
9573                                         struct btrfs_chunk);
9574                 if (btrfs_chunk_length(path.nodes[0], chunk) !=
9575                                                 bg_key.offset) {
9576                         error(
9577         "block group[%llu %llu] related chunk item length does not match",
9578                                 bg_key.objectid, bg_key.offset);
9579                         err |= REFERENCER_MISMATCH;
9580                 }
9581         }
9582         btrfs_release_path(&path);
9583
9584         /* Search from the block group bytenr */
9585         extent_key.objectid = bg_key.objectid;
9586         extent_key.type = 0;
9587         extent_key.offset = 0;
9588
9589         btrfs_init_path(&path);
9590         ret = btrfs_search_slot(NULL, extent_root, &extent_key, &path, 0, 0);
9591         if (ret < 0)
9592                 goto out;
9593
9594         /* Iterate extent tree to account used space */
9595         while (1) {
9596                 leaf = path.nodes[0];
9597                 btrfs_item_key_to_cpu(leaf, &extent_key, path.slots[0]);
9598                 if (extent_key.objectid >= bg_key.objectid + bg_key.offset)
9599                         break;
9600
9601                 if (extent_key.type != BTRFS_METADATA_ITEM_KEY &&
9602                     extent_key.type != BTRFS_EXTENT_ITEM_KEY)
9603                         goto next;
9604                 if (extent_key.objectid < bg_key.objectid)
9605                         goto next;
9606
9607                 if (extent_key.type == BTRFS_METADATA_ITEM_KEY)
9608                         total += nodesize;
9609                 else
9610                         total += extent_key.offset;
9611
9612                 ei = btrfs_item_ptr(leaf, path.slots[0],
9613                                     struct btrfs_extent_item);
9614                 flags = btrfs_extent_flags(leaf, ei);
9615                 if (flags & BTRFS_EXTENT_FLAG_DATA) {
9616                         if (!(bg_flags & BTRFS_BLOCK_GROUP_DATA)) {
9617                                 error(
9618                         "bad extent[%llu, %llu) type mismatch with chunk",
9619                                         extent_key.objectid,
9620                                         extent_key.objectid + extent_key.offset);
9621                                 err |= CHUNK_TYPE_MISMATCH;
9622                         }
9623                 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
9624                         if (!(bg_flags & (BTRFS_BLOCK_GROUP_SYSTEM |
9625                                     BTRFS_BLOCK_GROUP_METADATA))) {
9626                                 error(
9627                         "bad extent[%llu, %llu) type mismatch with chunk",
9628                                         extent_key.objectid,
9629                                         extent_key.objectid + nodesize);
9630                                 err |= CHUNK_TYPE_MISMATCH;
9631                         }
9632                 }
9633 next:
9634                 ret = btrfs_next_item(extent_root, &path);
9635                 if (ret)
9636                         break;
9637         }
9638
9639 out:
9640         btrfs_release_path(&path);
9641
9642         if (total != used) {
9643                 error(
9644                 "block group[%llu %llu] used %llu but extent items used %llu",
9645                         bg_key.objectid, bg_key.offset, used, total);
9646                 err |= ACCOUNTING_MISMATCH;
9647         }
9648         return err;
9649 }
9650
9651 /*
9652  * Check a chunk item.
9653  * Including checking all referred dev_extents and block group
9654  */
9655 static int check_chunk_item(struct btrfs_fs_info *fs_info,
9656                             struct extent_buffer *eb, int slot)
9657 {
9658         struct btrfs_root *extent_root = fs_info->extent_root;
9659         struct btrfs_root *dev_root = fs_info->dev_root;
9660         struct btrfs_path path;
9661         struct btrfs_key chunk_key;
9662         struct btrfs_key bg_key;
9663         struct btrfs_key devext_key;
9664         struct btrfs_chunk *chunk;
9665         struct extent_buffer *leaf;
9666         struct btrfs_block_group_item *bi;
9667         struct btrfs_block_group_item bg_item;
9668         struct btrfs_dev_extent *ptr;
9669         u32 sectorsize = btrfs_super_sectorsize(fs_info->super_copy);
9670         u64 length;
9671         u64 chunk_end;
9672         u64 type;
9673         u64 profile;
9674         int num_stripes;
9675         u64 offset;
9676         u64 objectid;
9677         int i;
9678         int ret;
9679         int err = 0;
9680
9681         btrfs_item_key_to_cpu(eb, &chunk_key, slot);
9682         chunk = btrfs_item_ptr(eb, slot, struct btrfs_chunk);
9683         length = btrfs_chunk_length(eb, chunk);
9684         chunk_end = chunk_key.offset + length;
9685         if (!IS_ALIGNED(length, sectorsize)) {
9686                 error("chunk[%llu %llu) not aligned to %u",
9687                         chunk_key.offset, chunk_end, sectorsize);
9688                 err |= BYTES_UNALIGNED;
9689                 goto out;
9690         }
9691
9692         type = btrfs_chunk_type(eb, chunk);
9693         profile = type & BTRFS_BLOCK_GROUP_PROFILE_MASK;
9694         if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
9695                 error("chunk[%llu %llu) has no chunk type",
9696                         chunk_key.offset, chunk_end);
9697                 err |= UNKNOWN_TYPE;
9698         }
9699         if (profile && (profile & (profile - 1))) {
9700                 error("chunk[%llu %llu) multiple profiles detected: %llx",
9701                         chunk_key.offset, chunk_end, profile);
9702                 err |= UNKNOWN_TYPE;
9703         }
9704
9705         bg_key.objectid = chunk_key.offset;
9706         bg_key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9707         bg_key.offset = length;
9708
9709         btrfs_init_path(&path);
9710         ret = btrfs_search_slot(NULL, extent_root, &bg_key, &path, 0, 0);
9711         if (ret) {
9712                 error(
9713                 "chunk[%llu %llu) did not find the related block group item",
9714                         chunk_key.offset, chunk_end);
9715                 err |= REFERENCER_MISSING;
9716         } else{
9717                 leaf = path.nodes[0];
9718                 bi = btrfs_item_ptr(leaf, path.slots[0],
9719                                     struct btrfs_block_group_item);
9720                 read_extent_buffer(leaf, &bg_item, (unsigned long)bi,
9721                                    sizeof(bg_item));
9722                 if (btrfs_block_group_flags(&bg_item) != type) {
9723                         error(
9724 "chunk[%llu %llu) related block group item flags mismatch, wanted: %llu, have: %llu",
9725                                 chunk_key.offset, chunk_end, type,
9726                                 btrfs_block_group_flags(&bg_item));
9727                         err |= REFERENCER_MISSING;
9728                 }
9729         }
9730
9731         num_stripes = btrfs_chunk_num_stripes(eb, chunk);
9732         for (i = 0; i < num_stripes; i++) {
9733                 btrfs_release_path(&path);
9734                 btrfs_init_path(&path);
9735                 devext_key.objectid = btrfs_stripe_devid_nr(eb, chunk, i);
9736                 devext_key.type = BTRFS_DEV_EXTENT_KEY;
9737                 devext_key.offset = btrfs_stripe_offset_nr(eb, chunk, i);
9738
9739                 ret = btrfs_search_slot(NULL, dev_root, &devext_key, &path,
9740                                         0, 0);
9741                 if (ret)
9742                         goto not_match_dev;
9743
9744                 leaf = path.nodes[0];
9745                 ptr = btrfs_item_ptr(leaf, path.slots[0],
9746                                      struct btrfs_dev_extent);
9747                 objectid = btrfs_dev_extent_chunk_objectid(leaf, ptr);
9748                 offset = btrfs_dev_extent_chunk_offset(leaf, ptr);
9749                 if (objectid != chunk_key.objectid ||
9750                     offset != chunk_key.offset ||
9751                     btrfs_dev_extent_length(leaf, ptr) != length)
9752                         goto not_match_dev;
9753                 continue;
9754 not_match_dev:
9755                 err |= BACKREF_MISSING;
9756                 error(
9757                 "chunk[%llu %llu) stripe %d did not find the related dev extent",
9758                         chunk_key.objectid, chunk_end, i);
9759                 continue;
9760         }
9761         btrfs_release_path(&path);
9762 out:
9763         return err;
9764 }
9765
9766 /*
9767  * Main entry function to check known items and update related accounting info
9768  */
9769 static int check_leaf_items(struct btrfs_root *root, struct extent_buffer *eb)
9770 {
9771         struct btrfs_fs_info *fs_info = root->fs_info;
9772         struct btrfs_key key;
9773         int slot = 0;
9774         int type;
9775         struct btrfs_extent_data_ref *dref;
9776         int ret;
9777         int err = 0;
9778
9779 next:
9780         btrfs_item_key_to_cpu(eb, &key, slot);
9781         type = key.type;
9782
9783         switch (type) {
9784         case BTRFS_EXTENT_DATA_KEY:
9785                 ret = check_extent_data_item(root, eb, slot);
9786                 err |= ret;
9787                 break;
9788         case BTRFS_BLOCK_GROUP_ITEM_KEY:
9789                 ret = check_block_group_item(fs_info, eb, slot);
9790                 err |= ret;
9791                 break;
9792         case BTRFS_DEV_ITEM_KEY:
9793                 ret = check_dev_item(fs_info, eb, slot);
9794                 err |= ret;
9795                 break;
9796         case BTRFS_CHUNK_ITEM_KEY:
9797                 ret = check_chunk_item(fs_info, eb, slot);
9798                 err |= ret;
9799                 break;
9800         case BTRFS_DEV_EXTENT_KEY:
9801                 ret = check_dev_extent_item(fs_info, eb, slot);
9802                 err |= ret;
9803                 break;
9804         case BTRFS_EXTENT_ITEM_KEY:
9805         case BTRFS_METADATA_ITEM_KEY:
9806                 ret = check_extent_item(fs_info, eb, slot);
9807                 err |= ret;
9808                 break;
9809         case BTRFS_EXTENT_CSUM_KEY:
9810                 total_csum_bytes += btrfs_item_size_nr(eb, slot);
9811                 break;
9812         case BTRFS_TREE_BLOCK_REF_KEY:
9813                 ret = check_tree_block_backref(fs_info, key.offset,
9814                                                key.objectid, -1);
9815                 err |= ret;
9816                 break;
9817         case BTRFS_EXTENT_DATA_REF_KEY:
9818                 dref = btrfs_item_ptr(eb, slot, struct btrfs_extent_data_ref);
9819                 ret = check_extent_data_backref(fs_info,
9820                                 btrfs_extent_data_ref_root(eb, dref),
9821                                 btrfs_extent_data_ref_objectid(eb, dref),
9822                                 btrfs_extent_data_ref_offset(eb, dref),
9823                                 key.objectid, 0,
9824                                 btrfs_extent_data_ref_count(eb, dref));
9825                 err |= ret;
9826                 break;
9827         case BTRFS_SHARED_BLOCK_REF_KEY:
9828                 ret = check_shared_block_backref(fs_info, key.offset,
9829                                                  key.objectid, -1);
9830                 err |= ret;
9831                 break;
9832         case BTRFS_SHARED_DATA_REF_KEY:
9833                 ret = check_shared_data_backref(fs_info, key.offset,
9834                                                 key.objectid);
9835                 err |= ret;
9836                 break;
9837         default:
9838                 break;
9839         }
9840
9841         if (++slot < btrfs_header_nritems(eb))
9842                 goto next;
9843
9844         return err;
9845 }
9846
9847 /*
9848  * Helper function for later fs/subvol tree check.  To determine if a tree
9849  * block should be checked.
9850  * This function will ensure only the direct referencer with lowest rootid to
9851  * check a fs/subvolume tree block.
9852  *
9853  * Backref check at extent tree would detect errors like missing subvolume
9854  * tree, so we can do aggressive check to reduce duplicated checks.
9855  */
9856 static int should_check(struct btrfs_root *root, struct extent_buffer *eb)
9857 {
9858         struct btrfs_root *extent_root = root->fs_info->extent_root;
9859         struct btrfs_key key;
9860         struct btrfs_path path;
9861         struct extent_buffer *leaf;
9862         int slot;
9863         struct btrfs_extent_item *ei;
9864         unsigned long ptr;
9865         unsigned long end;
9866         int type;
9867         u32 item_size;
9868         u64 offset;
9869         struct btrfs_extent_inline_ref *iref;
9870         int ret;
9871
9872         btrfs_init_path(&path);
9873         key.objectid = btrfs_header_bytenr(eb);
9874         key.type = BTRFS_METADATA_ITEM_KEY;
9875         key.offset = (u64)-1;
9876
9877         /*
9878          * Any failure in backref resolving means we can't determine
9879          * whom the tree block belongs to.
9880          * So in that case, we need to check that tree block
9881          */
9882         ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
9883         if (ret < 0)
9884                 goto need_check;
9885
9886         ret = btrfs_previous_extent_item(extent_root, &path,
9887                                          btrfs_header_bytenr(eb));
9888         if (ret)
9889                 goto need_check;
9890
9891         leaf = path.nodes[0];
9892         slot = path.slots[0];
9893         btrfs_item_key_to_cpu(leaf, &key, slot);
9894         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
9895
9896         if (key.type == BTRFS_METADATA_ITEM_KEY) {
9897                 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
9898         } else {
9899                 struct btrfs_tree_block_info *info;
9900
9901                 info = (struct btrfs_tree_block_info *)(ei + 1);
9902                 iref = (struct btrfs_extent_inline_ref *)(info + 1);
9903         }
9904
9905         item_size = btrfs_item_size_nr(leaf, slot);
9906         ptr = (unsigned long)iref;
9907         end = (unsigned long)ei + item_size;
9908         while (ptr < end) {
9909                 iref = (struct btrfs_extent_inline_ref *)ptr;
9910                 type = btrfs_extent_inline_ref_type(leaf, iref);
9911                 offset = btrfs_extent_inline_ref_offset(leaf, iref);
9912
9913                 /*
9914                  * We only check the tree block if current root is
9915                  * the lowest referencer of it.
9916                  */
9917                 if (type == BTRFS_TREE_BLOCK_REF_KEY &&
9918                     offset < root->objectid) {
9919                         btrfs_release_path(&path);
9920                         return 0;
9921                 }
9922
9923                 ptr += btrfs_extent_inline_ref_size(type);
9924         }
9925         /*
9926          * Normally we should also check keyed tree block ref, but that may be
9927          * very time consuming.  Inlined ref should already make us skip a lot
9928          * of refs now.  So skip search keyed tree block ref.
9929          */
9930
9931 need_check:
9932         btrfs_release_path(&path);
9933         return 1;
9934 }
9935
9936 /*
9937  * Traversal function for tree block. We will do:
9938  * 1) Skip shared fs/subvolume tree blocks
9939  * 2) Update related bytes accounting
9940  * 3) Pre-order traversal
9941  */
9942 static int traverse_tree_block(struct btrfs_root *root,
9943                                 struct extent_buffer *node)
9944 {
9945         struct extent_buffer *eb;
9946         struct btrfs_key key;
9947         struct btrfs_key drop_key;
9948         int level;
9949         u64 nr;
9950         int i;
9951         int err = 0;
9952         int ret;
9953
9954         /*
9955          * Skip shared fs/subvolume tree block, in that case they will
9956          * be checked by referencer with lowest rootid
9957          */
9958         if (is_fstree(root->objectid) && !should_check(root, node))
9959                 return 0;
9960
9961         /* Update bytes accounting */
9962         total_btree_bytes += node->len;
9963         if (fs_root_objectid(btrfs_header_owner(node)))
9964                 total_fs_tree_bytes += node->len;
9965         if (btrfs_header_owner(node) == BTRFS_EXTENT_TREE_OBJECTID)
9966                 total_extent_tree_bytes += node->len;
9967         if (!found_old_backref &&
9968             btrfs_header_owner(node) == BTRFS_TREE_RELOC_OBJECTID &&
9969             btrfs_header_backref_rev(node) == BTRFS_MIXED_BACKREF_REV &&
9970             !btrfs_header_flag(node, BTRFS_HEADER_FLAG_RELOC))
9971                 found_old_backref = 1;
9972
9973         /* pre-order tranversal, check itself first */
9974         level = btrfs_header_level(node);
9975         ret = check_tree_block_ref(root, node, btrfs_header_bytenr(node),
9976                                    btrfs_header_level(node),
9977                                    btrfs_header_owner(node));
9978         err |= ret;
9979         if (err)
9980                 error(
9981         "check %s failed root %llu bytenr %llu level %d, force continue check",
9982                         level ? "node":"leaf", root->objectid,
9983                         btrfs_header_bytenr(node), btrfs_header_level(node));
9984
9985         if (!level) {
9986                 btree_space_waste += btrfs_leaf_free_space(root, node);
9987                 ret = check_leaf_items(root, node);
9988                 err |= ret;
9989                 return err;
9990         }
9991
9992         nr = btrfs_header_nritems(node);
9993         btrfs_disk_key_to_cpu(&drop_key, &root->root_item.drop_progress);
9994         btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) - nr) *
9995                 sizeof(struct btrfs_key_ptr);
9996
9997         /* Then check all its children */
9998         for (i = 0; i < nr; i++) {
9999                 u64 blocknr = btrfs_node_blockptr(node, i);
10000
10001                 btrfs_node_key_to_cpu(node, &key, i);
10002                 if (level == root->root_item.drop_level &&
10003                     is_dropped_key(&key, &drop_key))
10004                         continue;
10005
10006                 /*
10007                  * As a btrfs tree has most 8 levels (0..7), so it's quite safe
10008                  * to call the function itself.
10009                  */
10010                 eb = read_tree_block(root, blocknr, root->nodesize, 0);
10011                 if (extent_buffer_uptodate(eb)) {
10012                         ret = traverse_tree_block(root, eb);
10013                         err |= ret;
10014                 }
10015                 free_extent_buffer(eb);
10016         }
10017
10018         return err;
10019 }
10020
10021 /*
10022  * Low memory usage version check_chunks_and_extents.
10023  */
10024 static int check_chunks_and_extents_v2(struct btrfs_root *root)
10025 {
10026         struct btrfs_path path;
10027         struct btrfs_key key;
10028         struct btrfs_root *root1;
10029         struct btrfs_root *cur_root;
10030         int err = 0;
10031         int ret;
10032
10033         root1 = root->fs_info->chunk_root;
10034         ret = traverse_tree_block(root1, root1->node);
10035         err |= ret;
10036
10037         root1 = root->fs_info->tree_root;
10038         ret = traverse_tree_block(root1, root1->node);
10039         err |= ret;
10040
10041         btrfs_init_path(&path);
10042         key.objectid = BTRFS_EXTENT_TREE_OBJECTID;
10043         key.offset = 0;
10044         key.type = BTRFS_ROOT_ITEM_KEY;
10045
10046         ret = btrfs_search_slot(NULL, root1, &key, &path, 0, 0);
10047         if (ret) {
10048                 error("cannot find extent treet in tree_root");
10049                 goto out;
10050         }
10051
10052         while (1) {
10053                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
10054                 if (key.type != BTRFS_ROOT_ITEM_KEY)
10055                         goto next;
10056                 key.offset = (u64)-1;
10057
10058                 cur_root = btrfs_read_fs_root(root->fs_info, &key);
10059                 if (IS_ERR(cur_root) || !cur_root) {
10060                         error("failed to read tree: %lld", key.objectid);
10061                         goto next;
10062                 }
10063
10064                 ret = traverse_tree_block(cur_root, cur_root->node);
10065                 err |= ret;
10066
10067 next:
10068                 ret = btrfs_next_item(root1, &path);
10069                 if (ret)
10070                         goto out;
10071         }
10072
10073 out:
10074         btrfs_release_path(&path);
10075         return err;
10076 }
10077
10078 static int btrfs_fsck_reinit_root(struct btrfs_trans_handle *trans,
10079                            struct btrfs_root *root, int overwrite)
10080 {
10081         struct extent_buffer *c;
10082         struct extent_buffer *old = root->node;
10083         int level;
10084         int ret;
10085         struct btrfs_disk_key disk_key = {0,0,0};
10086
10087         level = 0;
10088
10089         if (overwrite) {
10090                 c = old;
10091                 extent_buffer_get(c);
10092                 goto init;
10093         }
10094         c = btrfs_alloc_free_block(trans, root,
10095                                    root->nodesize,
10096                                    root->root_key.objectid,
10097                                    &disk_key, level, 0, 0);
10098         if (IS_ERR(c)) {
10099                 c = old;
10100                 extent_buffer_get(c);
10101                 overwrite = 1;
10102         }
10103 init:
10104         memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
10105         btrfs_set_header_level(c, level);
10106         btrfs_set_header_bytenr(c, c->start);
10107         btrfs_set_header_generation(c, trans->transid);
10108         btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
10109         btrfs_set_header_owner(c, root->root_key.objectid);
10110
10111         write_extent_buffer(c, root->fs_info->fsid,
10112                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
10113
10114         write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
10115                             btrfs_header_chunk_tree_uuid(c),
10116                             BTRFS_UUID_SIZE);
10117
10118         btrfs_mark_buffer_dirty(c);
10119         /*
10120          * this case can happen in the following case:
10121          *
10122          * 1.overwrite previous root.
10123          *
10124          * 2.reinit reloc data root, this is because we skip pin
10125          * down reloc data tree before which means we can allocate
10126          * same block bytenr here.
10127          */
10128         if (old->start == c->start) {
10129                 btrfs_set_root_generation(&root->root_item,
10130                                           trans->transid);
10131                 root->root_item.level = btrfs_header_level(root->node);
10132                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
10133                                         &root->root_key, &root->root_item);
10134                 if (ret) {
10135                         free_extent_buffer(c);
10136                         return ret;
10137                 }
10138         }
10139         free_extent_buffer(old);
10140         root->node = c;
10141         add_root_to_dirty_list(root);
10142         return 0;
10143 }
10144
10145 static int pin_down_tree_blocks(struct btrfs_fs_info *fs_info,
10146                                 struct extent_buffer *eb, int tree_root)
10147 {
10148         struct extent_buffer *tmp;
10149         struct btrfs_root_item *ri;
10150         struct btrfs_key key;
10151         u64 bytenr;
10152         u32 nodesize;
10153         int level = btrfs_header_level(eb);
10154         int nritems;
10155         int ret;
10156         int i;
10157
10158         /*
10159          * If we have pinned this block before, don't pin it again.
10160          * This can not only avoid forever loop with broken filesystem
10161          * but also give us some speedups.
10162          */
10163         if (test_range_bit(&fs_info->pinned_extents, eb->start,
10164                            eb->start + eb->len - 1, EXTENT_DIRTY, 0))
10165                 return 0;
10166
10167         btrfs_pin_extent(fs_info, eb->start, eb->len);
10168
10169         nodesize = btrfs_super_nodesize(fs_info->super_copy);
10170         nritems = btrfs_header_nritems(eb);
10171         for (i = 0; i < nritems; i++) {
10172                 if (level == 0) {
10173                         btrfs_item_key_to_cpu(eb, &key, i);
10174                         if (key.type != BTRFS_ROOT_ITEM_KEY)
10175                                 continue;
10176                         /* Skip the extent root and reloc roots */
10177                         if (key.objectid == BTRFS_EXTENT_TREE_OBJECTID ||
10178                             key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
10179                             key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
10180                                 continue;
10181                         ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
10182                         bytenr = btrfs_disk_root_bytenr(eb, ri);
10183
10184                         /*
10185                          * If at any point we start needing the real root we
10186                          * will have to build a stump root for the root we are
10187                          * in, but for now this doesn't actually use the root so
10188                          * just pass in extent_root.
10189                          */
10190                         tmp = read_tree_block(fs_info->extent_root, bytenr,
10191                                               nodesize, 0);
10192                         if (!extent_buffer_uptodate(tmp)) {
10193                                 fprintf(stderr, "Error reading root block\n");
10194                                 return -EIO;
10195                         }
10196                         ret = pin_down_tree_blocks(fs_info, tmp, 0);
10197                         free_extent_buffer(tmp);
10198                         if (ret)
10199                                 return ret;
10200                 } else {
10201                         bytenr = btrfs_node_blockptr(eb, i);
10202
10203                         /* If we aren't the tree root don't read the block */
10204                         if (level == 1 && !tree_root) {
10205                                 btrfs_pin_extent(fs_info, bytenr, nodesize);
10206                                 continue;
10207                         }
10208
10209                         tmp = read_tree_block(fs_info->extent_root, bytenr,
10210                                               nodesize, 0);
10211                         if (!extent_buffer_uptodate(tmp)) {
10212                                 fprintf(stderr, "Error reading tree block\n");
10213                                 return -EIO;
10214                         }
10215                         ret = pin_down_tree_blocks(fs_info, tmp, tree_root);
10216                         free_extent_buffer(tmp);
10217                         if (ret)
10218                                 return ret;
10219                 }
10220         }
10221
10222         return 0;
10223 }
10224
10225 static int pin_metadata_blocks(struct btrfs_fs_info *fs_info)
10226 {
10227         int ret;
10228
10229         ret = pin_down_tree_blocks(fs_info, fs_info->chunk_root->node, 0);
10230         if (ret)
10231                 return ret;
10232
10233         return pin_down_tree_blocks(fs_info, fs_info->tree_root->node, 1);
10234 }
10235
10236 static int reset_block_groups(struct btrfs_fs_info *fs_info)
10237 {
10238         struct btrfs_block_group_cache *cache;
10239         struct btrfs_path *path;
10240         struct extent_buffer *leaf;
10241         struct btrfs_chunk *chunk;
10242         struct btrfs_key key;
10243         int ret;
10244         u64 start;
10245
10246         path = btrfs_alloc_path();
10247         if (!path)
10248                 return -ENOMEM;
10249
10250         key.objectid = 0;
10251         key.type = BTRFS_CHUNK_ITEM_KEY;
10252         key.offset = 0;
10253
10254         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
10255         if (ret < 0) {
10256                 btrfs_free_path(path);
10257                 return ret;
10258         }
10259
10260         /*
10261          * We do this in case the block groups were screwed up and had alloc
10262          * bits that aren't actually set on the chunks.  This happens with
10263          * restored images every time and could happen in real life I guess.
10264          */
10265         fs_info->avail_data_alloc_bits = 0;
10266         fs_info->avail_metadata_alloc_bits = 0;
10267         fs_info->avail_system_alloc_bits = 0;
10268
10269         /* First we need to create the in-memory block groups */
10270         while (1) {
10271                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
10272                         ret = btrfs_next_leaf(fs_info->chunk_root, path);
10273                         if (ret < 0) {
10274                                 btrfs_free_path(path);
10275                                 return ret;
10276                         }
10277                         if (ret) {
10278                                 ret = 0;
10279                                 break;
10280                         }
10281                 }
10282                 leaf = path->nodes[0];
10283                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
10284                 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
10285                         path->slots[0]++;
10286                         continue;
10287                 }
10288
10289                 chunk = btrfs_item_ptr(leaf, path->slots[0],
10290                                        struct btrfs_chunk);
10291                 btrfs_add_block_group(fs_info, 0,
10292                                       btrfs_chunk_type(leaf, chunk),
10293                                       key.objectid, key.offset,
10294                                       btrfs_chunk_length(leaf, chunk));
10295                 set_extent_dirty(&fs_info->free_space_cache, key.offset,
10296                                  key.offset + btrfs_chunk_length(leaf, chunk),
10297                                  GFP_NOFS);
10298                 path->slots[0]++;
10299         }
10300         start = 0;
10301         while (1) {
10302                 cache = btrfs_lookup_first_block_group(fs_info, start);
10303                 if (!cache)
10304                         break;
10305                 cache->cached = 1;
10306                 start = cache->key.objectid + cache->key.offset;
10307         }
10308
10309         btrfs_free_path(path);
10310         return 0;
10311 }
10312
10313 static int reset_balance(struct btrfs_trans_handle *trans,
10314                          struct btrfs_fs_info *fs_info)
10315 {
10316         struct btrfs_root *root = fs_info->tree_root;
10317         struct btrfs_path *path;
10318         struct extent_buffer *leaf;
10319         struct btrfs_key key;
10320         int del_slot, del_nr = 0;
10321         int ret;
10322         int found = 0;
10323
10324         path = btrfs_alloc_path();
10325         if (!path)
10326                 return -ENOMEM;
10327
10328         key.objectid = BTRFS_BALANCE_OBJECTID;
10329         key.type = BTRFS_BALANCE_ITEM_KEY;
10330         key.offset = 0;
10331
10332         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10333         if (ret) {
10334                 if (ret > 0)
10335                         ret = 0;
10336                 if (!ret)
10337                         goto reinit_data_reloc;
10338                 else
10339                         goto out;
10340         }
10341
10342         ret = btrfs_del_item(trans, root, path);
10343         if (ret)
10344                 goto out;
10345         btrfs_release_path(path);
10346
10347         key.objectid = BTRFS_TREE_RELOC_OBJECTID;
10348         key.type = BTRFS_ROOT_ITEM_KEY;
10349         key.offset = 0;
10350
10351         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10352         if (ret < 0)
10353                 goto out;
10354         while (1) {
10355                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
10356                         if (!found)
10357                                 break;
10358
10359                         if (del_nr) {
10360                                 ret = btrfs_del_items(trans, root, path,
10361                                                       del_slot, del_nr);
10362                                 del_nr = 0;
10363                                 if (ret)
10364                                         goto out;
10365                         }
10366                         key.offset++;
10367                         btrfs_release_path(path);
10368
10369                         found = 0;
10370                         ret = btrfs_search_slot(trans, root, &key, path,
10371                                                 -1, 1);
10372                         if (ret < 0)
10373                                 goto out;
10374                         continue;
10375                 }
10376                 found = 1;
10377                 leaf = path->nodes[0];
10378                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
10379                 if (key.objectid > BTRFS_TREE_RELOC_OBJECTID)
10380                         break;
10381                 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
10382                         path->slots[0]++;
10383                         continue;
10384                 }
10385                 if (!del_nr) {
10386                         del_slot = path->slots[0];
10387                         del_nr = 1;
10388                 } else {
10389                         del_nr++;
10390                 }
10391                 path->slots[0]++;
10392         }
10393
10394         if (del_nr) {
10395                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
10396                 if (ret)
10397                         goto out;
10398         }
10399         btrfs_release_path(path);
10400
10401 reinit_data_reloc:
10402         key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
10403         key.type = BTRFS_ROOT_ITEM_KEY;
10404         key.offset = (u64)-1;
10405         root = btrfs_read_fs_root(fs_info, &key);
10406         if (IS_ERR(root)) {
10407                 fprintf(stderr, "Error reading data reloc tree\n");
10408                 ret = PTR_ERR(root);
10409                 goto out;
10410         }
10411         record_root_in_trans(trans, root);
10412         ret = btrfs_fsck_reinit_root(trans, root, 0);
10413         if (ret)
10414                 goto out;
10415         ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
10416 out:
10417         btrfs_free_path(path);
10418         return ret;
10419 }
10420
10421 static int reinit_extent_tree(struct btrfs_trans_handle *trans,
10422                               struct btrfs_fs_info *fs_info)
10423 {
10424         u64 start = 0;
10425         int ret;
10426
10427         /*
10428          * The only reason we don't do this is because right now we're just
10429          * walking the trees we find and pinning down their bytes, we don't look
10430          * at any of the leaves.  In order to do mixed groups we'd have to check
10431          * the leaves of any fs roots and pin down the bytes for any file
10432          * extents we find.  Not hard but why do it if we don't have to?
10433          */
10434         if (btrfs_fs_incompat(fs_info, BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)) {
10435                 fprintf(stderr, "We don't support re-initing the extent tree "
10436                         "for mixed block groups yet, please notify a btrfs "
10437                         "developer you want to do this so they can add this "
10438                         "functionality.\n");
10439                 return -EINVAL;
10440         }
10441
10442         /*
10443          * first we need to walk all of the trees except the extent tree and pin
10444          * down the bytes that are in use so we don't overwrite any existing
10445          * metadata.
10446          */
10447         ret = pin_metadata_blocks(fs_info);
10448         if (ret) {
10449                 fprintf(stderr, "error pinning down used bytes\n");
10450                 return ret;
10451         }
10452
10453         /*
10454          * Need to drop all the block groups since we're going to recreate all
10455          * of them again.
10456          */
10457         btrfs_free_block_groups(fs_info);
10458         ret = reset_block_groups(fs_info);
10459         if (ret) {
10460                 fprintf(stderr, "error resetting the block groups\n");
10461                 return ret;
10462         }
10463
10464         /* Ok we can allocate now, reinit the extent root */
10465         ret = btrfs_fsck_reinit_root(trans, fs_info->extent_root, 0);
10466         if (ret) {
10467                 fprintf(stderr, "extent root initialization failed\n");
10468                 /*
10469                  * When the transaction code is updated we should end the
10470                  * transaction, but for now progs only knows about commit so
10471                  * just return an error.
10472                  */
10473                 return ret;
10474         }
10475
10476         /*
10477          * Now we have all the in-memory block groups setup so we can make
10478          * allocations properly, and the metadata we care about is safe since we
10479          * pinned all of it above.
10480          */
10481         while (1) {
10482                 struct btrfs_block_group_cache *cache;
10483
10484                 cache = btrfs_lookup_first_block_group(fs_info, start);
10485                 if (!cache)
10486                         break;
10487                 start = cache->key.objectid + cache->key.offset;
10488                 ret = btrfs_insert_item(trans, fs_info->extent_root,
10489                                         &cache->key, &cache->item,
10490                                         sizeof(cache->item));
10491                 if (ret) {
10492                         fprintf(stderr, "Error adding block group\n");
10493                         return ret;
10494                 }
10495                 btrfs_extent_post_op(trans, fs_info->extent_root);
10496         }
10497
10498         ret = reset_balance(trans, fs_info);
10499         if (ret)
10500                 fprintf(stderr, "error resetting the pending balance\n");
10501
10502         return ret;
10503 }
10504
10505 static int recow_extent_buffer(struct btrfs_root *root, struct extent_buffer *eb)
10506 {
10507         struct btrfs_path *path;
10508         struct btrfs_trans_handle *trans;
10509         struct btrfs_key key;
10510         int ret;
10511
10512         printf("Recowing metadata block %llu\n", eb->start);
10513         key.objectid = btrfs_header_owner(eb);
10514         key.type = BTRFS_ROOT_ITEM_KEY;
10515         key.offset = (u64)-1;
10516
10517         root = btrfs_read_fs_root(root->fs_info, &key);
10518         if (IS_ERR(root)) {
10519                 fprintf(stderr, "Couldn't find owner root %llu\n",
10520                         key.objectid);
10521                 return PTR_ERR(root);
10522         }
10523
10524         path = btrfs_alloc_path();
10525         if (!path)
10526                 return -ENOMEM;
10527
10528         trans = btrfs_start_transaction(root, 1);
10529         if (IS_ERR(trans)) {
10530                 btrfs_free_path(path);
10531                 return PTR_ERR(trans);
10532         }
10533
10534         path->lowest_level = btrfs_header_level(eb);
10535         if (path->lowest_level)
10536                 btrfs_node_key_to_cpu(eb, &key, 0);
10537         else
10538                 btrfs_item_key_to_cpu(eb, &key, 0);
10539
10540         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
10541         btrfs_commit_transaction(trans, root);
10542         btrfs_free_path(path);
10543         return ret;
10544 }
10545
10546 static int delete_bad_item(struct btrfs_root *root, struct bad_item *bad)
10547 {
10548         struct btrfs_path *path;
10549         struct btrfs_trans_handle *trans;
10550         struct btrfs_key key;
10551         int ret;
10552
10553         printf("Deleting bad item [%llu,%u,%llu]\n", bad->key.objectid,
10554                bad->key.type, bad->key.offset);
10555         key.objectid = bad->root_id;
10556         key.type = BTRFS_ROOT_ITEM_KEY;
10557         key.offset = (u64)-1;
10558
10559         root = btrfs_read_fs_root(root->fs_info, &key);
10560         if (IS_ERR(root)) {
10561                 fprintf(stderr, "Couldn't find owner root %llu\n",
10562                         key.objectid);
10563                 return PTR_ERR(root);
10564         }
10565
10566         path = btrfs_alloc_path();
10567         if (!path)
10568                 return -ENOMEM;
10569
10570         trans = btrfs_start_transaction(root, 1);
10571         if (IS_ERR(trans)) {
10572                 btrfs_free_path(path);
10573                 return PTR_ERR(trans);
10574         }
10575
10576         ret = btrfs_search_slot(trans, root, &bad->key, path, -1, 1);
10577         if (ret) {
10578                 if (ret > 0)
10579                         ret = 0;
10580                 goto out;
10581         }
10582         ret = btrfs_del_item(trans, root, path);
10583 out:
10584         btrfs_commit_transaction(trans, root);
10585         btrfs_free_path(path);
10586         return ret;
10587 }
10588
10589 static int zero_log_tree(struct btrfs_root *root)
10590 {
10591         struct btrfs_trans_handle *trans;
10592         int ret;
10593
10594         trans = btrfs_start_transaction(root, 1);
10595         if (IS_ERR(trans)) {
10596                 ret = PTR_ERR(trans);
10597                 return ret;
10598         }
10599         btrfs_set_super_log_root(root->fs_info->super_copy, 0);
10600         btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
10601         ret = btrfs_commit_transaction(trans, root);
10602         return ret;
10603 }
10604
10605 static int populate_csum(struct btrfs_trans_handle *trans,
10606                          struct btrfs_root *csum_root, char *buf, u64 start,
10607                          u64 len)
10608 {
10609         u64 offset = 0;
10610         u64 sectorsize;
10611         int ret = 0;
10612
10613         while (offset < len) {
10614                 sectorsize = csum_root->sectorsize;
10615                 ret = read_extent_data(csum_root, buf, start + offset,
10616                                        &sectorsize, 0);
10617                 if (ret)
10618                         break;
10619                 ret = btrfs_csum_file_block(trans, csum_root, start + len,
10620                                             start + offset, buf, sectorsize);
10621                 if (ret)
10622                         break;
10623                 offset += sectorsize;
10624         }
10625         return ret;
10626 }
10627
10628 static int fill_csum_tree_from_one_fs_root(struct btrfs_trans_handle *trans,
10629                                       struct btrfs_root *csum_root,
10630                                       struct btrfs_root *cur_root)
10631 {
10632         struct btrfs_path *path;
10633         struct btrfs_key key;
10634         struct extent_buffer *node;
10635         struct btrfs_file_extent_item *fi;
10636         char *buf = NULL;
10637         u64 start = 0;
10638         u64 len = 0;
10639         int slot = 0;
10640         int ret = 0;
10641
10642         path = btrfs_alloc_path();
10643         if (!path)
10644                 return -ENOMEM;
10645         buf = malloc(cur_root->fs_info->csum_root->sectorsize);
10646         if (!buf) {
10647                 ret = -ENOMEM;
10648                 goto out;
10649         }
10650
10651         key.objectid = 0;
10652         key.offset = 0;
10653         key.type = 0;
10654
10655         ret = btrfs_search_slot(NULL, cur_root, &key, path, 0, 0);
10656         if (ret < 0)
10657                 goto out;
10658         /* Iterate all regular file extents and fill its csum */
10659         while (1) {
10660                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
10661
10662                 if (key.type != BTRFS_EXTENT_DATA_KEY)
10663                         goto next;
10664                 node = path->nodes[0];
10665                 slot = path->slots[0];
10666                 fi = btrfs_item_ptr(node, slot, struct btrfs_file_extent_item);
10667                 if (btrfs_file_extent_type(node, fi) != BTRFS_FILE_EXTENT_REG)
10668                         goto next;
10669                 start = btrfs_file_extent_disk_bytenr(node, fi);
10670                 len = btrfs_file_extent_disk_num_bytes(node, fi);
10671
10672                 ret = populate_csum(trans, csum_root, buf, start, len);
10673                 if (ret == -EEXIST)
10674                         ret = 0;
10675                 if (ret < 0)
10676                         goto out;
10677 next:
10678                 /*
10679                  * TODO: if next leaf is corrupted, jump to nearest next valid
10680                  * leaf.
10681                  */
10682                 ret = btrfs_next_item(cur_root, path);
10683                 if (ret < 0)
10684                         goto out;
10685                 if (ret > 0) {
10686                         ret = 0;
10687                         goto out;
10688                 }
10689         }
10690
10691 out:
10692         btrfs_free_path(path);
10693         free(buf);
10694         return ret;
10695 }
10696
10697 static int fill_csum_tree_from_fs(struct btrfs_trans_handle *trans,
10698                                   struct btrfs_root *csum_root)
10699 {
10700         struct btrfs_fs_info *fs_info = csum_root->fs_info;
10701         struct btrfs_path *path;
10702         struct btrfs_root *tree_root = fs_info->tree_root;
10703         struct btrfs_root *cur_root;
10704         struct extent_buffer *node;
10705         struct btrfs_key key;
10706         int slot = 0;
10707         int ret = 0;
10708
10709         path = btrfs_alloc_path();
10710         if (!path)
10711                 return -ENOMEM;
10712
10713         key.objectid = BTRFS_FS_TREE_OBJECTID;
10714         key.offset = 0;
10715         key.type = BTRFS_ROOT_ITEM_KEY;
10716
10717         ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
10718         if (ret < 0)
10719                 goto out;
10720         if (ret > 0) {
10721                 ret = -ENOENT;
10722                 goto out;
10723         }
10724
10725         while (1) {
10726                 node = path->nodes[0];
10727                 slot = path->slots[0];
10728                 btrfs_item_key_to_cpu(node, &key, slot);
10729                 if (key.objectid > BTRFS_LAST_FREE_OBJECTID)
10730                         goto out;
10731                 if (key.type != BTRFS_ROOT_ITEM_KEY)
10732                         goto next;
10733                 if (!is_fstree(key.objectid))
10734                         goto next;
10735                 key.offset = (u64)-1;
10736
10737                 cur_root = btrfs_read_fs_root(fs_info, &key);
10738                 if (IS_ERR(cur_root) || !cur_root) {
10739                         fprintf(stderr, "Fail to read fs/subvol tree: %lld\n",
10740                                 key.objectid);
10741                         goto out;
10742                 }
10743                 ret = fill_csum_tree_from_one_fs_root(trans, csum_root,
10744                                 cur_root);
10745                 if (ret < 0)
10746                         goto out;
10747 next:
10748                 ret = btrfs_next_item(tree_root, path);
10749                 if (ret > 0) {
10750                         ret = 0;
10751                         goto out;
10752                 }
10753                 if (ret < 0)
10754                         goto out;
10755         }
10756
10757 out:
10758         btrfs_free_path(path);
10759         return ret;
10760 }
10761
10762 static int fill_csum_tree_from_extent(struct btrfs_trans_handle *trans,
10763                                       struct btrfs_root *csum_root)
10764 {
10765         struct btrfs_root *extent_root = csum_root->fs_info->extent_root;
10766         struct btrfs_path *path;
10767         struct btrfs_extent_item *ei;
10768         struct extent_buffer *leaf;
10769         char *buf;
10770         struct btrfs_key key;
10771         int ret;
10772
10773         path = btrfs_alloc_path();
10774         if (!path)
10775                 return -ENOMEM;
10776
10777         key.objectid = 0;
10778         key.type = BTRFS_EXTENT_ITEM_KEY;
10779         key.offset = 0;
10780
10781         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
10782         if (ret < 0) {
10783                 btrfs_free_path(path);
10784                 return ret;
10785         }
10786
10787         buf = malloc(csum_root->sectorsize);
10788         if (!buf) {
10789                 btrfs_free_path(path);
10790                 return -ENOMEM;
10791         }
10792
10793         while (1) {
10794                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
10795                         ret = btrfs_next_leaf(extent_root, path);
10796                         if (ret < 0)
10797                                 break;
10798                         if (ret) {
10799                                 ret = 0;
10800                                 break;
10801                         }
10802                 }
10803                 leaf = path->nodes[0];
10804
10805                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
10806                 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
10807                         path->slots[0]++;
10808                         continue;
10809                 }
10810
10811                 ei = btrfs_item_ptr(leaf, path->slots[0],
10812                                     struct btrfs_extent_item);
10813                 if (!(btrfs_extent_flags(leaf, ei) &
10814                       BTRFS_EXTENT_FLAG_DATA)) {
10815                         path->slots[0]++;
10816                         continue;
10817                 }
10818
10819                 ret = populate_csum(trans, csum_root, buf, key.objectid,
10820                                     key.offset);
10821                 if (ret)
10822                         break;
10823                 path->slots[0]++;
10824         }
10825
10826         btrfs_free_path(path);
10827         free(buf);
10828         return ret;
10829 }
10830
10831 /*
10832  * Recalculate the csum and put it into the csum tree.
10833  *
10834  * Extent tree init will wipe out all the extent info, so in that case, we
10835  * can't depend on extent tree, but use fs tree.  If search_fs_tree is set, we
10836  * will use fs/subvol trees to init the csum tree.
10837  */
10838 static int fill_csum_tree(struct btrfs_trans_handle *trans,
10839                           struct btrfs_root *csum_root,
10840                           int search_fs_tree)
10841 {
10842         if (search_fs_tree)
10843                 return fill_csum_tree_from_fs(trans, csum_root);
10844         else
10845                 return fill_csum_tree_from_extent(trans, csum_root);
10846 }
10847
10848 static void free_roots_info_cache(void)
10849 {
10850         if (!roots_info_cache)
10851                 return;
10852
10853         while (!cache_tree_empty(roots_info_cache)) {
10854                 struct cache_extent *entry;
10855                 struct root_item_info *rii;
10856
10857                 entry = first_cache_extent(roots_info_cache);
10858                 if (!entry)
10859                         break;
10860                 remove_cache_extent(roots_info_cache, entry);
10861                 rii = container_of(entry, struct root_item_info, cache_extent);
10862                 free(rii);
10863         }
10864
10865         free(roots_info_cache);
10866         roots_info_cache = NULL;
10867 }
10868
10869 static int build_roots_info_cache(struct btrfs_fs_info *info)
10870 {
10871         int ret = 0;
10872         struct btrfs_key key;
10873         struct extent_buffer *leaf;
10874         struct btrfs_path *path;
10875
10876         if (!roots_info_cache) {
10877                 roots_info_cache = malloc(sizeof(*roots_info_cache));
10878                 if (!roots_info_cache)
10879                         return -ENOMEM;
10880                 cache_tree_init(roots_info_cache);
10881         }
10882
10883         path = btrfs_alloc_path();
10884         if (!path)
10885                 return -ENOMEM;
10886
10887         key.objectid = 0;
10888         key.type = BTRFS_EXTENT_ITEM_KEY;
10889         key.offset = 0;
10890
10891         ret = btrfs_search_slot(NULL, info->extent_root, &key, path, 0, 0);
10892         if (ret < 0)
10893                 goto out;
10894         leaf = path->nodes[0];
10895
10896         while (1) {
10897                 struct btrfs_key found_key;
10898                 struct btrfs_extent_item *ei;
10899                 struct btrfs_extent_inline_ref *iref;
10900                 int slot = path->slots[0];
10901                 int type;
10902                 u64 flags;
10903                 u64 root_id;
10904                 u8 level;
10905                 struct cache_extent *entry;
10906                 struct root_item_info *rii;
10907
10908                 if (slot >= btrfs_header_nritems(leaf)) {
10909                         ret = btrfs_next_leaf(info->extent_root, path);
10910                         if (ret < 0) {
10911                                 break;
10912                         } else if (ret) {
10913                                 ret = 0;
10914                                 break;
10915                         }
10916                         leaf = path->nodes[0];
10917                         slot = path->slots[0];
10918                 }
10919
10920                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10921
10922                 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
10923                     found_key.type != BTRFS_METADATA_ITEM_KEY)
10924                         goto next;
10925
10926                 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
10927                 flags = btrfs_extent_flags(leaf, ei);
10928
10929                 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
10930                     !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
10931                         goto next;
10932
10933                 if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
10934                         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
10935                         level = found_key.offset;
10936                 } else {
10937                         struct btrfs_tree_block_info *binfo;
10938
10939                         binfo = (struct btrfs_tree_block_info *)(ei + 1);
10940                         iref = (struct btrfs_extent_inline_ref *)(binfo + 1);
10941                         level = btrfs_tree_block_level(leaf, binfo);
10942                 }
10943
10944                 /*
10945                  * For a root extent, it must be of the following type and the
10946                  * first (and only one) iref in the item.
10947                  */
10948                 type = btrfs_extent_inline_ref_type(leaf, iref);
10949                 if (type != BTRFS_TREE_BLOCK_REF_KEY)
10950                         goto next;
10951
10952                 root_id = btrfs_extent_inline_ref_offset(leaf, iref);
10953                 entry = lookup_cache_extent(roots_info_cache, root_id, 1);
10954                 if (!entry) {
10955                         rii = malloc(sizeof(struct root_item_info));
10956                         if (!rii) {
10957                                 ret = -ENOMEM;
10958                                 goto out;
10959                         }
10960                         rii->cache_extent.start = root_id;
10961                         rii->cache_extent.size = 1;
10962                         rii->level = (u8)-1;
10963                         entry = &rii->cache_extent;
10964                         ret = insert_cache_extent(roots_info_cache, entry);
10965                         ASSERT(ret == 0);
10966                 } else {
10967                         rii = container_of(entry, struct root_item_info,
10968                                            cache_extent);
10969                 }
10970
10971                 ASSERT(rii->cache_extent.start == root_id);
10972                 ASSERT(rii->cache_extent.size == 1);
10973
10974                 if (level > rii->level || rii->level == (u8)-1) {
10975                         rii->level = level;
10976                         rii->bytenr = found_key.objectid;
10977                         rii->gen = btrfs_extent_generation(leaf, ei);
10978                         rii->node_count = 1;
10979                 } else if (level == rii->level) {
10980                         rii->node_count++;
10981                 }
10982 next:
10983                 path->slots[0]++;
10984         }
10985
10986 out:
10987         btrfs_free_path(path);
10988
10989         return ret;
10990 }
10991
10992 static int maybe_repair_root_item(struct btrfs_fs_info *info,
10993                                   struct btrfs_path *path,
10994                                   const struct btrfs_key *root_key,
10995                                   const int read_only_mode)
10996 {
10997         const u64 root_id = root_key->objectid;
10998         struct cache_extent *entry;
10999         struct root_item_info *rii;
11000         struct btrfs_root_item ri;
11001         unsigned long offset;
11002
11003         entry = lookup_cache_extent(roots_info_cache, root_id, 1);
11004         if (!entry) {
11005                 fprintf(stderr,
11006                         "Error: could not find extent items for root %llu\n",
11007                         root_key->objectid);
11008                 return -ENOENT;
11009         }
11010
11011         rii = container_of(entry, struct root_item_info, cache_extent);
11012         ASSERT(rii->cache_extent.start == root_id);
11013         ASSERT(rii->cache_extent.size == 1);
11014
11015         if (rii->node_count != 1) {
11016                 fprintf(stderr,
11017                         "Error: could not find btree root extent for root %llu\n",
11018                         root_id);
11019                 return -ENOENT;
11020         }
11021
11022         offset = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
11023         read_extent_buffer(path->nodes[0], &ri, offset, sizeof(ri));
11024
11025         if (btrfs_root_bytenr(&ri) != rii->bytenr ||
11026             btrfs_root_level(&ri) != rii->level ||
11027             btrfs_root_generation(&ri) != rii->gen) {
11028
11029                 /*
11030                  * If we're in repair mode but our caller told us to not update
11031                  * the root item, i.e. just check if it needs to be updated, don't
11032                  * print this message, since the caller will call us again shortly
11033                  * for the same root item without read only mode (the caller will
11034                  * open a transaction first).
11035                  */
11036                 if (!(read_only_mode && repair))
11037                         fprintf(stderr,
11038                                 "%sroot item for root %llu,"
11039                                 " current bytenr %llu, current gen %llu, current level %u,"
11040                                 " new bytenr %llu, new gen %llu, new level %u\n",
11041                                 (read_only_mode ? "" : "fixing "),
11042                                 root_id,
11043                                 btrfs_root_bytenr(&ri), btrfs_root_generation(&ri),
11044                                 btrfs_root_level(&ri),
11045                                 rii->bytenr, rii->gen, rii->level);
11046
11047                 if (btrfs_root_generation(&ri) > rii->gen) {
11048                         fprintf(stderr,
11049                                 "root %llu has a root item with a more recent gen (%llu) compared to the found root node (%llu)\n",
11050                                 root_id, btrfs_root_generation(&ri), rii->gen);
11051                         return -EINVAL;
11052                 }
11053
11054                 if (!read_only_mode) {
11055                         btrfs_set_root_bytenr(&ri, rii->bytenr);
11056                         btrfs_set_root_level(&ri, rii->level);
11057                         btrfs_set_root_generation(&ri, rii->gen);
11058                         write_extent_buffer(path->nodes[0], &ri,
11059                                             offset, sizeof(ri));
11060                 }
11061
11062                 return 1;
11063         }
11064
11065         return 0;
11066 }
11067
11068 /*
11069  * A regression introduced in the 3.17 kernel (more specifically in 3.17-rc2),
11070  * caused read-only snapshots to be corrupted if they were created at a moment
11071  * when the source subvolume/snapshot had orphan items. The issue was that the
11072  * on-disk root items became incorrect, referring to the pre orphan cleanup root
11073  * node instead of the post orphan cleanup root node.
11074  * So this function, and its callees, just detects and fixes those cases. Even
11075  * though the regression was for read-only snapshots, this function applies to
11076  * any snapshot/subvolume root.
11077  * This must be run before any other repair code - not doing it so, makes other
11078  * repair code delete or modify backrefs in the extent tree for example, which
11079  * will result in an inconsistent fs after repairing the root items.
11080  */
11081 static int repair_root_items(struct btrfs_fs_info *info)
11082 {
11083         struct btrfs_path *path = NULL;
11084         struct btrfs_key key;
11085         struct extent_buffer *leaf;
11086         struct btrfs_trans_handle *trans = NULL;
11087         int ret = 0;
11088         int bad_roots = 0;
11089         int need_trans = 0;
11090
11091         ret = build_roots_info_cache(info);
11092         if (ret)
11093                 goto out;
11094
11095         path = btrfs_alloc_path();
11096         if (!path) {
11097                 ret = -ENOMEM;
11098                 goto out;
11099         }
11100
11101         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
11102         key.type = BTRFS_ROOT_ITEM_KEY;
11103         key.offset = 0;
11104
11105 again:
11106         /*
11107          * Avoid opening and committing transactions if a leaf doesn't have
11108          * any root items that need to be fixed, so that we avoid rotating
11109          * backup roots unnecessarily.
11110          */
11111         if (need_trans) {
11112                 trans = btrfs_start_transaction(info->tree_root, 1);
11113                 if (IS_ERR(trans)) {
11114                         ret = PTR_ERR(trans);
11115                         goto out;
11116                 }
11117         }
11118
11119         ret = btrfs_search_slot(trans, info->tree_root, &key, path,
11120                                 0, trans ? 1 : 0);
11121         if (ret < 0)
11122                 goto out;
11123         leaf = path->nodes[0];
11124
11125         while (1) {
11126                 struct btrfs_key found_key;
11127
11128                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
11129                         int no_more_keys = find_next_key(path, &key);
11130
11131                         btrfs_release_path(path);
11132                         if (trans) {
11133                                 ret = btrfs_commit_transaction(trans,
11134                                                                info->tree_root);
11135                                 trans = NULL;
11136                                 if (ret < 0)
11137                                         goto out;
11138                         }
11139                         need_trans = 0;
11140                         if (no_more_keys)
11141                                 break;
11142                         goto again;
11143                 }
11144
11145                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
11146
11147                 if (found_key.type != BTRFS_ROOT_ITEM_KEY)
11148                         goto next;
11149                 if (found_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
11150                         goto next;
11151
11152                 ret = maybe_repair_root_item(info, path, &found_key,
11153                                              trans ? 0 : 1);
11154                 if (ret < 0)
11155                         goto out;
11156                 if (ret) {
11157                         if (!trans && repair) {
11158                                 need_trans = 1;
11159                                 key = found_key;
11160                                 btrfs_release_path(path);
11161                                 goto again;
11162                         }
11163                         bad_roots++;
11164                 }
11165 next:
11166                 path->slots[0]++;
11167         }
11168         ret = 0;
11169 out:
11170         free_roots_info_cache();
11171         btrfs_free_path(path);
11172         if (trans)
11173                 btrfs_commit_transaction(trans, info->tree_root);
11174         if (ret < 0)
11175                 return ret;
11176
11177         return bad_roots;
11178 }
11179
11180 static int clear_free_space_cache(struct btrfs_fs_info *fs_info)
11181 {
11182         struct btrfs_trans_handle *trans;
11183         struct btrfs_block_group_cache *bg_cache;
11184         u64 current = 0;
11185         int ret = 0;
11186
11187         /* Clear all free space cache inodes and its extent data */
11188         while (1) {
11189                 bg_cache = btrfs_lookup_first_block_group(fs_info, current);
11190                 if (!bg_cache)
11191                         break;
11192                 ret = btrfs_clear_free_space_cache(fs_info, bg_cache);
11193                 if (ret < 0)
11194                         return ret;
11195                 current = bg_cache->key.objectid + bg_cache->key.offset;
11196         }
11197
11198         /* Don't forget to set cache_generation to -1 */
11199         trans = btrfs_start_transaction(fs_info->tree_root, 0);
11200         if (IS_ERR(trans)) {
11201                 error("failed to update super block cache generation");
11202                 return PTR_ERR(trans);
11203         }
11204         btrfs_set_super_cache_generation(fs_info->super_copy, (u64)-1);
11205         btrfs_commit_transaction(trans, fs_info->tree_root);
11206
11207         return ret;
11208 }
11209
11210 const char * const cmd_check_usage[] = {
11211         "btrfs check [options] <device>",
11212         "Check structural integrity of a filesystem (unmounted).",
11213         "Check structural integrity of an unmounted filesystem. Verify internal",
11214         "trees' consistency and item connectivity. In the repair mode try to",
11215         "fix the problems found. ",
11216         "WARNING: the repair mode is considered dangerous",
11217         "",
11218         "-s|--super <superblock>     use this superblock copy",
11219         "-b|--backup                 use the first valid backup root copy",
11220         "--repair                    try to repair the filesystem",
11221         "--readonly                  run in read-only mode (default)",
11222         "--init-csum-tree            create a new CRC tree",
11223         "--init-extent-tree          create a new extent tree",
11224         "--mode <MODE>               allows choice of memory/IO trade-offs",
11225         "                            where MODE is one of:",
11226         "                            original - read inodes and extents to memory (requires",
11227         "                                       more memory, does less IO)",
11228         "                            lowmem   - try to use less memory but read blocks again",
11229         "                                       when needed",
11230         "--check-data-csum           verify checksums of data blocks",
11231         "-Q|--qgroup-report          print a report on qgroup consistency",
11232         "-E|--subvol-extents <subvolid>",
11233         "                            print subvolume extents and sharing state",
11234         "-r|--tree-root <bytenr>     use the given bytenr for the tree root",
11235         "--chunk-root <bytenr>       use the given bytenr for the chunk tree root",
11236         "-p|--progress               indicate progress",
11237         "--clear-space-cache v1|v2   clear space cache for v1 or v2",
11238         "                            NOTE: v1 support implemented",
11239         NULL
11240 };
11241
11242 int cmd_check(int argc, char **argv)
11243 {
11244         struct cache_tree root_cache;
11245         struct btrfs_root *root;
11246         struct btrfs_fs_info *info;
11247         u64 bytenr = 0;
11248         u64 subvolid = 0;
11249         u64 tree_root_bytenr = 0;
11250         u64 chunk_root_bytenr = 0;
11251         char uuidbuf[BTRFS_UUID_UNPARSED_SIZE];
11252         int ret;
11253         u64 num;
11254         int init_csum_tree = 0;
11255         int readonly = 0;
11256         int clear_space_cache = 0;
11257         int qgroup_report = 0;
11258         int qgroups_repaired = 0;
11259         unsigned ctree_flags = OPEN_CTREE_EXCLUSIVE;
11260
11261         while(1) {
11262                 int c;
11263                 enum { GETOPT_VAL_REPAIR = 257, GETOPT_VAL_INIT_CSUM,
11264                         GETOPT_VAL_INIT_EXTENT, GETOPT_VAL_CHECK_CSUM,
11265                         GETOPT_VAL_READONLY, GETOPT_VAL_CHUNK_TREE,
11266                         GETOPT_VAL_MODE, GETOPT_VAL_CLEAR_SPACE_CACHE };
11267                 static const struct option long_options[] = {
11268                         { "super", required_argument, NULL, 's' },
11269                         { "repair", no_argument, NULL, GETOPT_VAL_REPAIR },
11270                         { "readonly", no_argument, NULL, GETOPT_VAL_READONLY },
11271                         { "init-csum-tree", no_argument, NULL,
11272                                 GETOPT_VAL_INIT_CSUM },
11273                         { "init-extent-tree", no_argument, NULL,
11274                                 GETOPT_VAL_INIT_EXTENT },
11275                         { "check-data-csum", no_argument, NULL,
11276                                 GETOPT_VAL_CHECK_CSUM },
11277                         { "backup", no_argument, NULL, 'b' },
11278                         { "subvol-extents", required_argument, NULL, 'E' },
11279                         { "qgroup-report", no_argument, NULL, 'Q' },
11280                         { "tree-root", required_argument, NULL, 'r' },
11281                         { "chunk-root", required_argument, NULL,
11282                                 GETOPT_VAL_CHUNK_TREE },
11283                         { "progress", no_argument, NULL, 'p' },
11284                         { "mode", required_argument, NULL,
11285                                 GETOPT_VAL_MODE },
11286                         { "clear-space-cache", required_argument, NULL,
11287                                 GETOPT_VAL_CLEAR_SPACE_CACHE},
11288                         { NULL, 0, NULL, 0}
11289                 };
11290
11291                 c = getopt_long(argc, argv, "as:br:p", long_options, NULL);
11292                 if (c < 0)
11293                         break;
11294                 switch(c) {
11295                         case 'a': /* ignored */ break;
11296                         case 'b':
11297                                 ctree_flags |= OPEN_CTREE_BACKUP_ROOT;
11298                                 break;
11299                         case 's':
11300                                 num = arg_strtou64(optarg);
11301                                 if (num >= BTRFS_SUPER_MIRROR_MAX) {
11302                                         error(
11303                                         "super mirror should be less than %d",
11304                                                 BTRFS_SUPER_MIRROR_MAX);
11305                                         exit(1);
11306                                 }
11307                                 bytenr = btrfs_sb_offset(((int)num));
11308                                 printf("using SB copy %llu, bytenr %llu\n", num,
11309                                        (unsigned long long)bytenr);
11310                                 break;
11311                         case 'Q':
11312                                 qgroup_report = 1;
11313                                 break;
11314                         case 'E':
11315                                 subvolid = arg_strtou64(optarg);
11316                                 break;
11317                         case 'r':
11318                                 tree_root_bytenr = arg_strtou64(optarg);
11319                                 break;
11320                         case GETOPT_VAL_CHUNK_TREE:
11321                                 chunk_root_bytenr = arg_strtou64(optarg);
11322                                 break;
11323                         case 'p':
11324                                 ctx.progress_enabled = true;
11325                                 break;
11326                         case '?':
11327                         case 'h':
11328                                 usage(cmd_check_usage);
11329                         case GETOPT_VAL_REPAIR:
11330                                 printf("enabling repair mode\n");
11331                                 repair = 1;
11332                                 ctree_flags |= OPEN_CTREE_WRITES;
11333                                 break;
11334                         case GETOPT_VAL_READONLY:
11335                                 readonly = 1;
11336                                 break;
11337                         case GETOPT_VAL_INIT_CSUM:
11338                                 printf("Creating a new CRC tree\n");
11339                                 init_csum_tree = 1;
11340                                 repair = 1;
11341                                 ctree_flags |= OPEN_CTREE_WRITES;
11342                                 break;
11343                         case GETOPT_VAL_INIT_EXTENT:
11344                                 init_extent_tree = 1;
11345                                 ctree_flags |= (OPEN_CTREE_WRITES |
11346                                                 OPEN_CTREE_NO_BLOCK_GROUPS);
11347                                 repair = 1;
11348                                 break;
11349                         case GETOPT_VAL_CHECK_CSUM:
11350                                 check_data_csum = 1;
11351                                 break;
11352                         case GETOPT_VAL_MODE:
11353                                 check_mode = parse_check_mode(optarg);
11354                                 if (check_mode == CHECK_MODE_UNKNOWN) {
11355                                         error("unknown mode: %s", optarg);
11356                                         exit(1);
11357                                 }
11358                                 break;
11359                         case GETOPT_VAL_CLEAR_SPACE_CACHE:
11360                                 if (strcmp(optarg, "v1") != 0) {
11361                                         error(
11362                         "only v1 support implmented, unrecognized value %s",
11363                         optarg);
11364                                         exit(1);
11365                                 }
11366                                 clear_space_cache = 1;
11367                                 ctree_flags |= OPEN_CTREE_WRITES;
11368                                 break;
11369                 }
11370         }
11371
11372         if (check_argc_exact(argc - optind, 1))
11373                 usage(cmd_check_usage);
11374
11375         if (ctx.progress_enabled) {
11376                 ctx.tp = TASK_NOTHING;
11377                 ctx.info = task_init(print_status_check, print_status_return, &ctx);
11378         }
11379
11380         /* This check is the only reason for --readonly to exist */
11381         if (readonly && repair) {
11382                 error("repair options are not compatible with --readonly");
11383                 exit(1);
11384         }
11385
11386         /*
11387          * Not supported yet
11388          */
11389         if (repair && check_mode == CHECK_MODE_LOWMEM) {
11390                 error("low memory mode doesn't support repair yet");
11391                 exit(1);
11392         }
11393
11394         radix_tree_init();
11395         cache_tree_init(&root_cache);
11396
11397         if((ret = check_mounted(argv[optind])) < 0) {
11398                 error("could not check mount status: %s", strerror(-ret));
11399                 goto err_out;
11400         } else if(ret) {
11401                 error("%s is currently mounted, aborting", argv[optind]);
11402                 ret = -EBUSY;
11403                 goto err_out;
11404         }
11405
11406         /* only allow partial opening under repair mode */
11407         if (repair)
11408                 ctree_flags |= OPEN_CTREE_PARTIAL;
11409
11410         info = open_ctree_fs_info(argv[optind], bytenr, tree_root_bytenr,
11411                                   chunk_root_bytenr, ctree_flags);
11412         if (!info) {
11413                 error("cannot open file system");
11414                 ret = -EIO;
11415                 goto err_out;
11416         }
11417
11418         global_info = info;
11419         root = info->fs_root;
11420         if (clear_space_cache) {
11421                 if (btrfs_fs_compat_ro(info,
11422                                 BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE)) {
11423                         error(
11424                         "free space cache v2 detected, clearing not implemented");
11425                         ret = 1;
11426                         goto close_out;
11427                 }
11428                 printf("Clearing free space cache\n");
11429                 ret = clear_free_space_cache(info);
11430                 if (ret) {
11431                         error("failed to clear free space cache");
11432                         ret = 1;
11433                 } else {
11434                         printf("Free space cache cleared\n");
11435                 }
11436                 goto close_out;
11437         }
11438
11439         /*
11440          * repair mode will force us to commit transaction which
11441          * will make us fail to load log tree when mounting.
11442          */
11443         if (repair && btrfs_super_log_root(info->super_copy)) {
11444                 ret = ask_user("repair mode will force to clear out log tree, are you sure?");
11445                 if (!ret) {
11446                         ret = 1;
11447                         goto close_out;
11448                 }
11449                 ret = zero_log_tree(root);
11450                 if (ret) {
11451                         error("failed to zero log tree: %d", ret);
11452                         goto close_out;
11453                 }
11454         }
11455
11456         uuid_unparse(info->super_copy->fsid, uuidbuf);
11457         if (qgroup_report) {
11458                 printf("Print quota groups for %s\nUUID: %s\n", argv[optind],
11459                        uuidbuf);
11460                 ret = qgroup_verify_all(info);
11461                 if (ret == 0)
11462                         report_qgroups(1);
11463                 goto close_out;
11464         }
11465         if (subvolid) {
11466                 printf("Print extent state for subvolume %llu on %s\nUUID: %s\n",
11467                        subvolid, argv[optind], uuidbuf);
11468                 ret = print_extent_state(info, subvolid);
11469                 goto close_out;
11470         }
11471         printf("Checking filesystem on %s\nUUID: %s\n", argv[optind], uuidbuf);
11472
11473         if (!extent_buffer_uptodate(info->tree_root->node) ||
11474             !extent_buffer_uptodate(info->dev_root->node) ||
11475             !extent_buffer_uptodate(info->chunk_root->node)) {
11476                 error("critical roots corrupted, unable to check the filesystem");
11477                 ret = -EIO;
11478                 goto close_out;
11479         }
11480
11481         if (init_extent_tree || init_csum_tree) {
11482                 struct btrfs_trans_handle *trans;
11483
11484                 trans = btrfs_start_transaction(info->extent_root, 0);
11485                 if (IS_ERR(trans)) {
11486                         error("error starting transaction");
11487                         ret = PTR_ERR(trans);
11488                         goto close_out;
11489                 }
11490
11491                 if (init_extent_tree) {
11492                         printf("Creating a new extent tree\n");
11493                         ret = reinit_extent_tree(trans, info);
11494                         if (ret)
11495                                 goto close_out;
11496                 }
11497
11498                 if (init_csum_tree) {
11499                         printf("Reinitialize checksum tree\n");
11500                         ret = btrfs_fsck_reinit_root(trans, info->csum_root, 0);
11501                         if (ret) {
11502                                 error("checksum tree initialization failed: %d",
11503                                                 ret);
11504                                 ret = -EIO;
11505                                 goto close_out;
11506                         }
11507
11508                         ret = fill_csum_tree(trans, info->csum_root,
11509                                              init_extent_tree);
11510                         if (ret) {
11511                                 error("checksum tree refilling failed: %d", ret);
11512                                 return -EIO;
11513                         }
11514                 }
11515                 /*
11516                  * Ok now we commit and run the normal fsck, which will add
11517                  * extent entries for all of the items it finds.
11518                  */
11519                 ret = btrfs_commit_transaction(trans, info->extent_root);
11520                 if (ret)
11521                         goto close_out;
11522         }
11523         if (!extent_buffer_uptodate(info->extent_root->node)) {
11524                 error("critical: extent_root, unable to check the filesystem");
11525                 ret = -EIO;
11526                 goto close_out;
11527         }
11528         if (!extent_buffer_uptodate(info->csum_root->node)) {
11529                 error("critical: csum_root, unable to check the filesystem");
11530                 ret = -EIO;
11531                 goto close_out;
11532         }
11533
11534         if (!ctx.progress_enabled)
11535                 printf("checking extents");
11536         if (check_mode == CHECK_MODE_LOWMEM)
11537                 ret = check_chunks_and_extents_v2(root);
11538         else
11539                 ret = check_chunks_and_extents(root);
11540         if (ret)
11541                 printf("Errors found in extent allocation tree or chunk allocation");
11542
11543         ret = repair_root_items(info);
11544         if (ret < 0)
11545                 goto close_out;
11546         if (repair) {
11547                 fprintf(stderr, "Fixed %d roots.\n", ret);
11548                 ret = 0;
11549         } else if (ret > 0) {
11550                 fprintf(stderr,
11551                        "Found %d roots with an outdated root item.\n",
11552                        ret);
11553                 fprintf(stderr,
11554                         "Please run a filesystem check with the option --repair to fix them.\n");
11555                 ret = 1;
11556                 goto close_out;
11557         }
11558
11559         if (!ctx.progress_enabled) {
11560                 if (btrfs_fs_compat_ro(info, BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE))
11561                         fprintf(stderr, "checking free space tree\n");
11562                 else
11563                         fprintf(stderr, "checking free space cache\n");
11564         }
11565         ret = check_space_cache(root);
11566         if (ret)
11567                 goto out;
11568
11569         /*
11570          * We used to have to have these hole extents in between our real
11571          * extents so if we don't have this flag set we need to make sure there
11572          * are no gaps in the file extents for inodes, otherwise we can just
11573          * ignore it when this happens.
11574          */
11575         no_holes = btrfs_fs_incompat(root->fs_info,
11576                                      BTRFS_FEATURE_INCOMPAT_NO_HOLES);
11577         if (!ctx.progress_enabled)
11578                 fprintf(stderr, "checking fs roots\n");
11579         ret = check_fs_roots(root, &root_cache);
11580         if (ret)
11581                 goto out;
11582
11583         fprintf(stderr, "checking csums\n");
11584         ret = check_csums(root);
11585         if (ret)
11586                 goto out;
11587
11588         fprintf(stderr, "checking root refs\n");
11589         ret = check_root_refs(root, &root_cache);
11590         if (ret)
11591                 goto out;
11592
11593         while (repair && !list_empty(&root->fs_info->recow_ebs)) {
11594                 struct extent_buffer *eb;
11595
11596                 eb = list_first_entry(&root->fs_info->recow_ebs,
11597                                       struct extent_buffer, recow);
11598                 list_del_init(&eb->recow);
11599                 ret = recow_extent_buffer(root, eb);
11600                 if (ret)
11601                         break;
11602         }
11603
11604         while (!list_empty(&delete_items)) {
11605                 struct bad_item *bad;
11606
11607                 bad = list_first_entry(&delete_items, struct bad_item, list);
11608                 list_del_init(&bad->list);
11609                 if (repair)
11610                         ret = delete_bad_item(root, bad);
11611                 free(bad);
11612         }
11613
11614         if (info->quota_enabled) {
11615                 int err;
11616                 fprintf(stderr, "checking quota groups\n");
11617                 err = qgroup_verify_all(info);
11618                 if (err)
11619                         goto out;
11620                 report_qgroups(0);
11621                 err = repair_qgroups(info, &qgroups_repaired);
11622                 if (err)
11623                         goto out;
11624         }
11625
11626         if (!list_empty(&root->fs_info->recow_ebs)) {
11627                 error("transid errors in file system");
11628                 ret = 1;
11629         }
11630 out:
11631         /* Don't override original ret */
11632         if (!ret && qgroups_repaired)
11633                 ret = qgroups_repaired;
11634
11635         if (found_old_backref) { /*
11636                  * there was a disk format change when mixed
11637                  * backref was in testing tree. The old format
11638                  * existed about one week.
11639                  */
11640                 printf("\n * Found old mixed backref format. "
11641                        "The old format is not supported! *"
11642                        "\n * Please mount the FS in readonly mode, "
11643                        "backup data and re-format the FS. *\n\n");
11644                 ret = 1;
11645         }
11646         printf("found %llu bytes used err is %d\n",
11647                (unsigned long long)bytes_used, ret);
11648         printf("total csum bytes: %llu\n",(unsigned long long)total_csum_bytes);
11649         printf("total tree bytes: %llu\n",
11650                (unsigned long long)total_btree_bytes);
11651         printf("total fs tree bytes: %llu\n",
11652                (unsigned long long)total_fs_tree_bytes);
11653         printf("total extent tree bytes: %llu\n",
11654                (unsigned long long)total_extent_tree_bytes);
11655         printf("btree space waste bytes: %llu\n",
11656                (unsigned long long)btree_space_waste);
11657         printf("file data blocks allocated: %llu\n referenced %llu\n",
11658                 (unsigned long long)data_bytes_allocated,
11659                 (unsigned long long)data_bytes_referenced);
11660
11661         free_qgroup_counts();
11662         free_root_recs_tree(&root_cache);
11663 close_out:
11664         close_ctree(root);
11665 err_out:
11666         if (ctx.progress_enabled)
11667                 task_deinit(ctx.info);
11668
11669         return ret;
11670 }