cmds-check.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <stdio.h>
  20 #include <stdlib.h>
  21 #include <unistd.h>
  22 #include <fcntl.h>
  23 #include <sys/types.h>
  24 #include <sys/stat.h>
  25 #include <unistd.h>
  26 #include <getopt.h>
  27 #include <uuid/uuid.h>
  28 #include "ctree.h"
  29 #include "volumes.h"
  30 #include "repair.h"
  31 #include "disk-io.h"
  32 #include "print-tree.h"
  33 #include "task-utils.h"
  34 #include "transaction.h"
  35 #include "utils.h"
  36 #include "commands.h"
  37 #include "free-space-cache.h"
  38 #include "free-space-tree.h"
  39 #include "btrfsck.h"
  40 #include "qgroup-verify.h"
  41 #include "rbtree-utils.h"
  42 #include "backref.h"
  43 #include "ulist.h"
  44
  45 enum task_position {
  46         TASK_EXTENTS,
  47         TASK_FREE_SPACE,
  48         TASK_FS_ROOTS,
  49         TASK_NOTHING, /* have to be the last element */
  50 };
  51
  52 struct task_ctx {
  53         int progress_enabled;
  54         enum task_position tp;
  55
  56         struct task_info *info;
  57 };
  58
  59 static u64 bytes_used = 0;
  60 static u64 total_csum_bytes = 0;
  61 static u64 total_btree_bytes = 0;
  62 static u64 total_fs_tree_bytes = 0;
  63 static u64 total_extent_tree_bytes = 0;
  64 static u64 btree_space_waste = 0;
  65 static u64 data_bytes_allocated = 0;
  66 static u64 data_bytes_referenced = 0;
  67 static int found_old_backref = 0;
  68 static LIST_HEAD(duplicate_extents);
  69 static LIST_HEAD(delete_items);
  70 static int no_holes = 0;
  71 static int init_extent_tree = 0;
  72 static int check_data_csum = 0;
  73 static struct btrfs_fs_info *global_info;
  74 static struct task_ctx ctx = { 0 };
  75 static struct cache_tree *roots_info_cache = NULL;
  76
  77 enum btrfs_check_mode {
  78         CHECK_MODE_ORIGINAL,
  79         CHECK_MODE_LOWMEM,
  80         CHECK_MODE_UNKNOWN,
  81         CHECK_MODE_DEFAULT = CHECK_MODE_ORIGINAL
  82 };
  83
  84 static enum btrfs_check_mode check_mode = CHECK_MODE_DEFAULT;
  85
  86 struct extent_backref {
  87         struct list_head list;
  88         unsigned int is_data:1;
  89         unsigned int found_extent_tree:1;
  90         unsigned int full_backref:1;
  91         unsigned int found_ref:1;
  92         unsigned int broken:1;
  93 };
  94
  95 static inline struct extent_backref* to_extent_backref(struct list_head *entry)
  96 {
  97         return list_entry(entry, struct extent_backref, list);
  98 }
  99
 100 struct data_backref {
 101         struct extent_backref node;
 102         union {
 103                 u64 parent;
 104                 u64 root;
 105         };
 106         u64 owner;
 107         u64 offset;
 108         u64 disk_bytenr;
 109         u64 bytes;
 110         u64 ram_bytes;
 111         u32 num_refs;
 112         u32 found_ref;
 113 };
 114
 115 static inline struct data_backref* to_data_backref(struct extent_backref *back)
 116 {
 117         return container_of(back, struct data_backref, node);
 118 }
 119
 120 /*
 121  * Much like data_backref, just removed the undetermined members
 122  * and change it to use list_head.
 123  * During extent scan, it is stored in root->orphan_data_extent.
 124  * During fs tree scan, it is then moved to inode_rec->orphan_data_extents.
 125  */
 126 struct orphan_data_extent {
 127         struct list_head list;
 128         u64 root;
 129         u64 objectid;
 130         u64 offset;
 131         u64 disk_bytenr;
 132         u64 disk_len;
 133 };
 134
 135 struct tree_backref {
 136         struct extent_backref node;
 137         union {
 138                 u64 parent;
 139                 u64 root;
 140         };
 141 };
 142
 143 static inline struct tree_backref* to_tree_backref(struct extent_backref *back)
 144 {
 145         return container_of(back, struct tree_backref, node);
 146 }
 147
 148 /* Explicit initialization for extent_record::flag_block_full_backref */
 149 enum { FLAG_UNSET = 2 };
 150
 151 struct extent_record {
 152         struct list_head backrefs;
 153         struct list_head dups;
 154         struct list_head list;
 155         struct cache_extent cache;
 156         struct btrfs_disk_key parent_key;
 157         u64 start;
 158         u64 max_size;
 159         u64 nr;
 160         u64 refs;
 161         u64 extent_item_refs;
 162         u64 generation;
 163         u64 parent_generation;
 164         u64 info_objectid;
 165         u32 num_duplicates;
 166         u8 info_level;
 167         unsigned int flag_block_full_backref:2;
 168         unsigned int found_rec:1;
 169         unsigned int content_checked:1;
 170         unsigned int owner_ref_checked:1;
 171         unsigned int is_root:1;
 172         unsigned int metadata:1;
 173         unsigned int bad_full_backref:1;
 174         unsigned int crossing_stripes:1;
 175         unsigned int wrong_chunk_type:1;
 176 };
 177
 178 static inline struct extent_record* to_extent_record(struct list_head *entry)
 179 {
 180         return container_of(entry, struct extent_record, list);
 181 }
 182
 183 struct inode_backref {
 184         struct list_head list;
 185         unsigned int found_dir_item:1;
 186         unsigned int found_dir_index:1;
 187         unsigned int found_inode_ref:1;
 188         u8 filetype;
 189         u8 ref_type;
 190         int errors;
 191         u64 dir;
 192         u64 index;
 193         u16 namelen;
 194         char name[0];
 195 };
 196
 197 static inline struct inode_backref* to_inode_backref(struct list_head *entry)
 198 {
 199         return list_entry(entry, struct inode_backref, list);
 200 }
 201
 202 struct root_item_record {
 203         struct list_head list;
 204         u64 objectid;
 205         u64 bytenr;
 206         u64 last_snapshot;
 207         u8 level;
 208         u8 drop_level;
 209         int level_size;
 210         struct btrfs_key drop_key;
 211 };
 212
 213 #define REF_ERR_NO_DIR_ITEM             (1 << 0)
 214 #define REF_ERR_NO_DIR_INDEX            (1 << 1)
 215 #define REF_ERR_NO_INODE_REF            (1 << 2)
 216 #define REF_ERR_DUP_DIR_ITEM            (1 << 3)
 217 #define REF_ERR_DUP_DIR_INDEX           (1 << 4)
 218 #define REF_ERR_DUP_INODE_REF           (1 << 5)
 219 #define REF_ERR_INDEX_UNMATCH           (1 << 6)
 220 #define REF_ERR_FILETYPE_UNMATCH        (1 << 7)
 221 #define REF_ERR_NAME_TOO_LONG           (1 << 8) // 100
 222 #define REF_ERR_NO_ROOT_REF             (1 << 9)
 223 #define REF_ERR_NO_ROOT_BACKREF         (1 << 10)
 224 #define REF_ERR_DUP_ROOT_REF            (1 << 11)
 225 #define REF_ERR_DUP_ROOT_BACKREF        (1 << 12)
 226
 227 struct file_extent_hole {
 228         struct rb_node node;
 229         u64 start;
 230         u64 len;
 231 };
 232
 233 struct inode_record {
 234         struct list_head backrefs;
 235         unsigned int checked:1;
 236         unsigned int merging:1;
 237         unsigned int found_inode_item:1;
 238         unsigned int found_dir_item:1;
 239         unsigned int found_file_extent:1;
 240         unsigned int found_csum_item:1;
 241         unsigned int some_csum_missing:1;
 242         unsigned int nodatasum:1;
 243         int errors;
 244
 245         u64 ino;
 246         u32 nlink;
 247         u32 imode;
 248         u64 isize;
 249         u64 nbytes;
 250
 251         u32 found_link;
 252         u64 found_size;
 253         u64 extent_start;
 254         u64 extent_end;
 255         struct rb_root holes;
 256         struct list_head orphan_extents;
 257
 258         u32 refs;
 259 };
 260
 261 #define I_ERR_NO_INODE_ITEM             (1 << 0)
 262 #define I_ERR_NO_ORPHAN_ITEM            (1 << 1)
 263 #define I_ERR_DUP_INODE_ITEM            (1 << 2)
 264 #define I_ERR_DUP_DIR_INDEX             (1 << 3)
 265 #define I_ERR_ODD_DIR_ITEM              (1 << 4)
 266 #define I_ERR_ODD_FILE_EXTENT           (1 << 5)
 267 #define I_ERR_BAD_FILE_EXTENT           (1 << 6)
 268 #define I_ERR_FILE_EXTENT_OVERLAP       (1 << 7)
 269 #define I_ERR_FILE_EXTENT_DISCOUNT      (1 << 8) // 100
 270 #define I_ERR_DIR_ISIZE_WRONG           (1 << 9)
 271 #define I_ERR_FILE_NBYTES_WRONG         (1 << 10) // 400
 272 #define I_ERR_ODD_CSUM_ITEM             (1 << 11)
 273 #define I_ERR_SOME_CSUM_MISSING         (1 << 12)
 274 #define I_ERR_LINK_COUNT_WRONG          (1 << 13)
 275 #define I_ERR_FILE_EXTENT_ORPHAN        (1 << 14)
 276
 277 struct root_backref {
 278         struct list_head list;
 279         unsigned int found_dir_item:1;
 280         unsigned int found_dir_index:1;
 281         unsigned int found_back_ref:1;
 282         unsigned int found_forward_ref:1;
 283         unsigned int reachable:1;
 284         int errors;
 285         u64 ref_root;
 286         u64 dir;
 287         u64 index;
 288         u16 namelen;
 289         char name[0];
 290 };
 291
 292 static inline struct root_backref* to_root_backref(struct list_head *entry)
 293 {
 294         return list_entry(entry, struct root_backref, list);
 295 }
 296
 297 struct root_record {
 298         struct list_head backrefs;
 299         struct cache_extent cache;
 300         unsigned int found_root_item:1;
 301         u64 objectid;
 302         u32 found_ref;
 303 };
 304
 305 struct ptr_node {
 306         struct cache_extent cache;
 307         void *data;
 308 };
 309
 310 struct shared_node {
 311         struct cache_extent cache;
 312         struct cache_tree root_cache;
 313         struct cache_tree inode_cache;
 314         struct inode_record *current;
 315         u32 refs;
 316 };
 317
 318 struct block_info {
 319         u64 start;
 320         u32 size;
 321 };
 322
 323 struct walk_control {
 324         struct cache_tree shared;
 325         struct shared_node *nodes[BTRFS_MAX_LEVEL];
 326         int active_node;
 327         int root_level;
 328 };
 329
 330 struct bad_item {
 331         struct btrfs_key key;
 332         u64 root_id;
 333         struct list_head list;
 334 };
 335
 336 struct extent_entry {
 337         u64 bytenr;
 338         u64 bytes;
 339         int count;
 340         int broken;
 341         struct list_head list;
 342 };
 343
 344 struct root_item_info {
 345         /* level of the root */
 346         u8 level;
 347         /* number of nodes at this level, must be 1 for a root */
 348         int node_count;
 349         u64 bytenr;
 350         u64 gen;
 351         struct cache_extent cache_extent;
 352 };
 353
 354 /*
 355  * Error bit for low memory mode check.
 356  *
 357  * Currently no caller cares about it yet.  Just internal use for error
 358  * classification.
 359  */
 360 #define BACKREF_MISSING         (1 << 0) /* Backref missing in extent tree */
 361 #define BACKREF_MISMATCH        (1 << 1) /* Backref exists but does not match */
 362 #define BYTES_UNALIGNED         (1 << 2) /* Some bytes are not aligned */
 363 #define REFERENCER_MISSING      (1 << 3) /* Referencer not found */
 364 #define REFERENCER_MISMATCH     (1 << 4) /* Referenceer found but does not match */
 365 #define CROSSING_STRIPE_BOUNDARY (1 << 4) /* For kernel scrub workaround */
 366 #define ITEM_SIZE_MISMATCH      (1 << 5) /* Bad item size */
 367 #define UNKNOWN_TYPE            (1 << 6) /* Unknown type */
 368 #define ACCOUNTING_MISMATCH     (1 << 7) /* Used space accounting error */
 369 #define CHUNK_TYPE_MISMATCH     (1 << 8)
 370
 371 static void *print_status_check(void *p)
 372 {
 373         struct task_ctx *priv = p;
 374         const char work_indicator[] = { '.', 'o', 'O', 'o' };
 375         uint32_t count = 0;
 376         static char *task_position_string[] = {
 377                 "checking extents",
 378                 "checking free space cache",
 379                 "checking fs roots",
 380         };
 381
 382         task_period_start(priv->info, 1000 /* 1s */);
 383
 384         if (priv->tp == TASK_NOTHING)
 385                 return NULL;
 386
 387         while (1) {
 388                 printf("%s [%c]\r", task_position_string[priv->tp],
 389                                 work_indicator[count % 4]);
 390                 count++;
 391                 fflush(stdout);
 392                 task_period_wait(priv->info);
 393         }
 394         return NULL;
 395 }
 396
 397 static int print_status_return(void *p)
 398 {
 399         printf("\n");
 400         fflush(stdout);
 401
 402         return 0;
 403 }
 404
 405 static enum btrfs_check_mode parse_check_mode(const char *str)
 406 {
 407         if (strcmp(str, "lowmem") == 0)
 408                 return CHECK_MODE_LOWMEM;
 409         if (strcmp(str, "orig") == 0)
 410                 return CHECK_MODE_ORIGINAL;
 411         if (strcmp(str, "original") == 0)
 412                 return CHECK_MODE_ORIGINAL;
 413
 414         return CHECK_MODE_UNKNOWN;
 415 }
 416
 417 /* Compatible function to allow reuse of old codes */
 418 static u64 first_extent_gap(struct rb_root *holes)
 419 {
 420         struct file_extent_hole *hole;
 421
 422         if (RB_EMPTY_ROOT(holes))
 423                 return (u64)-1;
 424
 425         hole = rb_entry(rb_first(holes), struct file_extent_hole, node);
 426         return hole->start;
 427 }
 428
 429 static int compare_hole(struct rb_node *node1, struct rb_node *node2)
 430 {
 431         struct file_extent_hole *hole1;
 432         struct file_extent_hole *hole2;
 433
 434         hole1 = rb_entry(node1, struct file_extent_hole, node);
 435         hole2 = rb_entry(node2, struct file_extent_hole, node);
 436
 437         if (hole1->start > hole2->start)
 438                 return -1;
 439         if (hole1->start < hole2->start)
 440                 return 1;
 441         /* Now hole1->start == hole2->start */
 442         if (hole1->len >= hole2->len)
 443                 /*
 444                  * Hole 1 will be merge center
 445                  * Same hole will be merged later
 446                  */
 447                 return -1;
 448         /* Hole 2 will be merge center */
 449         return 1;
 450 }
 451
 452 /*
 453  * Add a hole to the record
 454  *
 455  * This will do hole merge for copy_file_extent_holes(),
 456  * which will ensure there won't be continuous holes.
 457  */
 458 static int add_file_extent_hole(struct rb_root *holes,
 459                                 u64 start, u64 len)
 460 {
 461         struct file_extent_hole *hole;
 462         struct file_extent_hole *prev = NULL;
 463         struct file_extent_hole *next = NULL;
 464
 465         hole = malloc(sizeof(*hole));
 466         if (!hole)
 467                 return -ENOMEM;
 468         hole->start = start;
 469         hole->len = len;
 470         /* Since compare will not return 0, no -EEXIST will happen */
 471         rb_insert(holes, &hole->node, compare_hole);
 472
 473         /* simple merge with previous hole */
 474         if (rb_prev(&hole->node))
 475                 prev = rb_entry(rb_prev(&hole->node), struct file_extent_hole,
 476                                 node);
 477         if (prev && prev->start + prev->len >= hole->start) {
 478                 hole->len = hole->start + hole->len - prev->start;
 479                 hole->start = prev->start;
 480                 rb_erase(&prev->node, holes);
 481                 free(prev);
 482                 prev = NULL;
 483         }
 484
 485         /* iterate merge with next holes */
 486         while (1) {
 487                 if (!rb_next(&hole->node))
 488                         break;
 489                 next = rb_entry(rb_next(&hole->node), struct file_extent_hole,
 490                                         node);
 491                 if (hole->start + hole->len >= next->start) {
 492                         if (hole->start + hole->len <= next->start + next->len)
 493                                 hole->len = next->start + next->len -
 494                                             hole->start;
 495                         rb_erase(&next->node, holes);
 496                         free(next);
 497                         next = NULL;
 498                 } else
 499                         break;
 500         }
 501         return 0;
 502 }
 503
 504 static int compare_hole_range(struct rb_node *node, void *data)
 505 {
 506         struct file_extent_hole *hole;
 507         u64 start;
 508
 509         hole = (struct file_extent_hole *)data;
 510         start = hole->start;
 511
 512         hole = rb_entry(node, struct file_extent_hole, node);
 513         if (start < hole->start)
 514                 return -1;
 515         if (start >= hole->start && start < hole->start + hole->len)
 516                 return 0;
 517         return 1;
 518 }
 519
 520 /*
 521  * Delete a hole in the record
 522  *
 523  * This will do the hole split and is much restrict than add.
 524  */
 525 static int del_file_extent_hole(struct rb_root *holes,
 526                                 u64 start, u64 len)
 527 {
 528         struct file_extent_hole *hole;
 529         struct file_extent_hole tmp;
 530         u64 prev_start = 0;
 531         u64 prev_len = 0;
 532         u64 next_start = 0;
 533         u64 next_len = 0;
 534         struct rb_node *node;
 535         int have_prev = 0;
 536         int have_next = 0;
 537         int ret = 0;
 538
 539         tmp.start = start;
 540         tmp.len = len;
 541         node = rb_search(holes, &tmp, compare_hole_range, NULL);
 542         if (!node)
 543                 return -EEXIST;
 544         hole = rb_entry(node, struct file_extent_hole, node);
 545         if (start + len > hole->start + hole->len)
 546                 return -EEXIST;
 547
 548         /*
 549          * Now there will be no overlap, delete the hole and re-add the
 550          * split(s) if they exists.
 551          */
 552         if (start > hole->start) {
 553                 prev_start = hole->start;
 554                 prev_len = start - hole->start;
 555                 have_prev = 1;
 556         }
 557         if (hole->start + hole->len > start + len) {
 558                 next_start = start + len;
 559                 next_len = hole->start + hole->len - start - len;
 560                 have_next = 1;
 561         }
 562         rb_erase(node, holes);
 563         free(hole);
 564         if (have_prev) {
 565                 ret = add_file_extent_hole(holes, prev_start, prev_len);
 566                 if (ret < 0)
 567                         return ret;
 568         }
 569         if (have_next) {
 570                 ret = add_file_extent_hole(holes, next_start, next_len);
 571                 if (ret < 0)
 572                         return ret;
 573         }
 574         return 0;
 575 }
 576
 577 static int copy_file_extent_holes(struct rb_root *dst,
 578                                   struct rb_root *src)
 579 {
 580         struct file_extent_hole *hole;
 581         struct rb_node *node;
 582         int ret = 0;
 583
 584         node = rb_first(src);
 585         while (node) {
 586                 hole = rb_entry(node, struct file_extent_hole, node);
 587                 ret = add_file_extent_hole(dst, hole->start, hole->len);
 588                 if (ret)
 589                         break;
 590                 node = rb_next(node);
 591         }
 592         return ret;
 593 }
 594
 595 static void free_file_extent_holes(struct rb_root *holes)
 596 {
 597         struct rb_node *node;
 598         struct file_extent_hole *hole;
 599
 600         node = rb_first(holes);
 601         while (node) {
 602                 hole = rb_entry(node, struct file_extent_hole, node);
 603                 rb_erase(node, holes);
 604                 free(hole);
 605                 node = rb_first(holes);
 606         }
 607 }
 608
 609 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info);
 610
 611 static void record_root_in_trans(struct btrfs_trans_handle *trans,
 612                                  struct btrfs_root *root)
 613 {
 614         if (root->last_trans != trans->transid) {
 615                 root->track_dirty = 1;
 616                 root->last_trans = trans->transid;
 617                 root->commit_root = root->node;
 618                 extent_buffer_get(root->node);
 619         }
 620 }
 621
 622 static u8 imode_to_type(u32 imode)
 623 {
 624 #define S_SHIFT 12
 625         static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
 626                 [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
 627                 [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
 628                 [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
 629                 [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
 630                 [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
 631                 [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
 632                 [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
 633         };
 634
 635         return btrfs_type_by_mode[(imode & S_IFMT) >> S_SHIFT];
 636 #undef S_SHIFT
 637 }
 638
 639 static int device_record_compare(struct rb_node *node1, struct rb_node *node2)
 640 {
 641         struct device_record *rec1;
 642         struct device_record *rec2;
 643
 644         rec1 = rb_entry(node1, struct device_record, node);
 645         rec2 = rb_entry(node2, struct device_record, node);
 646         if (rec1->devid > rec2->devid)
 647                 return -1;
 648         else if (rec1->devid < rec2->devid)
 649                 return 1;
 650         else
 651                 return 0;
 652 }
 653
 654 static struct inode_record *clone_inode_rec(struct inode_record *orig_rec)
 655 {
 656         struct inode_record *rec;
 657         struct inode_backref *backref;
 658         struct inode_backref *orig;
 659         struct inode_backref *tmp;
 660         struct orphan_data_extent *src_orphan;
 661         struct orphan_data_extent *dst_orphan;
 662         struct rb_node *rb;
 663         size_t size;
 664         int ret;
 665
 666         rec = malloc(sizeof(*rec));
 667         if (!rec)
 668                 return ERR_PTR(-ENOMEM);
 669         memcpy(rec, orig_rec, sizeof(*rec));
 670         rec->refs = 1;
 671         INIT_LIST_HEAD(&rec->backrefs);
 672         INIT_LIST_HEAD(&rec->orphan_extents);
 673         rec->holes = RB_ROOT;
 674
 675         list_for_each_entry(orig, &orig_rec->backrefs, list) {
 676                 size = sizeof(*orig) + orig->namelen + 1;
 677                 backref = malloc(size);
 678                 if (!backref) {
 679                         ret = -ENOMEM;
 680                         goto cleanup;
 681                 }
 682                 memcpy(backref, orig, size);
 683                 list_add_tail(&backref->list, &rec->backrefs);
 684         }
 685         list_for_each_entry(src_orphan, &orig_rec->orphan_extents, list) {
 686                 dst_orphan = malloc(sizeof(*dst_orphan));
 687                 if (!dst_orphan) {
 688                         ret = -ENOMEM;
 689                         goto cleanup;
 690                 }
 691                 memcpy(dst_orphan, src_orphan, sizeof(*src_orphan));
 692                 list_add_tail(&dst_orphan->list, &rec->orphan_extents);
 693         }
 694         ret = copy_file_extent_holes(&rec->holes, &orig_rec->holes);
 695         if (ret < 0)
 696                 goto cleanup_rb;
 697
 698         return rec;
 699
 700 cleanup_rb:
 701         rb = rb_first(&rec->holes);
 702         while (rb) {
 703                 struct file_extent_hole *hole;
 704
 705                 hole = rb_entry(rb, struct file_extent_hole, node);
 706                 rb = rb_next(rb);
 707                 free(hole);
 708         }
 709
 710 cleanup:
 711         if (!list_empty(&rec->backrefs))
 712                 list_for_each_entry_safe(orig, tmp, &rec->backrefs, list) {
 713                         list_del(&orig->list);
 714                         free(orig);
 715                 }
 716
 717         if (!list_empty(&rec->orphan_extents))
 718                 list_for_each_entry_safe(orig, tmp, &rec->orphan_extents, list) {
 719                         list_del(&orig->list);
 720                         free(orig);
 721                 }
 722
 723         free(rec);
 724
 725         return ERR_PTR(ret);
 726 }
 727
 728 static void print_orphan_data_extents(struct list_head *orphan_extents,
 729                                       u64 objectid)
 730 {
 731         struct orphan_data_extent *orphan;
 732
 733         if (list_empty(orphan_extents))
 734                 return;
 735         printf("The following data extent is lost in tree %llu:\n",
 736                objectid);
 737         list_for_each_entry(orphan, orphan_extents, list) {
 738                 printf("\tinode: %llu, offset:%llu, disk_bytenr: %llu, disk_len: %llu\n",
 739                        orphan->objectid, orphan->offset, orphan->disk_bytenr,
 740                        orphan->disk_len);
 741         }
 742 }
 743
 744 static void print_inode_error(struct btrfs_root *root, struct inode_record *rec)
 745 {
 746         u64 root_objectid = root->root_key.objectid;
 747         int errors = rec->errors;
 748
 749         if (!errors)
 750                 return;
 751         /* reloc root errors, we print its corresponding fs root objectid*/
 752         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
 753                 root_objectid = root->root_key.offset;
 754                 fprintf(stderr, "reloc");
 755         }
 756         fprintf(stderr, "root %llu inode %llu errors %x",
 757                 (unsigned long long) root_objectid,
 758                 (unsigned long long) rec->ino, rec->errors);
 759
 760         if (errors & I_ERR_NO_INODE_ITEM)
 761                 fprintf(stderr, ", no inode item");
 762         if (errors & I_ERR_NO_ORPHAN_ITEM)
 763                 fprintf(stderr, ", no orphan item");
 764         if (errors & I_ERR_DUP_INODE_ITEM)
 765                 fprintf(stderr, ", dup inode item");
 766         if (errors & I_ERR_DUP_DIR_INDEX)
 767                 fprintf(stderr, ", dup dir index");
 768         if (errors & I_ERR_ODD_DIR_ITEM)
 769                 fprintf(stderr, ", odd dir item");
 770         if (errors & I_ERR_ODD_FILE_EXTENT)
 771                 fprintf(stderr, ", odd file extent");
 772         if (errors & I_ERR_BAD_FILE_EXTENT)
 773                 fprintf(stderr, ", bad file extent");
 774         if (errors & I_ERR_FILE_EXTENT_OVERLAP)
 775                 fprintf(stderr, ", file extent overlap");
 776         if (errors & I_ERR_FILE_EXTENT_DISCOUNT)
 777                 fprintf(stderr, ", file extent discount");
 778         if (errors & I_ERR_DIR_ISIZE_WRONG)
 779                 fprintf(stderr, ", dir isize wrong");
 780         if (errors & I_ERR_FILE_NBYTES_WRONG)
 781                 fprintf(stderr, ", nbytes wrong");
 782         if (errors & I_ERR_ODD_CSUM_ITEM)
 783                 fprintf(stderr, ", odd csum item");
 784         if (errors & I_ERR_SOME_CSUM_MISSING)
 785                 fprintf(stderr, ", some csum missing");
 786         if (errors & I_ERR_LINK_COUNT_WRONG)
 787                 fprintf(stderr, ", link count wrong");
 788         if (errors & I_ERR_FILE_EXTENT_ORPHAN)
 789                 fprintf(stderr, ", orphan file extent");
 790         fprintf(stderr, "\n");
 791         /* Print the orphan extents if needed */
 792         if (errors & I_ERR_FILE_EXTENT_ORPHAN)
 793                 print_orphan_data_extents(&rec->orphan_extents, root->objectid);
 794
 795         /* Print the holes if needed */
 796         if (errors & I_ERR_FILE_EXTENT_DISCOUNT) {
 797                 struct file_extent_hole *hole;
 798                 struct rb_node *node;
 799                 int found = 0;
 800
 801                 node = rb_first(&rec->holes);
 802                 fprintf(stderr, "Found file extent holes:\n");
 803                 while (node) {
 804                         found = 1;
 805                         hole = rb_entry(node, struct file_extent_hole, node);
 806                         fprintf(stderr, "\tstart: %llu, len: %llu\n",
 807                                 hole->start, hole->len);
 808                         node = rb_next(node);
 809                 }
 810                 if (!found)
 811                         fprintf(stderr, "\tstart: 0, len: %llu\n",
 812                                 round_up(rec->isize, root->sectorsize));
 813         }
 814 }
 815
 816 static void print_ref_error(int errors)
 817 {
 818         if (errors & REF_ERR_NO_DIR_ITEM)
 819                 fprintf(stderr, ", no dir item");
 820         if (errors & REF_ERR_NO_DIR_INDEX)
 821                 fprintf(stderr, ", no dir index");
 822         if (errors & REF_ERR_NO_INODE_REF)
 823                 fprintf(stderr, ", no inode ref");
 824         if (errors & REF_ERR_DUP_DIR_ITEM)
 825                 fprintf(stderr, ", dup dir item");
 826         if (errors & REF_ERR_DUP_DIR_INDEX)
 827                 fprintf(stderr, ", dup dir index");
 828         if (errors & REF_ERR_DUP_INODE_REF)
 829                 fprintf(stderr, ", dup inode ref");
 830         if (errors & REF_ERR_INDEX_UNMATCH)
 831                 fprintf(stderr, ", index mismatch");
 832         if (errors & REF_ERR_FILETYPE_UNMATCH)
 833                 fprintf(stderr, ", filetype mismatch");
 834         if (errors & REF_ERR_NAME_TOO_LONG)
 835                 fprintf(stderr, ", name too long");
 836         if (errors & REF_ERR_NO_ROOT_REF)
 837                 fprintf(stderr, ", no root ref");
 838         if (errors & REF_ERR_NO_ROOT_BACKREF)
 839                 fprintf(stderr, ", no root backref");
 840         if (errors & REF_ERR_DUP_ROOT_REF)
 841                 fprintf(stderr, ", dup root ref");
 842         if (errors & REF_ERR_DUP_ROOT_BACKREF)
 843                 fprintf(stderr, ", dup root backref");
 844         fprintf(stderr, "\n");
 845 }
 846
 847 static struct inode_record *get_inode_rec(struct cache_tree *inode_cache,
 848                                           u64 ino, int mod)
 849 {
 850         struct ptr_node *node;
 851         struct cache_extent *cache;
 852         struct inode_record *rec = NULL;
 853         int ret;
 854
 855         cache = lookup_cache_extent(inode_cache, ino, 1);
 856         if (cache) {
 857                 node = container_of(cache, struct ptr_node, cache);
 858                 rec = node->data;
 859                 if (mod && rec->refs > 1) {
 860                         node->data = clone_inode_rec(rec);
 861                         if (IS_ERR(node->data))
 862                                 return node->data;
 863                         rec->refs--;
 864                         rec = node->data;
 865                 }
 866         } else if (mod) {
 867                 rec = calloc(1, sizeof(*rec));
 868                 if (!rec)
 869                         return ERR_PTR(-ENOMEM);
 870                 rec->ino = ino;
 871                 rec->extent_start = (u64)-1;
 872                 rec->refs = 1;
 873                 INIT_LIST_HEAD(&rec->backrefs);
 874                 INIT_LIST_HEAD(&rec->orphan_extents);
 875                 rec->holes = RB_ROOT;
 876
 877                 node = malloc(sizeof(*node));
 878                 if (!node) {
 879                         free(rec);
 880                         return ERR_PTR(-ENOMEM);
 881                 }
 882                 node->cache.start = ino;
 883                 node->cache.size = 1;
 884                 node->data = rec;
 885
 886                 if (ino == BTRFS_FREE_INO_OBJECTID)
 887                         rec->found_link = 1;
 888
 889                 ret = insert_cache_extent(inode_cache, &node->cache);
 890                 if (ret)
 891                         return ERR_PTR(-EEXIST);
 892         }
 893         return rec;
 894 }
 895
 896 static void free_orphan_data_extents(struct list_head *orphan_extents)
 897 {
 898         struct orphan_data_extent *orphan;
 899
 900         while (!list_empty(orphan_extents)) {
 901                 orphan = list_entry(orphan_extents->next,
 902                                     struct orphan_data_extent, list);
 903                 list_del(&orphan->list);
 904                 free(orphan);
 905         }
 906 }
 907
 908 static void free_inode_rec(struct inode_record *rec)
 909 {
 910         struct inode_backref *backref;
 911
 912         if (--rec->refs > 0)
 913                 return;
 914
 915         while (!list_empty(&rec->backrefs)) {
 916                 backref = to_inode_backref(rec->backrefs.next);
 917                 list_del(&backref->list);
 918                 free(backref);
 919         }
 920         free_orphan_data_extents(&rec->orphan_extents);
 921         free_file_extent_holes(&rec->holes);
 922         free(rec);
 923 }
 924
 925 static int can_free_inode_rec(struct inode_record *rec)
 926 {
 927         if (!rec->errors && rec->checked && rec->found_inode_item &&
 928             rec->nlink == rec->found_link && list_empty(&rec->backrefs))
 929                 return 1;
 930         return 0;
 931 }
 932
 933 static void maybe_free_inode_rec(struct cache_tree *inode_cache,
 934                                  struct inode_record *rec)
 935 {
 936         struct cache_extent *cache;
 937         struct inode_backref *tmp, *backref;
 938         struct ptr_node *node;
 939         u8 filetype;
 940
 941         if (!rec->found_inode_item)
 942                 return;
 943
 944         filetype = imode_to_type(rec->imode);
 945         list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
 946                 if (backref->found_dir_item && backref->found_dir_index) {
 947                         if (backref->filetype != filetype)
 948                                 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
 949                         if (!backref->errors && backref->found_inode_ref &&
 950                             rec->nlink == rec->found_link) {
 951                                 list_del(&backref->list);
 952                                 free(backref);
 953                         }
 954                 }
 955         }
 956
 957         if (!rec->checked || rec->merging)
 958                 return;
 959
 960         if (S_ISDIR(rec->imode)) {
 961                 if (rec->found_size != rec->isize)
 962                         rec->errors |= I_ERR_DIR_ISIZE_WRONG;
 963                 if (rec->found_file_extent)
 964                         rec->errors |= I_ERR_ODD_FILE_EXTENT;
 965         } else if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
 966                 if (rec->found_dir_item)
 967                         rec->errors |= I_ERR_ODD_DIR_ITEM;
 968                 if (rec->found_size != rec->nbytes)
 969                         rec->errors |= I_ERR_FILE_NBYTES_WRONG;
 970                 if (rec->nlink > 0 && !no_holes &&
 971                     (rec->extent_end < rec->isize ||
 972                      first_extent_gap(&rec->holes) < rec->isize))
 973                         rec->errors |= I_ERR_FILE_EXTENT_DISCOUNT;
 974         }
 975
 976         if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
 977                 if (rec->found_csum_item && rec->nodatasum)
 978                         rec->errors |= I_ERR_ODD_CSUM_ITEM;
 979                 if (rec->some_csum_missing && !rec->nodatasum)
 980                         rec->errors |= I_ERR_SOME_CSUM_MISSING;
 981         }
 982
 983         BUG_ON(rec->refs != 1);
 984         if (can_free_inode_rec(rec)) {
 985                 cache = lookup_cache_extent(inode_cache, rec->ino, 1);
 986                 node = container_of(cache, struct ptr_node, cache);
 987                 BUG_ON(node->data != rec);
 988                 remove_cache_extent(inode_cache, &node->cache);
 989                 free(node);
 990                 free_inode_rec(rec);
 991         }
 992 }
 993
 994 static int check_orphan_item(struct btrfs_root *root, u64 ino)
 995 {
 996         struct btrfs_path path;
 997         struct btrfs_key key;
 998         int ret;
 999
1000         key.objectid = BTRFS_ORPHAN_OBJECTID;
1001         key.type = BTRFS_ORPHAN_ITEM_KEY;
1002         key.offset = ino;
1003
1004         btrfs_init_path(&path);
1005         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
1006         btrfs_release_path(&path);
1007         if (ret > 0)
1008                 ret = -ENOENT;
1009         return ret;
1010 }
1011
1012 static int process_inode_item(struct extent_buffer *eb,
1013                               int slot, struct btrfs_key *key,
1014                               struct shared_node *active_node)
1015 {
1016         struct inode_record *rec;
1017         struct btrfs_inode_item *item;
1018
1019         rec = active_node->current;
1020         BUG_ON(rec->ino != key->objectid || rec->refs > 1);
1021         if (rec->found_inode_item) {
1022                 rec->errors |= I_ERR_DUP_INODE_ITEM;
1023                 return 1;
1024         }
1025         item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
1026         rec->nlink = btrfs_inode_nlink(eb, item);
1027         rec->isize = btrfs_inode_size(eb, item);
1028         rec->nbytes = btrfs_inode_nbytes(eb, item);
1029         rec->imode = btrfs_inode_mode(eb, item);
1030         if (btrfs_inode_flags(eb, item) & BTRFS_INODE_NODATASUM)
1031                 rec->nodatasum = 1;
1032         rec->found_inode_item = 1;
1033         if (rec->nlink == 0)
1034                 rec->errors |= I_ERR_NO_ORPHAN_ITEM;
1035         maybe_free_inode_rec(&active_node->inode_cache, rec);
1036         return 0;
1037 }
1038
1039 static struct inode_backref *get_inode_backref(struct inode_record *rec,
1040                                                 const char *name,
1041                                                 int namelen, u64 dir)
1042 {
1043         struct inode_backref *backref;
1044
1045         list_for_each_entry(backref, &rec->backrefs, list) {
1046                 if (rec->ino == BTRFS_MULTIPLE_OBJECTIDS)
1047                         break;
1048                 if (backref->dir != dir || backref->namelen != namelen)
1049                         continue;
1050                 if (memcmp(name, backref->name, namelen))
1051                         continue;
1052                 return backref;
1053         }
1054
1055         backref = malloc(sizeof(*backref) + namelen + 1);
1056         if (!backref)
1057                 return NULL;
1058         memset(backref, 0, sizeof(*backref));
1059         backref->dir = dir;
1060         backref->namelen = namelen;
1061         memcpy(backref->name, name, namelen);
1062         backref->name[namelen] = '\0';
1063         list_add_tail(&backref->list, &rec->backrefs);
1064         return backref;
1065 }
1066
1067 static int add_inode_backref(struct cache_tree *inode_cache,
1068                              u64 ino, u64 dir, u64 index,
1069                              const char *name, int namelen,
1070                              u8 filetype, u8 itemtype, int errors)
1071 {
1072         struct inode_record *rec;
1073         struct inode_backref *backref;
1074
1075         rec = get_inode_rec(inode_cache, ino, 1);
1076         BUG_ON(IS_ERR(rec));
1077         backref = get_inode_backref(rec, name, namelen, dir);
1078         BUG_ON(!backref);
1079         if (errors)
1080                 backref->errors |= errors;
1081         if (itemtype == BTRFS_DIR_INDEX_KEY) {
1082                 if (backref->found_dir_index)
1083                         backref->errors |= REF_ERR_DUP_DIR_INDEX;
1084                 if (backref->found_inode_ref && backref->index != index)
1085                         backref->errors |= REF_ERR_INDEX_UNMATCH;
1086                 if (backref->found_dir_item && backref->filetype != filetype)
1087                         backref->errors |= REF_ERR_FILETYPE_UNMATCH;
1088
1089                 backref->index = index;
1090                 backref->filetype = filetype;
1091                 backref->found_dir_index = 1;
1092         } else if (itemtype == BTRFS_DIR_ITEM_KEY) {
1093                 rec->found_link++;
1094                 if (backref->found_dir_item)
1095                         backref->errors |= REF_ERR_DUP_DIR_ITEM;
1096                 if (backref->found_dir_index && backref->filetype != filetype)
1097                         backref->errors |= REF_ERR_FILETYPE_UNMATCH;
1098
1099                 backref->filetype = filetype;
1100                 backref->found_dir_item = 1;
1101         } else if ((itemtype == BTRFS_INODE_REF_KEY) ||
1102                    (itemtype == BTRFS_INODE_EXTREF_KEY)) {
1103                 if (backref->found_inode_ref)
1104                         backref->errors |= REF_ERR_DUP_INODE_REF;
1105                 if (backref->found_dir_index && backref->index != index)
1106                         backref->errors |= REF_ERR_INDEX_UNMATCH;
1107                 else
1108                         backref->index = index;
1109
1110                 backref->ref_type = itemtype;
1111                 backref->found_inode_ref = 1;
1112         } else {
1113                 BUG_ON(1);
1114         }
1115
1116         maybe_free_inode_rec(inode_cache, rec);
1117         return 0;
1118 }
1119
1120 static int merge_inode_recs(struct inode_record *src, struct inode_record *dst,
1121                             struct cache_tree *dst_cache)
1122 {
1123         struct inode_backref *backref;
1124         u32 dir_count = 0;
1125         int ret = 0;
1126
1127         dst->merging = 1;
1128         list_for_each_entry(backref, &src->backrefs, list) {
1129                 if (backref->found_dir_index) {
1130                         add_inode_backref(dst_cache, dst->ino, backref->dir,
1131                                         backref->index, backref->name,
1132                                         backref->namelen, backref->filetype,
1133                                         BTRFS_DIR_INDEX_KEY, backref->errors);
1134                 }
1135                 if (backref->found_dir_item) {
1136                         dir_count++;
1137                         add_inode_backref(dst_cache, dst->ino,
1138                                         backref->dir, 0, backref->name,
1139                                         backref->namelen, backref->filetype,
1140                                         BTRFS_DIR_ITEM_KEY, backref->errors);
1141                 }
1142                 if (backref->found_inode_ref) {
1143                         add_inode_backref(dst_cache, dst->ino,
1144                                         backref->dir, backref->index,
1145                                         backref->name, backref->namelen, 0,
1146                                         backref->ref_type, backref->errors);
1147                 }
1148         }
1149
1150         if (src->found_dir_item)
1151                 dst->found_dir_item = 1;
1152         if (src->found_file_extent)
1153                 dst->found_file_extent = 1;
1154         if (src->found_csum_item)
1155                 dst->found_csum_item = 1;
1156         if (src->some_csum_missing)
1157                 dst->some_csum_missing = 1;
1158         if (first_extent_gap(&dst->holes) > first_extent_gap(&src->holes)) {
1159                 ret = copy_file_extent_holes(&dst->holes, &src->holes);
1160                 if (ret < 0)
1161                         return ret;
1162         }
1163
1164         BUG_ON(src->found_link < dir_count);
1165         dst->found_link += src->found_link - dir_count;
1166         dst->found_size += src->found_size;
1167         if (src->extent_start != (u64)-1) {
1168                 if (dst->extent_start == (u64)-1) {
1169                         dst->extent_start = src->extent_start;
1170                         dst->extent_end = src->extent_end;
1171                 } else {
1172                         if (dst->extent_end > src->extent_start)
1173                                 dst->errors |= I_ERR_FILE_EXTENT_OVERLAP;
1174                         else if (dst->extent_end < src->extent_start) {
1175                                 ret = add_file_extent_hole(&dst->holes,
1176                                         dst->extent_end,
1177                                         src->extent_start - dst->extent_end);
1178                         }
1179                         if (dst->extent_end < src->extent_end)
1180                                 dst->extent_end = src->extent_end;
1181                 }
1182         }
1183
1184         dst->errors |= src->errors;
1185         if (src->found_inode_item) {
1186                 if (!dst->found_inode_item) {
1187                         dst->nlink = src->nlink;
1188                         dst->isize = src->isize;
1189                         dst->nbytes = src->nbytes;
1190                         dst->imode = src->imode;
1191                         dst->nodatasum = src->nodatasum;
1192                         dst->found_inode_item = 1;
1193                 } else {
1194                         dst->errors |= I_ERR_DUP_INODE_ITEM;
1195                 }
1196         }
1197         dst->merging = 0;
1198
1199         return 0;
1200 }
1201
1202 static int splice_shared_node(struct shared_node *src_node,
1203                               struct shared_node *dst_node)
1204 {
1205         struct cache_extent *cache;
1206         struct ptr_node *node, *ins;
1207         struct cache_tree *src, *dst;
1208         struct inode_record *rec, *conflict;
1209         u64 current_ino = 0;
1210         int splice = 0;
1211         int ret;
1212
1213         if (--src_node->refs == 0)
1214                 splice = 1;
1215         if (src_node->current)
1216                 current_ino = src_node->current->ino;
1217
1218         src = &src_node->root_cache;
1219         dst = &dst_node->root_cache;
1220 again:
1221         cache = search_cache_extent(src, 0);
1222         while (cache) {
1223                 node = container_of(cache, struct ptr_node, cache);
1224                 rec = node->data;
1225                 cache = next_cache_extent(cache);
1226
1227                 if (splice) {
1228                         remove_cache_extent(src, &node->cache);
1229                         ins = node;
1230                 } else {
1231                         ins = malloc(sizeof(*ins));
1232                         BUG_ON(!ins);
1233                         ins->cache.start = node->cache.start;
1234                         ins->cache.size = node->cache.size;
1235                         ins->data = rec;
1236                         rec->refs++;
1237                 }
1238                 ret = insert_cache_extent(dst, &ins->cache);
1239                 if (ret == -EEXIST) {
1240                         conflict = get_inode_rec(dst, rec->ino, 1);
1241                         BUG_ON(IS_ERR(conflict));
1242                         merge_inode_recs(rec, conflict, dst);
1243                         if (rec->checked) {
1244                                 conflict->checked = 1;
1245                                 if (dst_node->current == conflict)
1246                                         dst_node->current = NULL;
1247                         }
1248                         maybe_free_inode_rec(dst, conflict);
1249                         free_inode_rec(rec);
1250                         free(ins);
1251                 } else {
1252                         BUG_ON(ret);
1253                 }
1254         }
1255
1256         if (src == &src_node->root_cache) {
1257                 src = &src_node->inode_cache;
1258                 dst = &dst_node->inode_cache;
1259                 goto again;
1260         }
1261
1262         if (current_ino > 0 && (!dst_node->current ||
1263             current_ino > dst_node->current->ino)) {
1264                 if (dst_node->current) {
1265                         dst_node->current->checked = 1;
1266                         maybe_free_inode_rec(dst, dst_node->current);
1267                 }
1268                 dst_node->current = get_inode_rec(dst, current_ino, 1);
1269                 BUG_ON(IS_ERR(dst_node->current));
1270         }
1271         return 0;
1272 }
1273
1274 static void free_inode_ptr(struct cache_extent *cache)
1275 {
1276         struct ptr_node *node;
1277         struct inode_record *rec;
1278
1279         node = container_of(cache, struct ptr_node, cache);
1280         rec = node->data;
1281         free_inode_rec(rec);
1282         free(node);
1283 }
1284
1285 FREE_EXTENT_CACHE_BASED_TREE(inode_recs, free_inode_ptr);
1286
1287 static struct shared_node *find_shared_node(struct cache_tree *shared,
1288                                             u64 bytenr)
1289 {
1290         struct cache_extent *cache;
1291         struct shared_node *node;
1292
1293         cache = lookup_cache_extent(shared, bytenr, 1);
1294         if (cache) {
1295                 node = container_of(cache, struct shared_node, cache);
1296                 return node;
1297         }
1298         return NULL;
1299 }
1300
1301 static int add_shared_node(struct cache_tree *shared, u64 bytenr, u32 refs)
1302 {
1303         int ret;
1304         struct shared_node *node;
1305
1306         node = calloc(1, sizeof(*node));
1307         if (!node)
1308                 return -ENOMEM;
1309         node->cache.start = bytenr;
1310         node->cache.size = 1;
1311         cache_tree_init(&node->root_cache);
1312         cache_tree_init(&node->inode_cache);
1313         node->refs = refs;
1314
1315         ret = insert_cache_extent(shared, &node->cache);
1316
1317         return ret;
1318 }
1319
1320 static int enter_shared_node(struct btrfs_root *root, u64 bytenr, u32 refs,
1321                              struct walk_control *wc, int level)
1322 {
1323         struct shared_node *node;
1324         struct shared_node *dest;
1325         int ret;
1326
1327         if (level == wc->active_node)
1328                 return 0;
1329
1330         BUG_ON(wc->active_node <= level);
1331         node = find_shared_node(&wc->shared, bytenr);
1332         if (!node) {
1333                 ret = add_shared_node(&wc->shared, bytenr, refs);
1334                 BUG_ON(ret);
1335                 node = find_shared_node(&wc->shared, bytenr);
1336                 wc->nodes[level] = node;
1337                 wc->active_node = level;
1338                 return 0;
1339         }
1340
1341         if (wc->root_level == wc->active_node &&
1342             btrfs_root_refs(&root->root_item) == 0) {
1343                 if (--node->refs == 0) {
1344                         free_inode_recs_tree(&node->root_cache);
1345                         free_inode_recs_tree(&node->inode_cache);
1346                         remove_cache_extent(&wc->shared, &node->cache);
1347                         free(node);
1348                 }
1349                 return 1;
1350         }
1351
1352         dest = wc->nodes[wc->active_node];
1353         splice_shared_node(node, dest);
1354         if (node->refs == 0) {
1355                 remove_cache_extent(&wc->shared, &node->cache);
1356                 free(node);
1357         }
1358         return 1;
1359 }
1360
1361 static int leave_shared_node(struct btrfs_root *root,
1362                              struct walk_control *wc, int level)
1363 {
1364         struct shared_node *node;
1365         struct shared_node *dest;
1366         int i;
1367
1368         if (level == wc->root_level)
1369                 return 0;
1370
1371         for (i = level + 1; i < BTRFS_MAX_LEVEL; i++) {
1372                 if (wc->nodes[i])
1373                         break;
1374         }
1375         BUG_ON(i >= BTRFS_MAX_LEVEL);
1376
1377         node = wc->nodes[wc->active_node];
1378         wc->nodes[wc->active_node] = NULL;
1379         wc->active_node = i;
1380
1381         dest = wc->nodes[wc->active_node];
1382         if (wc->active_node < wc->root_level ||
1383             btrfs_root_refs(&root->root_item) > 0) {
1384                 BUG_ON(node->refs <= 1);
1385                 splice_shared_node(node, dest);
1386         } else {
1387                 BUG_ON(node->refs < 2);
1388                 node->refs--;
1389         }
1390         return 0;
1391 }
1392
1393 /*
1394  * Returns:
1395  * < 0 - on error
1396  * 1   - if the root with id child_root_id is a child of root parent_root_id
1397  * 0   - if the root child_root_id isn't a child of the root parent_root_id but
1398  *       has other root(s) as parent(s)
1399  * 2   - if the root child_root_id doesn't have any parent roots
1400  */
1401 static int is_child_root(struct btrfs_root *root, u64 parent_root_id,
1402                          u64 child_root_id)
1403 {
1404         struct btrfs_path path;
1405         struct btrfs_key key;
1406         struct extent_buffer *leaf;
1407         int has_parent = 0;
1408         int ret;
1409
1410         btrfs_init_path(&path);
1411
1412         key.objectid = parent_root_id;
1413         key.type = BTRFS_ROOT_REF_KEY;
1414         key.offset = child_root_id;
1415         ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
1416                                 0, 0);
1417         if (ret < 0)
1418                 return ret;
1419         btrfs_release_path(&path);
1420         if (!ret)
1421                 return 1;
1422
1423         key.objectid = child_root_id;
1424         key.type = BTRFS_ROOT_BACKREF_KEY;
1425         key.offset = 0;
1426         ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
1427                                 0, 0);
1428         if (ret < 0)
1429                 goto out;
1430
1431         while (1) {
1432                 leaf = path.nodes[0];
1433                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1434                         ret = btrfs_next_leaf(root->fs_info->tree_root, &path);
1435                         if (ret)
1436                                 break;
1437                         leaf = path.nodes[0];
1438                 }
1439
1440                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1441                 if (key.objectid != child_root_id ||
1442                     key.type != BTRFS_ROOT_BACKREF_KEY)
1443                         break;
1444
1445                 has_parent = 1;
1446
1447                 if (key.offset == parent_root_id) {
1448                         btrfs_release_path(&path);
1449                         return 1;
1450                 }
1451
1452                 path.slots[0]++;
1453         }
1454 out:
1455         btrfs_release_path(&path);
1456         if (ret < 0)
1457                 return ret;
1458         return has_parent ? 0 : 2;
1459 }
1460
1461 static int process_dir_item(struct btrfs_root *root,
1462                             struct extent_buffer *eb,
1463                             int slot, struct btrfs_key *key,
1464                             struct shared_node *active_node)
1465 {
1466         u32 total;
1467         u32 cur = 0;
1468         u32 len;
1469         u32 name_len;
1470         u32 data_len;
1471         int error;
1472         int nritems = 0;
1473         u8 filetype;
1474         struct btrfs_dir_item *di;
1475         struct inode_record *rec;
1476         struct cache_tree *root_cache;
1477         struct cache_tree *inode_cache;
1478         struct btrfs_key location;
1479         char namebuf[BTRFS_NAME_LEN];
1480
1481         root_cache = &active_node->root_cache;
1482         inode_cache = &active_node->inode_cache;
1483         rec = active_node->current;
1484         rec->found_dir_item = 1;
1485
1486         di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1487         total = btrfs_item_size_nr(eb, slot);
1488         while (cur < total) {
1489                 nritems++;
1490                 btrfs_dir_item_key_to_cpu(eb, di, &location);
1491                 name_len = btrfs_dir_name_len(eb, di);
1492                 data_len = btrfs_dir_data_len(eb, di);
1493                 filetype = btrfs_dir_type(eb, di);
1494
1495                 rec->found_size += name_len;
1496                 if (name_len <= BTRFS_NAME_LEN) {
1497                         len = name_len;
1498                         error = 0;
1499                 } else {
1500                         len = BTRFS_NAME_LEN;
1501                         error = REF_ERR_NAME_TOO_LONG;
1502                 }
1503                 read_extent_buffer(eb, namebuf, (unsigned long)(di + 1), len);
1504
1505                 if (location.type == BTRFS_INODE_ITEM_KEY) {
1506                         add_inode_backref(inode_cache, location.objectid,
1507                                           key->objectid, key->offset, namebuf,
1508                                           len, filetype, key->type, error);
1509                 } else if (location.type == BTRFS_ROOT_ITEM_KEY) {
1510                         add_inode_backref(root_cache, location.objectid,
1511                                           key->objectid, key->offset,
1512                                           namebuf, len, filetype,
1513                                           key->type, error);
1514                 } else {
1515                         fprintf(stderr, "invalid location in dir item %u\n",
1516                                 location.type);
1517                         add_inode_backref(inode_cache, BTRFS_MULTIPLE_OBJECTIDS,
1518                                           key->objectid, key->offset, namebuf,
1519                                           len, filetype, key->type, error);
1520                 }
1521
1522                 len = sizeof(*di) + name_len + data_len;
1523                 di = (struct btrfs_dir_item *)((char *)di + len);
1524                 cur += len;
1525         }
1526         if (key->type == BTRFS_DIR_INDEX_KEY && nritems > 1)
1527                 rec->errors |= I_ERR_DUP_DIR_INDEX;
1528
1529         return 0;
1530 }
1531
1532 static int process_inode_ref(struct extent_buffer *eb,
1533                              int slot, struct btrfs_key *key,
1534                              struct shared_node *active_node)
1535 {
1536         u32 total;
1537         u32 cur = 0;
1538         u32 len;
1539         u32 name_len;
1540         u64 index;
1541         int error;
1542         struct cache_tree *inode_cache;
1543         struct btrfs_inode_ref *ref;
1544         char namebuf[BTRFS_NAME_LEN];
1545
1546         inode_cache = &active_node->inode_cache;
1547
1548         ref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1549         total = btrfs_item_size_nr(eb, slot);
1550         while (cur < total) {
1551                 name_len = btrfs_inode_ref_name_len(eb, ref);
1552                 index = btrfs_inode_ref_index(eb, ref);
1553                 if (name_len <= BTRFS_NAME_LEN) {
1554                         len = name_len;
1555                         error = 0;
1556                 } else {
1557                         len = BTRFS_NAME_LEN;
1558                         error = REF_ERR_NAME_TOO_LONG;
1559                 }
1560                 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
1561                 add_inode_backref(inode_cache, key->objectid, key->offset,
1562                                   index, namebuf, len, 0, key->type, error);
1563
1564                 len = sizeof(*ref) + name_len;
1565                 ref = (struct btrfs_inode_ref *)((char *)ref + len);
1566                 cur += len;
1567         }
1568         return 0;
1569 }
1570
1571 static int process_inode_extref(struct extent_buffer *eb,
1572                                 int slot, struct btrfs_key *key,
1573                                 struct shared_node *active_node)
1574 {
1575         u32 total;
1576         u32 cur = 0;
1577         u32 len;
1578         u32 name_len;
1579         u64 index;
1580         u64 parent;
1581         int error;
1582         struct cache_tree *inode_cache;
1583         struct btrfs_inode_extref *extref;
1584         char namebuf[BTRFS_NAME_LEN];
1585
1586         inode_cache = &active_node->inode_cache;
1587
1588         extref = btrfs_item_ptr(eb, slot, struct btrfs_inode_extref);
1589         total = btrfs_item_size_nr(eb, slot);
1590         while (cur < total) {
1591                 name_len = btrfs_inode_extref_name_len(eb, extref);
1592                 index = btrfs_inode_extref_index(eb, extref);
1593                 parent = btrfs_inode_extref_parent(eb, extref);
1594                 if (name_len <= BTRFS_NAME_LEN) {
1595                         len = name_len;
1596                         error = 0;
1597                 } else {
1598                         len = BTRFS_NAME_LEN;
1599                         error = REF_ERR_NAME_TOO_LONG;
1600                 }
1601                 read_extent_buffer(eb, namebuf,
1602                                    (unsigned long)(extref + 1), len);
1603                 add_inode_backref(inode_cache, key->objectid, parent,
1604                                   index, namebuf, len, 0, key->type, error);
1605
1606                 len = sizeof(*extref) + name_len;
1607                 extref = (struct btrfs_inode_extref *)((char *)extref + len);
1608                 cur += len;
1609         }
1610         return 0;
1611
1612 }
1613
1614 static int count_csum_range(struct btrfs_root *root, u64 start,
1615                             u64 len, u64 *found)
1616 {
1617         struct btrfs_key key;
1618         struct btrfs_path path;
1619         struct extent_buffer *leaf;
1620         int ret;
1621         size_t size;
1622         *found = 0;
1623         u64 csum_end;
1624         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
1625
1626         btrfs_init_path(&path);
1627
1628         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1629         key.offset = start;
1630         key.type = BTRFS_EXTENT_CSUM_KEY;
1631
1632         ret = btrfs_search_slot(NULL, root->fs_info->csum_root,
1633                                 &key, &path, 0, 0);
1634         if (ret < 0)
1635                 goto out;
1636         if (ret > 0 && path.slots[0] > 0) {
1637                 leaf = path.nodes[0];
1638                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0] - 1);
1639                 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
1640                     key.type == BTRFS_EXTENT_CSUM_KEY)
1641                         path.slots[0]--;
1642         }
1643
1644         while (len > 0) {
1645                 leaf = path.nodes[0];
1646                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1647                         ret = btrfs_next_leaf(root->fs_info->csum_root, &path);
1648                         if (ret > 0)
1649                                 break;
1650                         else if (ret < 0)
1651                                 goto out;
1652                         leaf = path.nodes[0];
1653                 }
1654
1655                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1656                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1657                     key.type != BTRFS_EXTENT_CSUM_KEY)
1658                         break;
1659
1660                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1661                 if (key.offset >= start + len)
1662                         break;
1663
1664                 if (key.offset > start)
1665                         start = key.offset;
1666
1667                 size = btrfs_item_size_nr(leaf, path.slots[0]);
1668                 csum_end = key.offset + (size / csum_size) * root->sectorsize;
1669                 if (csum_end > start) {
1670                         size = min(csum_end - start, len);
1671                         len -= size;
1672                         start += size;
1673                         *found += size;
1674                 }
1675
1676                 path.slots[0]++;
1677         }
1678 out:
1679         btrfs_release_path(&path);
1680         if (ret < 0)
1681                 return ret;
1682         return 0;
1683 }
1684
1685 static int process_file_extent(struct btrfs_root *root,
1686                                 struct extent_buffer *eb,
1687                                 int slot, struct btrfs_key *key,
1688                                 struct shared_node *active_node)
1689 {
1690         struct inode_record *rec;
1691         struct btrfs_file_extent_item *fi;
1692         u64 num_bytes = 0;
1693         u64 disk_bytenr = 0;
1694         u64 extent_offset = 0;
1695         u64 mask = root->sectorsize - 1;
1696         int extent_type;
1697         int ret;
1698
1699         rec = active_node->current;
1700         BUG_ON(rec->ino != key->objectid || rec->refs > 1);
1701         rec->found_file_extent = 1;
1702
1703         if (rec->extent_start == (u64)-1) {
1704                 rec->extent_start = key->offset;
1705                 rec->extent_end = key->offset;
1706         }
1707
1708         if (rec->extent_end > key->offset)
1709                 rec->errors |= I_ERR_FILE_EXTENT_OVERLAP;
1710         else if (rec->extent_end < key->offset) {
1711                 ret = add_file_extent_hole(&rec->holes, rec->extent_end,
1712                                            key->offset - rec->extent_end);
1713                 if (ret < 0)
1714                         return ret;
1715         }
1716
1717         fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
1718         extent_type = btrfs_file_extent_type(eb, fi);
1719
1720         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1721                 num_bytes = btrfs_file_extent_inline_len(eb, slot, fi);
1722                 if (num_bytes == 0)
1723                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1724                 rec->found_size += num_bytes;
1725                 num_bytes = (num_bytes + mask) & ~mask;
1726         } else if (extent_type == BTRFS_FILE_EXTENT_REG ||
1727                    extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1728                 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1729                 disk_bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1730                 extent_offset = btrfs_file_extent_offset(eb, fi);
1731                 if (num_bytes == 0 || (num_bytes & mask))
1732                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1733                 if (num_bytes + extent_offset >
1734                     btrfs_file_extent_ram_bytes(eb, fi))
1735                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1736                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC &&
1737                     (btrfs_file_extent_compression(eb, fi) ||
1738                      btrfs_file_extent_encryption(eb, fi) ||
1739                      btrfs_file_extent_other_encoding(eb, fi)))
1740                         rec->errors |= I_ERR_BAD_FILE_EXTENT;
1741                 if (disk_bytenr > 0)
1742                         rec->found_size += num_bytes;
1743         } else {
1744                 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1745         }
1746         rec->extent_end = key->offset + num_bytes;
1747
1748         /*
1749          * The data reloc tree will copy full extents into its inode and then
1750          * copy the corresponding csums.  Because the extent it copied could be
1751          * a preallocated extent that hasn't been written to yet there may be no
1752          * csums to copy, ergo we won't have csums for our file extent.  This is
1753          * ok so just don't bother checking csums if the inode belongs to the
1754          * data reloc tree.
1755          */
1756         if (disk_bytenr > 0 &&
1757             btrfs_header_owner(eb) != BTRFS_DATA_RELOC_TREE_OBJECTID) {
1758                 u64 found;
1759                 if (btrfs_file_extent_compression(eb, fi))
1760                         num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1761                 else
1762                         disk_bytenr += extent_offset;
1763
1764                 ret = count_csum_range(root, disk_bytenr, num_bytes, &found);
1765                 if (ret < 0)
1766                         return ret;
1767                 if (extent_type == BTRFS_FILE_EXTENT_REG) {
1768                         if (found > 0)
1769                                 rec->found_csum_item = 1;
1770                         if (found < num_bytes)
1771                                 rec->some_csum_missing = 1;
1772                 } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1773                         if (found > 0)
1774                                 rec->errors |= I_ERR_ODD_CSUM_ITEM;
1775                 }
1776         }
1777         return 0;
1778 }
1779
1780 static int process_one_leaf(struct btrfs_root *root, struct extent_buffer *eb,
1781                             struct walk_control *wc)
1782 {
1783         struct btrfs_key key;
1784         u32 nritems;
1785         int i;
1786         int ret = 0;
1787         struct cache_tree *inode_cache;
1788         struct shared_node *active_node;
1789
1790         if (wc->root_level == wc->active_node &&
1791             btrfs_root_refs(&root->root_item) == 0)
1792                 return 0;
1793
1794         active_node = wc->nodes[wc->active_node];
1795         inode_cache = &active_node->inode_cache;
1796         nritems = btrfs_header_nritems(eb);
1797         for (i = 0; i < nritems; i++) {
1798                 btrfs_item_key_to_cpu(eb, &key, i);
1799
1800                 if (key.objectid == BTRFS_FREE_SPACE_OBJECTID)
1801                         continue;
1802                 if (key.type == BTRFS_ORPHAN_ITEM_KEY)
1803                         continue;
1804
1805                 if (active_node->current == NULL ||
1806                     active_node->current->ino < key.objectid) {
1807                         if (active_node->current) {
1808                                 active_node->current->checked = 1;
1809                                 maybe_free_inode_rec(inode_cache,
1810                                                      active_node->current);
1811                         }
1812                         active_node->current = get_inode_rec(inode_cache,
1813                                                              key.objectid, 1);
1814                         BUG_ON(IS_ERR(active_node->current));
1815                 }
1816                 switch (key.type) {
1817                 case BTRFS_DIR_ITEM_KEY:
1818                 case BTRFS_DIR_INDEX_KEY:
1819                         ret = process_dir_item(root, eb, i, &key, active_node);
1820                         break;
1821                 case BTRFS_INODE_REF_KEY:
1822                         ret = process_inode_ref(eb, i, &key, active_node);
1823                         break;
1824                 case BTRFS_INODE_EXTREF_KEY:
1825                         ret = process_inode_extref(eb, i, &key, active_node);
1826                         break;
1827                 case BTRFS_INODE_ITEM_KEY:
1828                         ret = process_inode_item(eb, i, &key, active_node);
1829                         break;
1830                 case BTRFS_EXTENT_DATA_KEY:
1831                         ret = process_file_extent(root, eb, i, &key,
1832                                                   active_node);
1833                         break;
1834                 default:
1835                         break;
1836                 };
1837         }
1838         return ret;
1839 }
1840
1841 static void reada_walk_down(struct btrfs_root *root,
1842                             struct extent_buffer *node, int slot)
1843 {
1844         u64 bytenr;
1845         u64 ptr_gen;
1846         u32 nritems;
1847         u32 blocksize;
1848         int i;
1849         int level;
1850
1851         level = btrfs_header_level(node);
1852         if (level != 1)
1853                 return;
1854
1855         nritems = btrfs_header_nritems(node);
1856         blocksize = root->nodesize;
1857         for (i = slot; i < nritems; i++) {
1858                 bytenr = btrfs_node_blockptr(node, i);
1859                 ptr_gen = btrfs_node_ptr_generation(node, i);
1860                 readahead_tree_block(root, bytenr, blocksize, ptr_gen);
1861         }
1862 }
1863
1864 /*
1865  * Check the child node/leaf by the following condition:
1866  * 1. the first item key of the node/leaf should be the same with the one
1867  *    in parent.
1868  * 2. block in parent node should match the child node/leaf.
1869  * 3. generation of parent node and child's header should be consistent.
1870  *
1871  * Or the child node/leaf pointed by the key in parent is not valid.
1872  *
1873  * We hope to check leaf owner too, but since subvol may share leaves,
1874  * which makes leaf owner check not so strong, key check should be
1875  * sufficient enough for that case.
1876  */
1877 static int check_child_node(struct btrfs_root *root,
1878                             struct extent_buffer *parent, int slot,
1879                             struct extent_buffer *child)
1880 {
1881         struct btrfs_key parent_key;
1882         struct btrfs_key child_key;
1883         int ret = 0;
1884
1885         btrfs_node_key_to_cpu(parent, &parent_key, slot);
1886         if (btrfs_header_level(child) == 0)
1887                 btrfs_item_key_to_cpu(child, &child_key, 0);
1888         else
1889                 btrfs_node_key_to_cpu(child, &child_key, 0);
1890
1891         if (memcmp(&parent_key, &child_key, sizeof(parent_key))) {
1892                 ret = -EINVAL;
1893                 fprintf(stderr,
1894                         "Wrong key of child node/leaf, wanted: (%llu, %u, %llu), have: (%llu, %u, %llu)\n",
1895                         parent_key.objectid, parent_key.type, parent_key.offset,
1896                         child_key.objectid, child_key.type, child_key.offset);
1897         }
1898         if (btrfs_header_bytenr(child) != btrfs_node_blockptr(parent, slot)) {
1899                 ret = -EINVAL;
1900                 fprintf(stderr, "Wrong block of child node/leaf, wanted: %llu, have: %llu\n",
1901                         btrfs_node_blockptr(parent, slot),
1902                         btrfs_header_bytenr(child));
1903         }
1904         if (btrfs_node_ptr_generation(parent, slot) !=
1905             btrfs_header_generation(child)) {
1906                 ret = -EINVAL;
1907                 fprintf(stderr, "Wrong generation of child node/leaf, wanted: %llu, have: %llu\n",
1908                         btrfs_header_generation(child),
1909                         btrfs_node_ptr_generation(parent, slot));
1910         }
1911         return ret;
1912 }
1913
1914 struct node_refs {
1915         u64 bytenr[BTRFS_MAX_LEVEL];
1916         u64 refs[BTRFS_MAX_LEVEL];
1917 };
1918
1919 static int walk_down_tree(struct btrfs_root *root, struct btrfs_path *path,
1920                           struct walk_control *wc, int *level,
1921                           struct node_refs *nrefs)
1922 {
1923         enum btrfs_tree_block_status status;
1924         u64 bytenr;
1925         u64 ptr_gen;
1926         struct extent_buffer *next;
1927         struct extent_buffer *cur;
1928         u32 blocksize;
1929         int ret, err = 0;
1930         u64 refs;
1931
1932         WARN_ON(*level < 0);
1933         WARN_ON(*level >= BTRFS_MAX_LEVEL);
1934
1935         if (path->nodes[*level]->start == nrefs->bytenr[*level]) {
1936                 refs = nrefs->refs[*level];
1937                 ret = 0;
1938         } else {
1939                 ret = btrfs_lookup_extent_info(NULL, root,
1940                                        path->nodes[*level]->start,
1941                                        *level, 1, &refs, NULL);
1942                 if (ret < 0) {
1943                         err = ret;
1944                         goto out;
1945                 }
1946                 nrefs->bytenr[*level] = path->nodes[*level]->start;
1947                 nrefs->refs[*level] = refs;
1948         }
1949
1950         if (refs > 1) {
1951                 ret = enter_shared_node(root, path->nodes[*level]->start,
1952                                         refs, wc, *level);
1953                 if (ret > 0) {
1954                         err = ret;
1955                         goto out;
1956                 }
1957         }
1958
1959         while (*level >= 0) {
1960                 WARN_ON(*level < 0);
1961                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1962                 cur = path->nodes[*level];
1963
1964                 if (btrfs_header_level(cur) != *level)
1965                         WARN_ON(1);
1966
1967                 if (path->slots[*level] >= btrfs_header_nritems(cur))
1968                         break;
1969                 if (*level == 0) {
1970                         ret = process_one_leaf(root, cur, wc);
1971                         if (ret < 0)
1972                                 err = ret;
1973                         break;
1974                 }
1975                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1976                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1977                 blocksize = root->nodesize;
1978
1979                 if (bytenr == nrefs->bytenr[*level - 1]) {
1980                         refs = nrefs->refs[*level - 1];
1981                 } else {
1982                         ret = btrfs_lookup_extent_info(NULL, root, bytenr,
1983                                         *level - 1, 1, &refs, NULL);
1984                         if (ret < 0) {
1985                                 refs = 0;
1986                         } else {
1987                                 nrefs->bytenr[*level - 1] = bytenr;
1988                                 nrefs->refs[*level - 1] = refs;
1989                         }
1990                 }
1991
1992                 if (refs > 1) {
1993                         ret = enter_shared_node(root, bytenr, refs,
1994                                                 wc, *level - 1);
1995                         if (ret > 0) {
1996                                 path->slots[*level]++;
1997                                 continue;
1998                         }
1999                 }
2000
2001                 next = btrfs_find_tree_block(root, bytenr, blocksize);
2002                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2003                         free_extent_buffer(next);
2004                         reada_walk_down(root, cur, path->slots[*level]);
2005                         next = read_tree_block(root, bytenr, blocksize,
2006                                                ptr_gen);
2007                         if (!extent_buffer_uptodate(next)) {
2008                                 struct btrfs_key node_key;
2009
2010                                 btrfs_node_key_to_cpu(path->nodes[*level],
2011                                                       &node_key,
2012                                                       path->slots[*level]);
2013                                 btrfs_add_corrupt_extent_record(root->fs_info,
2014                                                 &node_key,
2015                                                 path->nodes[*level]->start,
2016                                                 root->nodesize, *level);
2017                                 err = -EIO;
2018                                 goto out;
2019                         }
2020                 }
2021
2022                 ret = check_child_node(root, cur, path->slots[*level], next);
2023                 if (ret) {
2024                         err = ret;
2025                         goto out;
2026                 }
2027
2028                 if (btrfs_is_leaf(next))
2029                         status = btrfs_check_leaf(root, NULL, next);
2030                 else
2031                         status = btrfs_check_node(root, NULL, next);
2032                 if (status != BTRFS_TREE_BLOCK_CLEAN) {
2033                         free_extent_buffer(next);
2034                         err = -EIO;
2035                         goto out;
2036                 }
2037
2038                 *level = *level - 1;
2039                 free_extent_buffer(path->nodes[*level]);
2040                 path->nodes[*level] = next;
2041                 path->slots[*level] = 0;
2042         }
2043 out:
2044         path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2045         return err;
2046 }
2047
2048 static int walk_up_tree(struct btrfs_root *root, struct btrfs_path *path,
2049                         struct walk_control *wc, int *level)
2050 {
2051         int i;
2052         struct extent_buffer *leaf;
2053
2054         for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2055                 leaf = path->nodes[i];
2056                 if (path->slots[i] + 1 < btrfs_header_nritems(leaf)) {
2057                         path->slots[i]++;
2058                         *level = i;
2059                         return 0;
2060                 } else {
2061                         free_extent_buffer(path->nodes[*level]);
2062                         path->nodes[*level] = NULL;
2063                         BUG_ON(*level > wc->active_node);
2064                         if (*level == wc->active_node)
2065                                 leave_shared_node(root, wc, *level);
2066                         *level = i + 1;
2067                 }
2068         }
2069         return 1;
2070 }
2071
2072 static int check_root_dir(struct inode_record *rec)
2073 {
2074         struct inode_backref *backref;
2075         int ret = -1;
2076
2077         if (!rec->found_inode_item || rec->errors)
2078                 goto out;
2079         if (rec->nlink != 1 || rec->found_link != 0)
2080                 goto out;
2081         if (list_empty(&rec->backrefs))
2082                 goto out;
2083         backref = to_inode_backref(rec->backrefs.next);
2084         if (!backref->found_inode_ref)
2085                 goto out;
2086         if (backref->index != 0 || backref->namelen != 2 ||
2087             memcmp(backref->name, "..", 2))
2088                 goto out;
2089         if (backref->found_dir_index || backref->found_dir_item)
2090                 goto out;
2091         ret = 0;
2092 out:
2093         return ret;
2094 }
2095
2096 static int repair_inode_isize(struct btrfs_trans_handle *trans,
2097                               struct btrfs_root *root, struct btrfs_path *path,
2098                               struct inode_record *rec)
2099 {
2100         struct btrfs_inode_item *ei;
2101         struct btrfs_key key;
2102         int ret;
2103
2104         key.objectid = rec->ino;
2105         key.type = BTRFS_INODE_ITEM_KEY;
2106         key.offset = (u64)-1;
2107
2108         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2109         if (ret < 0)
2110                 goto out;
2111         if (ret) {
2112                 if (!path->slots[0]) {
2113                         ret = -ENOENT;
2114                         goto out;
2115                 }
2116                 path->slots[0]--;
2117                 ret = 0;
2118         }
2119         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2120         if (key.objectid != rec->ino) {
2121                 ret = -ENOENT;
2122                 goto out;
2123         }
2124
2125         ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
2126                             struct btrfs_inode_item);
2127         btrfs_set_inode_size(path->nodes[0], ei, rec->found_size);
2128         btrfs_mark_buffer_dirty(path->nodes[0]);
2129         rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
2130         printf("reset isize for dir %Lu root %Lu\n", rec->ino,
2131                root->root_key.objectid);
2132 out:
2133         btrfs_release_path(path);
2134         return ret;
2135 }
2136
2137 static int repair_inode_orphan_item(struct btrfs_trans_handle *trans,
2138                                     struct btrfs_root *root,
2139                                     struct btrfs_path *path,
2140                                     struct inode_record *rec)
2141 {
2142         int ret;
2143
2144         ret = btrfs_add_orphan_item(trans, root, path, rec->ino);
2145         btrfs_release_path(path);
2146         if (!ret)
2147                 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
2148         return ret;
2149 }
2150
2151 static int repair_inode_nbytes(struct btrfs_trans_handle *trans,
2152                                struct btrfs_root *root,
2153                                struct btrfs_path *path,
2154                                struct inode_record *rec)
2155 {
2156         struct btrfs_inode_item *ei;
2157         struct btrfs_key key;
2158         int ret = 0;
2159
2160         key.objectid = rec->ino;
2161         key.type = BTRFS_INODE_ITEM_KEY;
2162         key.offset = 0;
2163
2164         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2165         if (ret) {
2166                 if (ret > 0)
2167                         ret = -ENOENT;
2168                 goto out;
2169         }
2170
2171         /* Since ret == 0, no need to check anything */
2172         ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
2173                             struct btrfs_inode_item);
2174         btrfs_set_inode_nbytes(path->nodes[0], ei, rec->found_size);
2175         btrfs_mark_buffer_dirty(path->nodes[0]);
2176         rec->errors &= ~I_ERR_FILE_NBYTES_WRONG;
2177         printf("reset nbytes for ino %llu root %llu\n",
2178                rec->ino, root->root_key.objectid);
2179 out:
2180         btrfs_release_path(path);
2181         return ret;
2182 }
2183
2184 static int add_missing_dir_index(struct btrfs_root *root,
2185                                  struct cache_tree *inode_cache,
2186                                  struct inode_record *rec,
2187                                  struct inode_backref *backref)
2188 {
2189         struct btrfs_path path;
2190         struct btrfs_trans_handle *trans;
2191         struct btrfs_dir_item *dir_item;
2192         struct extent_buffer *leaf;
2193         struct btrfs_key key;
2194         struct btrfs_disk_key disk_key;
2195         struct inode_record *dir_rec;
2196         unsigned long name_ptr;
2197         u32 data_size = sizeof(*dir_item) + backref->namelen;
2198         int ret;
2199
2200         trans = btrfs_start_transaction(root, 1);
2201         if (IS_ERR(trans))
2202                 return PTR_ERR(trans);
2203
2204         fprintf(stderr, "repairing missing dir index item for inode %llu\n",
2205                 (unsigned long long)rec->ino);
2206
2207         btrfs_init_path(&path);
2208         key.objectid = backref->dir;
2209         key.type = BTRFS_DIR_INDEX_KEY;
2210         key.offset = backref->index;
2211         ret = btrfs_insert_empty_item(trans, root, &path, &key, data_size);
2212         BUG_ON(ret);
2213
2214         leaf = path.nodes[0];
2215         dir_item = btrfs_item_ptr(leaf, path.slots[0], struct btrfs_dir_item);
2216
2217         disk_key.objectid = cpu_to_le64(rec->ino);
2218         disk_key.type = BTRFS_INODE_ITEM_KEY;
2219         disk_key.offset = 0;
2220
2221         btrfs_set_dir_item_key(leaf, dir_item, &disk_key);
2222         btrfs_set_dir_type(leaf, dir_item, imode_to_type(rec->imode));
2223         btrfs_set_dir_data_len(leaf, dir_item, 0);
2224         btrfs_set_dir_name_len(leaf, dir_item, backref->namelen);
2225         name_ptr = (unsigned long)(dir_item + 1);
2226         write_extent_buffer(leaf, backref->name, name_ptr, backref->namelen);
2227         btrfs_mark_buffer_dirty(leaf);
2228         btrfs_release_path(&path);
2229         btrfs_commit_transaction(trans, root);
2230
2231         backref->found_dir_index = 1;
2232         dir_rec = get_inode_rec(inode_cache, backref->dir, 0);
2233         BUG_ON(IS_ERR(dir_rec));
2234         if (!dir_rec)
2235                 return 0;
2236         dir_rec->found_size += backref->namelen;
2237         if (dir_rec->found_size == dir_rec->isize &&
2238             (dir_rec->errors & I_ERR_DIR_ISIZE_WRONG))
2239                 dir_rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
2240         if (dir_rec->found_size != dir_rec->isize)
2241                 dir_rec->errors |= I_ERR_DIR_ISIZE_WRONG;
2242
2243         return 0;
2244 }
2245
2246 static int delete_dir_index(struct btrfs_root *root,
2247                             struct cache_tree *inode_cache,
2248                             struct inode_record *rec,
2249                             struct inode_backref *backref)
2250 {
2251         struct btrfs_trans_handle *trans;
2252         struct btrfs_dir_item *di;
2253         struct btrfs_path path;
2254         int ret = 0;
2255
2256         trans = btrfs_start_transaction(root, 1);
2257         if (IS_ERR(trans))
2258                 return PTR_ERR(trans);
2259
2260         fprintf(stderr, "Deleting bad dir index [%llu,%u,%llu] root %llu\n",
2261                 (unsigned long long)backref->dir,
2262                 BTRFS_DIR_INDEX_KEY, (unsigned long long)backref->index,
2263                 (unsigned long long)root->objectid);
2264
2265         btrfs_init_path(&path);
2266         di = btrfs_lookup_dir_index(trans, root, &path, backref->dir,
2267                                     backref->name, backref->namelen,
2268                                     backref->index, -1);
2269         if (IS_ERR(di)) {
2270                 ret = PTR_ERR(di);
2271                 btrfs_release_path(&path);
2272                 btrfs_commit_transaction(trans, root);
2273                 if (ret == -ENOENT)
2274                         return 0;
2275                 return ret;
2276         }
2277
2278         if (!di)
2279                 ret = btrfs_del_item(trans, root, &path);
2280         else
2281                 ret = btrfs_delete_one_dir_name(trans, root, &path, di);
2282         BUG_ON(ret);
2283         btrfs_release_path(&path);
2284         btrfs_commit_transaction(trans, root);
2285         return ret;
2286 }
2287
2288 static int create_inode_item(struct btrfs_root *root,
2289                              struct inode_record *rec,
2290                              struct inode_backref *backref, int root_dir)
2291 {
2292         struct btrfs_trans_handle *trans;
2293         struct btrfs_inode_item inode_item;
2294         time_t now = time(NULL);
2295         int ret;
2296
2297         trans = btrfs_start_transaction(root, 1);
2298         if (IS_ERR(trans)) {
2299                 ret = PTR_ERR(trans);
2300                 return ret;
2301         }
2302
2303         fprintf(stderr, "root %llu inode %llu recreating inode item, this may "
2304                 "be incomplete, please check permissions and content after "
2305                 "the fsck completes.\n", (unsigned long long)root->objectid,
2306                 (unsigned long long)rec->ino);
2307
2308         memset(&inode_item, 0, sizeof(inode_item));
2309         btrfs_set_stack_inode_generation(&inode_item, trans->transid);
2310         if (root_dir)
2311                 btrfs_set_stack_inode_nlink(&inode_item, 1);
2312         else
2313                 btrfs_set_stack_inode_nlink(&inode_item, rec->found_link);
2314         btrfs_set_stack_inode_nbytes(&inode_item, rec->found_size);
2315         if (rec->found_dir_item) {
2316                 if (rec->found_file_extent)
2317                         fprintf(stderr, "root %llu inode %llu has both a dir "
2318                                 "item and extents, unsure if it is a dir or a "
2319                                 "regular file so setting it as a directory\n",
2320                                 (unsigned long long)root->objectid,
2321                                 (unsigned long long)rec->ino);
2322                 btrfs_set_stack_inode_mode(&inode_item, S_IFDIR | 0755);
2323                 btrfs_set_stack_inode_size(&inode_item, rec->found_size);
2324         } else if (!rec->found_dir_item) {
2325                 btrfs_set_stack_inode_size(&inode_item, rec->extent_end);
2326                 btrfs_set_stack_inode_mode(&inode_item, S_IFREG | 0755);
2327         }
2328         btrfs_set_stack_timespec_sec(&inode_item.atime, now);
2329         btrfs_set_stack_timespec_nsec(&inode_item.atime, 0);
2330         btrfs_set_stack_timespec_sec(&inode_item.ctime, now);
2331         btrfs_set_stack_timespec_nsec(&inode_item.ctime, 0);
2332         btrfs_set_stack_timespec_sec(&inode_item.mtime, now);
2333         btrfs_set_stack_timespec_nsec(&inode_item.mtime, 0);
2334         btrfs_set_stack_timespec_sec(&inode_item.otime, 0);
2335         btrfs_set_stack_timespec_nsec(&inode_item.otime, 0);
2336
2337         ret = btrfs_insert_inode(trans, root, rec->ino, &inode_item);
2338         BUG_ON(ret);
2339         btrfs_commit_transaction(trans, root);
2340         return 0;
2341 }
2342
2343 static int repair_inode_backrefs(struct btrfs_root *root,
2344                                  struct inode_record *rec,
2345                                  struct cache_tree *inode_cache,
2346                                  int delete)
2347 {
2348         struct inode_backref *tmp, *backref;
2349         u64 root_dirid = btrfs_root_dirid(&root->root_item);
2350         int ret = 0;
2351         int repaired = 0;
2352
2353         list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
2354                 if (!delete && rec->ino == root_dirid) {
2355                         if (!rec->found_inode_item) {
2356                                 ret = create_inode_item(root, rec, backref, 1);
2357                                 if (ret)
2358                                         break;
2359                                 repaired++;
2360                         }
2361                 }
2362
2363                 /* Index 0 for root dir's are special, don't mess with it */
2364                 if (rec->ino == root_dirid && backref->index == 0)
2365                         continue;
2366
2367                 if (delete &&
2368                     ((backref->found_dir_index && !backref->found_inode_ref) ||
2369                      (backref->found_dir_index && backref->found_inode_ref &&
2370                       (backref->errors & REF_ERR_INDEX_UNMATCH)))) {
2371                         ret = delete_dir_index(root, inode_cache, rec, backref);
2372                         if (ret)
2373                                 break;
2374                         repaired++;
2375                         list_del(&backref->list);
2376                         free(backref);
2377                 }
2378
2379                 if (!delete && !backref->found_dir_index &&
2380                     backref->found_dir_item && backref->found_inode_ref) {
2381                         ret = add_missing_dir_index(root, inode_cache, rec,
2382                                                     backref);
2383                         if (ret)
2384                                 break;
2385                         repaired++;
2386                         if (backref->found_dir_item &&
2387                             backref->found_dir_index &&
2388                             backref->found_dir_index) {
2389                                 if (!backref->errors &&
2390                                     backref->found_inode_ref) {
2391                                         list_del(&backref->list);
2392                                         free(backref);
2393                                 }
2394                         }
2395                 }
2396
2397                 if (!delete && (!backref->found_dir_index &&
2398                                 !backref->found_dir_item &&
2399                                 backref->found_inode_ref)) {
2400                         struct btrfs_trans_handle *trans;
2401                         struct btrfs_key location;
2402
2403                         ret = check_dir_conflict(root, backref->name,
2404                                                  backref->namelen,
2405                                                  backref->dir,
2406                                                  backref->index);
2407                         if (ret) {
2408                                 /*
2409                                  * let nlink fixing routine to handle it,
2410                                  * which can do it better.
2411                                  */
2412                                 ret = 0;
2413                                 break;
2414                         }
2415                         location.objectid = rec->ino;
2416                         location.type = BTRFS_INODE_ITEM_KEY;
2417                         location.offset = 0;
2418
2419                         trans = btrfs_start_transaction(root, 1);
2420                         if (IS_ERR(trans)) {
2421                                 ret = PTR_ERR(trans);
2422                                 break;
2423                         }
2424                         fprintf(stderr, "adding missing dir index/item pair "
2425                                 "for inode %llu\n",
2426                                 (unsigned long long)rec->ino);
2427                         ret = btrfs_insert_dir_item(trans, root, backref->name,
2428                                                     backref->namelen,
2429                                                     backref->dir, &location,
2430                                                     imode_to_type(rec->imode),
2431                                                     backref->index);
2432                         BUG_ON(ret);
2433                         btrfs_commit_transaction(trans, root);
2434                         repaired++;
2435                 }
2436
2437                 if (!delete && (backref->found_inode_ref &&
2438                                 backref->found_dir_index &&
2439                                 backref->found_dir_item &&
2440                                 !(backref->errors & REF_ERR_INDEX_UNMATCH) &&
2441                                 !rec->found_inode_item)) {
2442                         ret = create_inode_item(root, rec, backref, 0);
2443                         if (ret)
2444                                 break;
2445                         repaired++;
2446                 }
2447
2448         }
2449         return ret ? ret : repaired;
2450 }
2451
2452 /*
2453  * To determine the file type for nlink/inode_item repair
2454  *
2455  * Return 0 if file type is found and BTRFS_FT_* is stored into type.
2456  * Return -ENOENT if file type is not found.
2457  */
2458 static int find_file_type(struct inode_record *rec, u8 *type)
2459 {
2460         struct inode_backref *backref;
2461
2462         /* For inode item recovered case */
2463         if (rec->found_inode_item) {
2464                 *type = imode_to_type(rec->imode);
2465                 return 0;
2466         }
2467
2468         list_for_each_entry(backref, &rec->backrefs, list) {
2469                 if (backref->found_dir_index || backref->found_dir_item) {
2470                         *type = backref->filetype;
2471                         return 0;
2472                 }
2473         }
2474         return -ENOENT;
2475 }
2476
2477 /*
2478  * To determine the file name for nlink repair
2479  *
2480  * Return 0 if file name is found, set name and namelen.
2481  * Return -ENOENT if file name is not found.
2482  */
2483 static int find_file_name(struct inode_record *rec,
2484                           char *name, int *namelen)
2485 {
2486         struct inode_backref *backref;
2487
2488         list_for_each_entry(backref, &rec->backrefs, list) {
2489                 if (backref->found_dir_index || backref->found_dir_item ||
2490                     backref->found_inode_ref) {
2491                         memcpy(name, backref->name, backref->namelen);
2492                         *namelen = backref->namelen;
2493                         return 0;
2494                 }
2495         }
2496         return -ENOENT;
2497 }
2498
2499 /* Reset the nlink of the inode to the correct one */
2500 static int reset_nlink(struct btrfs_trans_handle *trans,
2501                        struct btrfs_root *root,
2502                        struct btrfs_path *path,
2503                        struct inode_record *rec)
2504 {
2505         struct inode_backref *backref;
2506         struct inode_backref *tmp;
2507         struct btrfs_key key;
2508         struct btrfs_inode_item *inode_item;
2509         int ret = 0;
2510
2511         /* We don't believe this either, reset it and iterate backref */
2512         rec->found_link = 0;
2513
2514         /* Remove all backref including the valid ones */
2515         list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
2516                 ret = btrfs_unlink(trans, root, rec->ino, backref->dir,
2517                                    backref->index, backref->name,
2518                                    backref->namelen, 0);
2519                 if (ret < 0)
2520                         goto out;
2521
2522                 /* remove invalid backref, so it won't be added back */
2523                 if (!(backref->found_dir_index &&
2524                       backref->found_dir_item &&
2525                       backref->found_inode_ref)) {
2526                         list_del(&backref->list);
2527                         free(backref);
2528                 } else {
2529                         rec->found_link++;
2530                 }
2531         }
2532
2533         /* Set nlink to 0 */
2534         key.objectid = rec->ino;
2535         key.type = BTRFS_INODE_ITEM_KEY;
2536         key.offset = 0;
2537         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2538         if (ret < 0)
2539                 goto out;
2540         if (ret > 0) {
2541                 ret = -ENOENT;
2542                 goto out;
2543         }
2544         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2545                                     struct btrfs_inode_item);
2546         btrfs_set_inode_nlink(path->nodes[0], inode_item, 0);
2547         btrfs_mark_buffer_dirty(path->nodes[0]);
2548         btrfs_release_path(path);
2549
2550         /*
2551          * Add back valid inode_ref/dir_item/dir_index,
2552          * add_link() will handle the nlink inc, so new nlink must be correct
2553          */
2554         list_for_each_entry(backref, &rec->backrefs, list) {
2555                 ret = btrfs_add_link(trans, root, rec->ino, backref->dir,
2556                                      backref->name, backref->namelen,
2557                                      backref->filetype, &backref->index, 1);
2558                 if (ret < 0)
2559                         goto out;
2560         }
2561 out:
2562         btrfs_release_path(path);
2563         return ret;
2564 }
2565
2566 static int repair_inode_nlinks(struct btrfs_trans_handle *trans,
2567                                struct btrfs_root *root,
2568                                struct btrfs_path *path,
2569                                struct inode_record *rec)
2570 {
2571         char *dir_name = "lost+found";
2572         char namebuf[BTRFS_NAME_LEN] = {0};
2573         u64 lost_found_ino;
2574         u32 mode = 0700;
2575         u8 type = 0;
2576         int namelen = 0;
2577         int name_recovered = 0;
2578         int type_recovered = 0;
2579         int ret = 0;
2580
2581         /*
2582          * Get file name and type first before these invalid inode ref
2583          * are deleted by remove_all_invalid_backref()
2584          */
2585         name_recovered = !find_file_name(rec, namebuf, &namelen);
2586         type_recovered = !find_file_type(rec, &type);
2587
2588         if (!name_recovered) {
2589                 printf("Can't get file name for inode %llu, using '%llu' as fallback\n",
2590                        rec->ino, rec->ino);
2591                 namelen = count_digits(rec->ino);
2592                 sprintf(namebuf, "%llu", rec->ino);
2593                 name_recovered = 1;
2594         }
2595         if (!type_recovered) {
2596                 printf("Can't get file type for inode %llu, using FILE as fallback\n",
2597                        rec->ino);
2598                 type = BTRFS_FT_REG_FILE;
2599                 type_recovered = 1;
2600         }
2601
2602         ret = reset_nlink(trans, root, path, rec);
2603         if (ret < 0) {
2604                 fprintf(stderr,
2605                         "Failed to reset nlink for inode %llu: %s\n",
2606                         rec->ino, strerror(-ret));
2607                 goto out;
2608         }
2609
2610         if (rec->found_link == 0) {
2611                 lost_found_ino = root->highest_inode;
2612                 if (lost_found_ino >= BTRFS_LAST_FREE_OBJECTID) {
2613                         ret = -EOVERFLOW;
2614                         goto out;
2615                 }
2616                 lost_found_ino++;
2617                 ret = btrfs_mkdir(trans, root, dir_name, strlen(dir_name),
2618                                   BTRFS_FIRST_FREE_OBJECTID, &lost_found_ino,
2619                                   mode);
2620                 if (ret < 0) {
2621                         fprintf(stderr, "Failed to create '%s' dir: %s\n",
2622                                 dir_name, strerror(-ret));
2623                         goto out;
2624                 }
2625                 ret = btrfs_add_link(trans, root, rec->ino, lost_found_ino,
2626                                      namebuf, namelen, type, NULL, 1);
2627                 /*
2628                  * Add ".INO" suffix several times to handle case where
2629                  * "FILENAME.INO" is already taken by another file.
2630                  */
2631                 while (ret == -EEXIST) {
2632                         /*
2633                          * Conflicting file name, add ".INO" as suffix * +1 for '.'
2634                          */
2635                         if (namelen + count_digits(rec->ino) + 1 >
2636                             BTRFS_NAME_LEN) {
2637                                 ret = -EFBIG;
2638                                 goto out;
2639                         }
2640                         snprintf(namebuf + namelen, BTRFS_NAME_LEN - namelen,
2641                                  ".%llu", rec->ino);
2642                         namelen += count_digits(rec->ino) + 1;
2643                         ret = btrfs_add_link(trans, root, rec->ino,
2644                                              lost_found_ino, namebuf,
2645                                              namelen, type, NULL, 1);
2646                 }
2647                 if (ret < 0) {
2648                         fprintf(stderr,
2649                                 "Failed to link the inode %llu to %s dir: %s\n",
2650                                 rec->ino, dir_name, strerror(-ret));
2651                         goto out;
2652                 }
2653                 /*
2654                  * Just increase the found_link, don't actually add the
2655                  * backref. This will make things easier and this inode
2656                  * record will be freed after the repair is done.
2657                  * So fsck will not report problem about this inode.
2658                  */
2659                 rec->found_link++;
2660                 printf("Moving file '%.*s' to '%s' dir since it has no valid backref\n",
2661                        namelen, namebuf, dir_name);
2662         }
2663         printf("Fixed the nlink of inode %llu\n", rec->ino);
2664 out:
2665         /*
2666          * Clear the flag anyway, or we will loop forever for the same inode
2667          * as it will not be removed from the bad inode list and the dead loop
2668          * happens.
2669          */
2670         rec->errors &= ~I_ERR_LINK_COUNT_WRONG;
2671         btrfs_release_path(path);
2672         return ret;
2673 }
2674
2675 /*
2676  * Check if there is any normal(reg or prealloc) file extent for given
2677  * ino.
2678  * This is used to determine the file type when neither its dir_index/item or
2679  * inode_item exists.
2680  *
2681  * This will *NOT* report error, if any error happens, just consider it does
2682  * not have any normal file extent.
2683  */
2684 static int find_normal_file_extent(struct btrfs_root *root, u64 ino)
2685 {
2686         struct btrfs_path path;
2687         struct btrfs_key key;
2688         struct btrfs_key found_key;
2689         struct btrfs_file_extent_item *fi;
2690         u8 type;
2691         int ret = 0;
2692
2693         btrfs_init_path(&path);
2694         key.objectid = ino;
2695         key.type = BTRFS_EXTENT_DATA_KEY;
2696         key.offset = 0;
2697
2698         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
2699         if (ret < 0) {
2700                 ret = 0;
2701                 goto out;
2702         }
2703         if (ret && path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
2704                 ret = btrfs_next_leaf(root, &path);
2705                 if (ret) {
2706                         ret = 0;
2707                         goto out;
2708                 }
2709         }
2710         while (1) {
2711                 btrfs_item_key_to_cpu(path.nodes[0], &found_key,
2712                                       path.slots[0]);
2713                 if (found_key.objectid != ino ||
2714                     found_key.type != BTRFS_EXTENT_DATA_KEY)
2715                         break;
2716                 fi = btrfs_item_ptr(path.nodes[0], path.slots[0],
2717                                     struct btrfs_file_extent_item);
2718                 type = btrfs_file_extent_type(path.nodes[0], fi);
2719                 if (type != BTRFS_FILE_EXTENT_INLINE) {
2720                         ret = 1;
2721                         goto out;
2722                 }
2723         }
2724 out:
2725         btrfs_release_path(&path);
2726         return ret;
2727 }
2728
2729 static u32 btrfs_type_to_imode(u8 type)
2730 {
2731         static u32 imode_by_btrfs_type[] = {
2732                 [BTRFS_FT_REG_FILE]     = S_IFREG,
2733                 [BTRFS_FT_DIR]          = S_IFDIR,
2734                 [BTRFS_FT_CHRDEV]       = S_IFCHR,
2735                 [BTRFS_FT_BLKDEV]       = S_IFBLK,
2736                 [BTRFS_FT_FIFO]         = S_IFIFO,
2737                 [BTRFS_FT_SOCK]         = S_IFSOCK,
2738                 [BTRFS_FT_SYMLINK]      = S_IFLNK,
2739         };
2740
2741         return imode_by_btrfs_type[(type)];
2742 }
2743
2744 static int repair_inode_no_item(struct btrfs_trans_handle *trans,
2745                                 struct btrfs_root *root,
2746                                 struct btrfs_path *path,
2747                                 struct inode_record *rec)
2748 {
2749         u8 filetype;
2750         u32 mode = 0700;
2751         int type_recovered = 0;
2752         int ret = 0;
2753
2754         printf("Trying to rebuild inode:%llu\n", rec->ino);
2755
2756         type_recovered = !find_file_type(rec, &filetype);
2757
2758         /*
2759          * Try to determine inode type if type not found.
2760          *
2761          * For found regular file extent, it must be FILE.
2762          * For found dir_item/index, it must be DIR.
2763          *
2764          * For undetermined one, use FILE as fallback.
2765          *
2766          * TODO:
2767          * 1. If found backref(inode_index/item is already handled) to it,
2768          *    it must be DIR.
2769          *    Need new inode-inode ref structure to allow search for that.
2770          */
2771         if (!type_recovered) {
2772                 if (rec->found_file_extent &&
2773                     find_normal_file_extent(root, rec->ino)) {
2774                         type_recovered = 1;
2775                         filetype = BTRFS_FT_REG_FILE;
2776                 } else if (rec->found_dir_item) {
2777                         type_recovered = 1;
2778                         filetype = BTRFS_FT_DIR;
2779                 } else if (!list_empty(&rec->orphan_extents)) {
2780                         type_recovered = 1;
2781                         filetype = BTRFS_FT_REG_FILE;
2782                 } else{
2783                         printf("Can't determine the filetype for inode %llu, assume it is a normal file\n",
2784                                rec->ino);
2785                         type_recovered = 1;
2786                         filetype = BTRFS_FT_REG_FILE;
2787                 }
2788         }
2789
2790         ret = btrfs_new_inode(trans, root, rec->ino,
2791                               mode | btrfs_type_to_imode(filetype));
2792         if (ret < 0)
2793                 goto out;
2794
2795         /*
2796          * Here inode rebuild is done, we only rebuild the inode item,
2797          * don't repair the nlink(like move to lost+found).
2798          * That is the job of nlink repair.
2799          *
2800          * We just fill the record and return
2801          */
2802         rec->found_dir_item = 1;
2803         rec->imode = mode | btrfs_type_to_imode(filetype);
2804         rec->nlink = 0;
2805         rec->errors &= ~I_ERR_NO_INODE_ITEM;
2806         /* Ensure the inode_nlinks repair function will be called */
2807         rec->errors |= I_ERR_LINK_COUNT_WRONG;
2808 out:
2809         return ret;
2810 }
2811
2812 static int repair_inode_orphan_extent(struct btrfs_trans_handle *trans,
2813                                       struct btrfs_root *root,
2814                                       struct btrfs_path *path,
2815                                       struct inode_record *rec)
2816 {
2817         struct orphan_data_extent *orphan;
2818         struct orphan_data_extent *tmp;
2819         int ret = 0;
2820
2821         list_for_each_entry_safe(orphan, tmp, &rec->orphan_extents, list) {
2822                 /*
2823                  * Check for conflicting file extents
2824                  *
2825                  * Here we don't know whether the extents is compressed or not,
2826                  * so we can only assume it not compressed nor data offset,
2827                  * and use its disk_len as extent length.
2828                  */
2829                 ret = btrfs_get_extent(NULL, root, path, orphan->objectid,
2830                                        orphan->offset, orphan->disk_len, 0);
2831                 btrfs_release_path(path);
2832                 if (ret < 0)
2833                         goto out;
2834                 if (!ret) {
2835                         fprintf(stderr,
2836                                 "orphan extent (%llu, %llu) conflicts, delete the orphan\n",
2837                                 orphan->disk_bytenr, orphan->disk_len);
2838                         ret = btrfs_free_extent(trans,
2839                                         root->fs_info->extent_root,
2840                                         orphan->disk_bytenr, orphan->disk_len,
2841                                         0, root->objectid, orphan->objectid,
2842                                         orphan->offset);
2843                         if (ret < 0)
2844                                 goto out;
2845                 }
2846                 ret = btrfs_insert_file_extent(trans, root, orphan->objectid,
2847                                 orphan->offset, orphan->disk_bytenr,
2848                                 orphan->disk_len, orphan->disk_len);
2849                 if (ret < 0)
2850                         goto out;
2851
2852                 /* Update file size info */
2853                 rec->found_size += orphan->disk_len;
2854                 if (rec->found_size == rec->nbytes)
2855                         rec->errors &= ~I_ERR_FILE_NBYTES_WRONG;
2856
2857                 /* Update the file extent hole info too */
2858                 ret = del_file_extent_hole(&rec->holes, orphan->offset,
2859                                            orphan->disk_len);
2860                 if (ret < 0)
2861                         goto out;
2862                 if (RB_EMPTY_ROOT(&rec->holes))
2863                         rec->errors &= ~I_ERR_FILE_EXTENT_DISCOUNT;
2864
2865                 list_del(&orphan->list);
2866                 free(orphan);
2867         }
2868         rec->errors &= ~I_ERR_FILE_EXTENT_ORPHAN;
2869 out:
2870         return ret;
2871 }
2872
2873 static int repair_inode_discount_extent(struct btrfs_trans_handle *trans,
2874                                         struct btrfs_root *root,
2875                                         struct btrfs_path *path,
2876                                         struct inode_record *rec)
2877 {
2878         struct rb_node *node;
2879         struct file_extent_hole *hole;
2880         int found = 0;
2881         int ret = 0;
2882
2883         node = rb_first(&rec->holes);
2884
2885         while (node) {
2886                 found = 1;
2887                 hole = rb_entry(node, struct file_extent_hole, node);
2888                 ret = btrfs_punch_hole(trans, root, rec->ino,
2889                                        hole->start, hole->len);
2890                 if (ret < 0)
2891                         goto out;
2892                 ret = del_file_extent_hole(&rec->holes, hole->start,
2893                                            hole->len);
2894                 if (ret < 0)
2895                         goto out;
2896                 if (RB_EMPTY_ROOT(&rec->holes))
2897                         rec->errors &= ~I_ERR_FILE_EXTENT_DISCOUNT;
2898                 node = rb_first(&rec->holes);
2899         }
2900         /* special case for a file losing all its file extent */
2901         if (!found) {
2902                 ret = btrfs_punch_hole(trans, root, rec->ino, 0,
2903                                        round_up(rec->isize, root->sectorsize));
2904                 if (ret < 0)
2905                         goto out;
2906         }
2907         printf("Fixed discount file extents for inode: %llu in root: %llu\n",
2908                rec->ino, root->objectid);
2909 out:
2910         return ret;
2911 }
2912
2913 static int try_repair_inode(struct btrfs_root *root, struct inode_record *rec)
2914 {
2915         struct btrfs_trans_handle *trans;
2916         struct btrfs_path path;
2917         int ret = 0;
2918
2919         if (!(rec->errors & (I_ERR_DIR_ISIZE_WRONG |
2920                              I_ERR_NO_ORPHAN_ITEM |
2921                              I_ERR_LINK_COUNT_WRONG |
2922                              I_ERR_NO_INODE_ITEM |
2923                              I_ERR_FILE_EXTENT_ORPHAN |
2924                              I_ERR_FILE_EXTENT_DISCOUNT|
2925                              I_ERR_FILE_NBYTES_WRONG)))
2926                 return rec->errors;
2927
2928         /*
2929          * For nlink repair, it may create a dir and add link, so
2930          * 2 for parent(256)'s dir_index and dir_item
2931          * 2 for lost+found dir's inode_item and inode_ref
2932          * 1 for the new inode_ref of the file
2933          * 2 for lost+found dir's dir_index and dir_item for the file
2934          */
2935         trans = btrfs_start_transaction(root, 7);
2936         if (IS_ERR(trans))
2937                 return PTR_ERR(trans);
2938
2939         btrfs_init_path(&path);
2940         if (rec->errors & I_ERR_NO_INODE_ITEM)
2941                 ret = repair_inode_no_item(trans, root, &path, rec);
2942         if (!ret && rec->errors & I_ERR_FILE_EXTENT_ORPHAN)
2943                 ret = repair_inode_orphan_extent(trans, root, &path, rec);
2944         if (!ret && rec->errors & I_ERR_FILE_EXTENT_DISCOUNT)
2945                 ret = repair_inode_discount_extent(trans, root, &path, rec);
2946         if (!ret && rec->errors & I_ERR_DIR_ISIZE_WRONG)
2947                 ret = repair_inode_isize(trans, root, &path, rec);
2948         if (!ret && rec->errors & I_ERR_NO_ORPHAN_ITEM)
2949                 ret = repair_inode_orphan_item(trans, root, &path, rec);
2950         if (!ret && rec->errors & I_ERR_LINK_COUNT_WRONG)
2951                 ret = repair_inode_nlinks(trans, root, &path, rec);
2952         if (!ret && rec->errors & I_ERR_FILE_NBYTES_WRONG)
2953                 ret = repair_inode_nbytes(trans, root, &path, rec);
2954         btrfs_commit_transaction(trans, root);
2955         btrfs_release_path(&path);
2956         return ret;
2957 }
2958
2959 static int check_inode_recs(struct btrfs_root *root,
2960                             struct cache_tree *inode_cache)
2961 {
2962         struct cache_extent *cache;
2963         struct ptr_node *node;
2964         struct inode_record *rec;
2965         struct inode_backref *backref;
2966         int stage = 0;
2967         int ret = 0;
2968         int err = 0;
2969         u64 error = 0;
2970         u64 root_dirid = btrfs_root_dirid(&root->root_item);
2971
2972         if (btrfs_root_refs(&root->root_item) == 0) {
2973                 if (!cache_tree_empty(inode_cache))
2974                         fprintf(stderr, "warning line %d\n", __LINE__);
2975                 return 0;
2976         }
2977
2978         /*
2979          * We need to record the highest inode number for later 'lost+found'
2980          * dir creation.
2981          * We must select an ino not used/referred by any existing inode, or
2982          * 'lost+found' ino may be a missing ino in a corrupted leaf,
2983          * this may cause 'lost+found' dir has wrong nlinks.
2984          */
2985         cache = last_cache_extent(inode_cache);
2986         if (cache) {
2987                 node = container_of(cache, struct ptr_node, cache);
2988                 rec = node->data;
2989                 if (rec->ino > root->highest_inode)
2990                         root->highest_inode = rec->ino;
2991         }
2992
2993         /*
2994          * We need to repair backrefs first because we could change some of the
2995          * errors in the inode recs.
2996          *
2997          * We also need to go through and delete invalid backrefs first and then
2998          * add the correct ones second.  We do this because we may get EEXIST
2999          * when adding back the correct index because we hadn't yet deleted the
3000          * invalid index.
3001          *
3002          * For example, if we were missing a dir index then the directories
3003          * isize would be wrong, so if we fixed the isize to what we thought it
3004          * would be and then fixed the backref we'd still have a invalid fs, so
3005          * we need to add back the dir index and then check to see if the isize
3006          * is still wrong.
3007          */
3008         while (stage < 3) {
3009                 stage++;
3010                 if (stage == 3 && !err)
3011                         break;
3012
3013                 cache = search_cache_extent(inode_cache, 0);
3014                 while (repair && cache) {
3015                         node = container_of(cache, struct ptr_node, cache);
3016                         rec = node->data;
3017                         cache = next_cache_extent(cache);
3018
3019                         /* Need to free everything up and rescan */
3020                         if (stage == 3) {
3021                                 remove_cache_extent(inode_cache, &node->cache);
3022                                 free(node);
3023                                 free_inode_rec(rec);
3024                                 continue;
3025                         }
3026
3027                         if (list_empty(&rec->backrefs))
3028                                 continue;
3029
3030                         ret = repair_inode_backrefs(root, rec, inode_cache,
3031                                                     stage == 1);
3032                         if (ret < 0) {
3033                                 err = ret;
3034                                 stage = 2;
3035                                 break;
3036                         } if (ret > 0) {
3037                                 err = -EAGAIN;
3038                         }
3039                 }
3040         }
3041         if (err)
3042                 return err;
3043
3044         rec = get_inode_rec(inode_cache, root_dirid, 0);
3045         BUG_ON(IS_ERR(rec));
3046         if (rec) {
3047                 ret = check_root_dir(rec);
3048                 if (ret) {
3049                         fprintf(stderr, "root %llu root dir %llu error\n",
3050                                 (unsigned long long)root->root_key.objectid,
3051                                 (unsigned long long)root_dirid);
3052                         print_inode_error(root, rec);
3053                         error++;
3054                 }
3055         } else {
3056                 if (repair) {
3057                         struct btrfs_trans_handle *trans;
3058
3059                         trans = btrfs_start_transaction(root, 1);
3060                         if (IS_ERR(trans)) {
3061                                 err = PTR_ERR(trans);
3062                                 return err;
3063                         }
3064
3065                         fprintf(stderr,
3066                                 "root %llu missing its root dir, recreating\n",
3067                                 (unsigned long long)root->objectid);
3068
3069                         ret = btrfs_make_root_dir(trans, root, root_dirid);
3070                         BUG_ON(ret);
3071
3072                         btrfs_commit_transaction(trans, root);
3073                         return -EAGAIN;
3074                 }
3075
3076                 fprintf(stderr, "root %llu root dir %llu not found\n",
3077                         (unsigned long long)root->root_key.objectid,
3078                         (unsigned long long)root_dirid);
3079         }
3080
3081         while (1) {
3082                 cache = search_cache_extent(inode_cache, 0);
3083                 if (!cache)
3084                         break;
3085                 node = container_of(cache, struct ptr_node, cache);
3086                 rec = node->data;
3087                 remove_cache_extent(inode_cache, &node->cache);
3088                 free(node);
3089                 if (rec->ino == root_dirid ||
3090                     rec->ino == BTRFS_ORPHAN_OBJECTID) {
3091                         free_inode_rec(rec);
3092                         continue;
3093                 }
3094
3095                 if (rec->errors & I_ERR_NO_ORPHAN_ITEM) {
3096                         ret = check_orphan_item(root, rec->ino);
3097                         if (ret == 0)
3098                                 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
3099                         if (can_free_inode_rec(rec)) {
3100                                 free_inode_rec(rec);
3101                                 continue;
3102                         }
3103                 }
3104
3105                 if (!rec->found_inode_item)
3106                         rec->errors |= I_ERR_NO_INODE_ITEM;
3107                 if (rec->found_link != rec->nlink)
3108                         rec->errors |= I_ERR_LINK_COUNT_WRONG;
3109                 if (repair) {
3110                         ret = try_repair_inode(root, rec);
3111                         if (ret == 0 && can_free_inode_rec(rec)) {
3112                                 free_inode_rec(rec);
3113                                 continue;
3114                         }
3115                         ret = 0;
3116                 }
3117
3118                 if (!(repair && ret == 0))
3119                         error++;
3120                 print_inode_error(root, rec);
3121                 list_for_each_entry(backref, &rec->backrefs, list) {
3122                         if (!backref->found_dir_item)
3123                                 backref->errors |= REF_ERR_NO_DIR_ITEM;
3124                         if (!backref->found_dir_index)
3125                                 backref->errors |= REF_ERR_NO_DIR_INDEX;
3126                         if (!backref->found_inode_ref)
3127                                 backref->errors |= REF_ERR_NO_INODE_REF;
3128                         fprintf(stderr, "\tunresolved ref dir %llu index %llu"
3129                                 " namelen %u name %s filetype %d errors %x",
3130                                 (unsigned long long)backref->dir,
3131                                 (unsigned long long)backref->index,
3132                                 backref->namelen, backref->name,
3133                                 backref->filetype, backref->errors);
3134                         print_ref_error(backref->errors);
3135                 }
3136                 free_inode_rec(rec);
3137         }
3138         return (error > 0) ? -1 : 0;
3139 }
3140
3141 static struct root_record *get_root_rec(struct cache_tree *root_cache,
3142                                         u64 objectid)
3143 {
3144         struct cache_extent *cache;
3145         struct root_record *rec = NULL;
3146         int ret;
3147
3148         cache = lookup_cache_extent(root_cache, objectid, 1);
3149         if (cache) {
3150                 rec = container_of(cache, struct root_record, cache);
3151         } else {
3152                 rec = calloc(1, sizeof(*rec));
3153                 if (!rec)
3154                         return ERR_PTR(-ENOMEM);
3155                 rec->objectid = objectid;
3156                 INIT_LIST_HEAD(&rec->backrefs);
3157                 rec->cache.start = objectid;
3158                 rec->cache.size = 1;
3159
3160                 ret = insert_cache_extent(root_cache, &rec->cache);
3161                 if (ret)
3162                         return ERR_PTR(-EEXIST);
3163         }
3164         return rec;
3165 }
3166
3167 static struct root_backref *get_root_backref(struct root_record *rec,
3168                                              u64 ref_root, u64 dir, u64 index,
3169                                              const char *name, int namelen)
3170 {
3171         struct root_backref *backref;
3172
3173         list_for_each_entry(backref, &rec->backrefs, list) {
3174                 if (backref->ref_root != ref_root || backref->dir != dir ||
3175                     backref->namelen != namelen)
3176                         continue;
3177                 if (memcmp(name, backref->name, namelen))
3178                         continue;
3179                 return backref;
3180         }
3181
3182         backref = calloc(1, sizeof(*backref) + namelen + 1);
3183         if (!backref)
3184                 return NULL;
3185         backref->ref_root = ref_root;
3186         backref->dir = dir;
3187         backref->index = index;
3188         backref->namelen = namelen;
3189         memcpy(backref->name, name, namelen);
3190         backref->name[namelen] = '\0';
3191         list_add_tail(&backref->list, &rec->backrefs);
3192         return backref;
3193 }
3194
3195 static void free_root_record(struct cache_extent *cache)
3196 {
3197         struct root_record *rec;
3198         struct root_backref *backref;
3199
3200         rec = container_of(cache, struct root_record, cache);
3201         while (!list_empty(&rec->backrefs)) {
3202                 backref = to_root_backref(rec->backrefs.next);
3203                 list_del(&backref->list);
3204                 free(backref);
3205         }
3206
3207         free(rec);
3208 }
3209
3210 FREE_EXTENT_CACHE_BASED_TREE(root_recs, free_root_record);
3211
3212 static int add_root_backref(struct cache_tree *root_cache,
3213                             u64 root_id, u64 ref_root, u64 dir, u64 index,
3214                             const char *name, int namelen,
3215                             int item_type, int errors)
3216 {
3217         struct root_record *rec;
3218         struct root_backref *backref;
3219
3220         rec = get_root_rec(root_cache, root_id);
3221         BUG_ON(IS_ERR(rec));
3222         backref = get_root_backref(rec, ref_root, dir, index, name, namelen);
3223         BUG_ON(!backref);
3224
3225         backref->errors |= errors;
3226
3227         if (item_type != BTRFS_DIR_ITEM_KEY) {
3228                 if (backref->found_dir_index || backref->found_back_ref ||
3229                     backref->found_forward_ref) {
3230                         if (backref->index != index)
3231                                 backref->errors |= REF_ERR_INDEX_UNMATCH;
3232                 } else {
3233                         backref->index = index;
3234                 }
3235         }
3236
3237         if (item_type == BTRFS_DIR_ITEM_KEY) {
3238                 if (backref->found_forward_ref)
3239                         rec->found_ref++;
3240                 backref->found_dir_item = 1;
3241         } else if (item_type == BTRFS_DIR_INDEX_KEY) {
3242                 backref->found_dir_index = 1;
3243         } else if (item_type == BTRFS_ROOT_REF_KEY) {
3244                 if (backref->found_forward_ref)
3245                         backref->errors |= REF_ERR_DUP_ROOT_REF;
3246                 else if (backref->found_dir_item)
3247                         rec->found_ref++;
3248                 backref->found_forward_ref = 1;
3249         } else if (item_type == BTRFS_ROOT_BACKREF_KEY) {
3250                 if (backref->found_back_ref)
3251                         backref->errors |= REF_ERR_DUP_ROOT_BACKREF;
3252                 backref->found_back_ref = 1;
3253         } else {
3254                 BUG_ON(1);
3255         }
3256
3257         if (backref->found_forward_ref && backref->found_dir_item)
3258                 backref->reachable = 1;
3259         return 0;
3260 }
3261
3262 static int merge_root_recs(struct btrfs_root *root,
3263                            struct cache_tree *src_cache,
3264                            struct cache_tree *dst_cache)
3265 {
3266         struct cache_extent *cache;
3267         struct ptr_node *node;
3268         struct inode_record *rec;
3269         struct inode_backref *backref;
3270         int ret = 0;
3271
3272         if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
3273                 free_inode_recs_tree(src_cache);
3274                 return 0;
3275         }
3276
3277         while (1) {
3278                 cache = search_cache_extent(src_cache, 0);
3279                 if (!cache)
3280                         break;
3281                 node = container_of(cache, struct ptr_node, cache);
3282                 rec = node->data;
3283                 remove_cache_extent(src_cache, &node->cache);
3284                 free(node);
3285
3286                 ret = is_child_root(root, root->objectid, rec->ino);
3287                 if (ret < 0)
3288                         break;
3289                 else if (ret == 0)
3290                         goto skip;
3291
3292                 list_for_each_entry(backref, &rec->backrefs, list) {
3293                         BUG_ON(backref->found_inode_ref);
3294                         if (backref->found_dir_item)
3295                                 add_root_backref(dst_cache, rec->ino,
3296                                         root->root_key.objectid, backref->dir,
3297                                         backref->index, backref->name,
3298                                         backref->namelen, BTRFS_DIR_ITEM_KEY,
3299                                         backref->errors);
3300                         if (backref->found_dir_index)
3301                                 add_root_backref(dst_cache, rec->ino,
3302                                         root->root_key.objectid, backref->dir,
3303                                         backref->index, backref->name,
3304                                         backref->namelen, BTRFS_DIR_INDEX_KEY,
3305                                         backref->errors);
3306                 }
3307 skip:
3308                 free_inode_rec(rec);
3309         }
3310         if (ret < 0)
3311                 return ret;
3312         return 0;
3313 }
3314
3315 static int check_root_refs(struct btrfs_root *root,
3316                            struct cache_tree *root_cache)
3317 {
3318         struct root_record *rec;
3319         struct root_record *ref_root;
3320         struct root_backref *backref;
3321         struct cache_extent *cache;
3322         int loop = 1;
3323         int ret;
3324         int error;
3325         int errors = 0;
3326
3327         rec = get_root_rec(root_cache, BTRFS_FS_TREE_OBJECTID);
3328         BUG_ON(IS_ERR(rec));
3329         rec->found_ref = 1;
3330
3331         /* fixme: this can not detect circular references */
3332         while (loop) {
3333                 loop = 0;
3334                 cache = search_cache_extent(root_cache, 0);
3335                 while (1) {
3336                         if (!cache)
3337                                 break;
3338                         rec = container_of(cache, struct root_record, cache);
3339                         cache = next_cache_extent(cache);
3340
3341                         if (rec->found_ref == 0)
3342                                 continue;
3343
3344                         list_for_each_entry(backref, &rec->backrefs, list) {
3345                                 if (!backref->reachable)
3346                                         continue;
3347
3348                                 ref_root = get_root_rec(root_cache,
3349                                                         backref->ref_root);
3350                                 BUG_ON(IS_ERR(ref_root));
3351                                 if (ref_root->found_ref > 0)
3352                                         continue;
3353
3354                                 backref->reachable = 0;
3355                                 rec->found_ref--;
3356                                 if (rec->found_ref == 0)
3357                                         loop = 1;
3358                         }
3359                 }
3360         }
3361
3362         cache = search_cache_extent(root_cache, 0);
3363         while (1) {
3364                 if (!cache)
3365                         break;
3366                 rec = container_of(cache, struct root_record, cache);
3367                 cache = next_cache_extent(cache);
3368
3369                 if (rec->found_ref == 0 &&
3370                     rec->objectid >= BTRFS_FIRST_FREE_OBJECTID &&
3371                     rec->objectid <= BTRFS_LAST_FREE_OBJECTID) {
3372                         ret = check_orphan_item(root->fs_info->tree_root,
3373                                                 rec->objectid);
3374                         if (ret == 0)
3375                                 continue;
3376
3377                         /*
3378                          * If we don't have a root item then we likely just have
3379                          * a dir item in a snapshot for this root but no actual
3380                          * ref key or anything so it's meaningless.
3381                          */
3382                         if (!rec->found_root_item)
3383                                 continue;
3384                         errors++;
3385                         fprintf(stderr, "fs tree %llu not referenced\n",
3386                                 (unsigned long long)rec->objectid);
3387                 }
3388
3389                 error = 0;
3390                 if (rec->found_ref > 0 && !rec->found_root_item)
3391                         error = 1;
3392                 list_for_each_entry(backref, &rec->backrefs, list) {
3393                         if (!backref->found_dir_item)
3394                                 backref->errors |= REF_ERR_NO_DIR_ITEM;
3395                         if (!backref->found_dir_index)
3396                                 backref->errors |= REF_ERR_NO_DIR_INDEX;
3397                         if (!backref->found_back_ref)
3398                                 backref->errors |= REF_ERR_NO_ROOT_BACKREF;
3399                         if (!backref->found_forward_ref)
3400                                 backref->errors |= REF_ERR_NO_ROOT_REF;
3401                         if (backref->reachable && backref->errors)
3402                                 error = 1;
3403                 }
3404                 if (!error)
3405                         continue;
3406
3407                 errors++;
3408                 fprintf(stderr, "fs tree %llu refs %u %s\n",
3409                         (unsigned long long)rec->objectid, rec->found_ref,
3410                          rec->found_root_item ? "" : "not found");
3411
3412                 list_for_each_entry(backref, &rec->backrefs, list) {
3413                         if (!backref->reachable)
3414                                 continue;
3415                         if (!backref->errors && rec->found_root_item)
3416                                 continue;
3417                         fprintf(stderr, "\tunresolved ref root %llu dir %llu"
3418                                 " index %llu namelen %u name %s errors %x\n",
3419                                 (unsigned long long)backref->ref_root,
3420                                 (unsigned long long)backref->dir,
3421                                 (unsigned long long)backref->index,
3422                                 backref->namelen, backref->name,
3423                                 backref->errors);
3424                         print_ref_error(backref->errors);
3425                 }
3426         }
3427         return errors > 0 ? 1 : 0;
3428 }
3429
3430 static int process_root_ref(struct extent_buffer *eb, int slot,
3431                             struct btrfs_key *key,
3432                             struct cache_tree *root_cache)
3433 {
3434         u64 dirid;
3435         u64 index;
3436         u32 len;
3437         u32 name_len;
3438         struct btrfs_root_ref *ref;
3439         char namebuf[BTRFS_NAME_LEN];
3440         int error;
3441
3442         ref = btrfs_item_ptr(eb, slot, struct btrfs_root_ref);
3443
3444         dirid = btrfs_root_ref_dirid(eb, ref);
3445         index = btrfs_root_ref_sequence(eb, ref);
3446         name_len = btrfs_root_ref_name_len(eb, ref);
3447
3448         if (name_len <= BTRFS_NAME_LEN) {
3449                 len = name_len;
3450                 error = 0;
3451         } else {
3452                 len = BTRFS_NAME_LEN;
3453                 error = REF_ERR_NAME_TOO_LONG;
3454         }
3455         read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
3456
3457         if (key->type == BTRFS_ROOT_REF_KEY) {
3458                 add_root_backref(root_cache, key->offset, key->objectid, dirid,
3459                                  index, namebuf, len, key->type, error);
3460         } else {
3461                 add_root_backref(root_cache, key->objectid, key->offset, dirid,
3462                                  index, namebuf, len, key->type, error);
3463         }
3464         return 0;
3465 }
3466
3467 static void free_corrupt_block(struct cache_extent *cache)
3468 {
3469         struct btrfs_corrupt_block *corrupt;
3470
3471         corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
3472         free(corrupt);
3473 }
3474
3475 FREE_EXTENT_CACHE_BASED_TREE(corrupt_blocks, free_corrupt_block);
3476
3477 /*
3478  * Repair the btree of the given root.
3479  *
3480  * The fix is to remove the node key in corrupt_blocks cache_tree.
3481  * and rebalance the tree.
3482  * After the fix, the btree should be writeable.
3483  */
3484 static int repair_btree(struct btrfs_root *root,
3485                         struct cache_tree *corrupt_blocks)
3486 {
3487         struct btrfs_trans_handle *trans;
3488         struct btrfs_path path;
3489         struct btrfs_corrupt_block *corrupt;
3490         struct cache_extent *cache;
3491         struct btrfs_key key;
3492         u64 offset;
3493         int level;
3494         int ret = 0;
3495
3496         if (cache_tree_empty(corrupt_blocks))
3497                 return 0;
3498
3499         trans = btrfs_start_transaction(root, 1);
3500         if (IS_ERR(trans)) {
3501                 ret = PTR_ERR(trans);
3502                 fprintf(stderr, "Error starting transaction: %s\n",
3503                         strerror(-ret));
3504                 return ret;
3505         }
3506         btrfs_init_path(&path);
3507         cache = first_cache_extent(corrupt_blocks);
3508         while (cache) {
3509                 corrupt = container_of(cache, struct btrfs_corrupt_block,
3510                                        cache);
3511                 level = corrupt->level;
3512                 path.lowest_level = level;
3513                 key.objectid = corrupt->key.objectid;
3514                 key.type = corrupt->key.type;
3515                 key.offset = corrupt->key.offset;
3516
3517                 /*
3518                  * Here we don't want to do any tree balance, since it may
3519                  * cause a balance with corrupted brother leaf/node,
3520                  * so ins_len set to 0 here.
3521                  * Balance will be done after all corrupt node/leaf is deleted.
3522                  */
3523                 ret = btrfs_search_slot(trans, root, &key, &path, 0, 1);
3524                 if (ret < 0)
3525                         goto out;
3526                 offset = btrfs_node_blockptr(path.nodes[level],
3527                                              path.slots[level]);
3528
3529                 /* Remove the ptr */
3530                 ret = btrfs_del_ptr(trans, root, &path, level,
3531                                     path.slots[level]);
3532                 if (ret < 0)
3533                         goto out;
3534                 /*
3535                  * Remove the corresponding extent
3536                  * return value is not concerned.
3537                  */
3538                 btrfs_release_path(&path);
3539                 ret = btrfs_free_extent(trans, root, offset, root->nodesize,
3540                                         0, root->root_key.objectid,
3541                                         level - 1, 0);
3542                 cache = next_cache_extent(cache);
3543         }
3544
3545         /* Balance the btree using btrfs_search_slot() */
3546         cache = first_cache_extent(corrupt_blocks);
3547         while (cache) {
3548                 corrupt = container_of(cache, struct btrfs_corrupt_block,
3549                                        cache);
3550                 memcpy(&key, &corrupt->key, sizeof(key));
3551                 ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
3552                 if (ret < 0)
3553                         goto out;
3554                 /* return will always >0 since it won't find the item */
3555                 ret = 0;
3556                 btrfs_release_path(&path);
3557                 cache = next_cache_extent(cache);
3558         }
3559 out:
3560         btrfs_commit_transaction(trans, root);
3561         btrfs_release_path(&path);
3562         return ret;
3563 }
3564
3565 static int check_fs_root(struct btrfs_root *root,
3566                          struct cache_tree *root_cache,
3567                          struct walk_control *wc)
3568 {
3569         int ret = 0;
3570         int err = 0;
3571         int wret;
3572         int level;
3573         struct btrfs_path path;
3574         struct shared_node root_node;
3575         struct root_record *rec;
3576         struct btrfs_root_item *root_item = &root->root_item;
3577         struct cache_tree corrupt_blocks;
3578         struct orphan_data_extent *orphan;
3579         struct orphan_data_extent *tmp;
3580         enum btrfs_tree_block_status status;
3581         struct node_refs nrefs;
3582
3583         /*
3584          * Reuse the corrupt_block cache tree to record corrupted tree block
3585          *
3586          * Unlike the usage in extent tree check, here we do it in a per
3587          * fs/subvol tree base.
3588          */
3589         cache_tree_init(&corrupt_blocks);
3590         root->fs_info->corrupt_blocks = &corrupt_blocks;
3591
3592         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
3593                 rec = get_root_rec(root_cache, root->root_key.objectid);
3594                 BUG_ON(IS_ERR(rec));
3595                 if (btrfs_root_refs(root_item) > 0)
3596                         rec->found_root_item = 1;
3597         }
3598
3599         btrfs_init_path(&path);
3600         memset(&root_node, 0, sizeof(root_node));
3601         cache_tree_init(&root_node.root_cache);
3602         cache_tree_init(&root_node.inode_cache);
3603         memset(&nrefs, 0, sizeof(nrefs));
3604
3605         /* Move the orphan extent record to corresponding inode_record */
3606         list_for_each_entry_safe(orphan, tmp,
3607                                  &root->orphan_data_extents, list) {
3608                 struct inode_record *inode;
3609
3610                 inode = get_inode_rec(&root_node.inode_cache, orphan->objectid,
3611                                       1);
3612                 BUG_ON(IS_ERR(inode));
3613                 inode->errors |= I_ERR_FILE_EXTENT_ORPHAN;
3614                 list_move(&orphan->list, &inode->orphan_extents);
3615         }
3616
3617         level = btrfs_header_level(root->node);
3618         memset(wc->nodes, 0, sizeof(wc->nodes));
3619         wc->nodes[level] = &root_node;
3620         wc->active_node = level;
3621         wc->root_level = level;
3622
3623         /* We may not have checked the root block, lets do that now */
3624         if (btrfs_is_leaf(root->node))
3625                 status = btrfs_check_leaf(root, NULL, root->node);
3626         else
3627                 status = btrfs_check_node(root, NULL, root->node);
3628         if (status != BTRFS_TREE_BLOCK_CLEAN)
3629                 return -EIO;
3630
3631         if (btrfs_root_refs(root_item) > 0 ||
3632             btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3633                 path.nodes[level] = root->node;
3634                 extent_buffer_get(root->node);
3635                 path.slots[level] = 0;
3636         } else {
3637                 struct btrfs_key key;
3638                 struct btrfs_disk_key found_key;
3639
3640                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3641                 level = root_item->drop_level;
3642                 path.lowest_level = level;
3643                 if (level > btrfs_header_level(root->node) ||
3644                     level >= BTRFS_MAX_LEVEL) {
3645                         error("ignoring invalid drop level: %u", level);
3646                         goto skip_walking;
3647                 }
3648                 wret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
3649                 if (wret < 0)
3650                         goto skip_walking;
3651                 btrfs_node_key(path.nodes[level], &found_key,
3652                                 path.slots[level]);
3653                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3654                                         sizeof(found_key)));
3655         }
3656
3657         while (1) {
3658                 wret = walk_down_tree(root, &path, wc, &level, &nrefs);
3659                 if (wret < 0)
3660                         ret = wret;
3661                 if (wret != 0)
3662                         break;
3663
3664                 wret = walk_up_tree(root, &path, wc, &level);
3665                 if (wret < 0)
3666                         ret = wret;
3667                 if (wret != 0)
3668                         break;
3669         }
3670 skip_walking:
3671         btrfs_release_path(&path);
3672
3673         if (!cache_tree_empty(&corrupt_blocks)) {
3674                 struct cache_extent *cache;
3675                 struct btrfs_corrupt_block *corrupt;
3676
3677                 printf("The following tree block(s) is corrupted in tree %llu:\n",
3678                        root->root_key.objectid);
3679                 cache = first_cache_extent(&corrupt_blocks);
3680                 while (cache) {
3681                         corrupt = container_of(cache,
3682                                                struct btrfs_corrupt_block,
3683                                                cache);
3684                         printf("\ttree block bytenr: %llu, level: %d, node key: (%llu, %u, %llu)\n",
3685                                cache->start, corrupt->level,
3686                                corrupt->key.objectid, corrupt->key.type,
3687                                corrupt->key.offset);
3688                         cache = next_cache_extent(cache);
3689                 }
3690                 if (repair) {
3691                         printf("Try to repair the btree for root %llu\n",
3692                                root->root_key.objectid);
3693                         ret = repair_btree(root, &corrupt_blocks);
3694                         if (ret < 0)
3695                                 fprintf(stderr, "Failed to repair btree: %s\n",
3696                                         strerror(-ret));
3697                         if (!ret)
3698                                 printf("Btree for root %llu is fixed\n",
3699                                        root->root_key.objectid);
3700                 }
3701         }
3702
3703         err = merge_root_recs(root, &root_node.root_cache, root_cache);
3704         if (err < 0)
3705                 ret = err;
3706
3707         if (root_node.current) {
3708                 root_node.current->checked = 1;
3709                 maybe_free_inode_rec(&root_node.inode_cache,
3710                                 root_node.current);
3711         }
3712
3713         err = check_inode_recs(root, &root_node.inode_cache);
3714         if (!ret)
3715                 ret = err;
3716
3717         free_corrupt_blocks_tree(&corrupt_blocks);
3718         root->fs_info->corrupt_blocks = NULL;
3719         free_orphan_data_extents(&root->orphan_data_extents);
3720         return ret;
3721 }
3722
3723 static int fs_root_objectid(u64 objectid)
3724 {
3725         if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
3726             objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3727                 return 1;
3728         return is_fstree(objectid);
3729 }
3730
3731 static int check_fs_roots(struct btrfs_root *root,
3732                           struct cache_tree *root_cache)
3733 {
3734         struct btrfs_path path;
3735         struct btrfs_key key;
3736         struct walk_control wc;
3737         struct extent_buffer *leaf, *tree_node;
3738         struct btrfs_root *tmp_root;
3739         struct btrfs_root *tree_root = root->fs_info->tree_root;
3740         int ret;
3741         int err = 0;
3742
3743         if (ctx.progress_enabled) {
3744                 ctx.tp = TASK_FS_ROOTS;
3745                 task_start(ctx.info);
3746         }
3747
3748         /*
3749          * Just in case we made any changes to the extent tree that weren't
3750          * reflected into the free space cache yet.
3751          */
3752         if (repair)
3753                 reset_cached_block_groups(root->fs_info);
3754         memset(&wc, 0, sizeof(wc));
3755         cache_tree_init(&wc.shared);
3756         btrfs_init_path(&path);
3757
3758 again:
3759         key.offset = 0;
3760         key.objectid = 0;
3761         key.type = BTRFS_ROOT_ITEM_KEY;
3762         ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
3763         if (ret < 0) {
3764                 err = 1;
3765                 goto out;
3766         }
3767         tree_node = tree_root->node;
3768         while (1) {
3769                 if (tree_node != tree_root->node) {
3770                         free_root_recs_tree(root_cache);
3771                         btrfs_release_path(&path);
3772                         goto again;
3773                 }
3774                 leaf = path.nodes[0];
3775                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
3776                         ret = btrfs_next_leaf(tree_root, &path);
3777                         if (ret) {
3778                                 if (ret < 0)
3779                                         err = 1;
3780                                 break;
3781                         }
3782                         leaf = path.nodes[0];
3783                 }
3784                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
3785                 if (key.type == BTRFS_ROOT_ITEM_KEY &&
3786                     fs_root_objectid(key.objectid)) {
3787                         if (key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
3788                                 tmp_root = btrfs_read_fs_root_no_cache(
3789                                                 root->fs_info, &key);
3790                         } else {
3791                                 key.offset = (u64)-1;
3792                                 tmp_root = btrfs_read_fs_root(
3793                                                 root->fs_info, &key);
3794                         }
3795                         if (IS_ERR(tmp_root)) {
3796                                 err = 1;
3797                                 goto next;
3798                         }
3799                         ret = check_fs_root(tmp_root, root_cache, &wc);
3800                         if (ret == -EAGAIN) {
3801                                 free_root_recs_tree(root_cache);
3802                                 btrfs_release_path(&path);
3803                                 goto again;
3804                         }
3805                         if (ret)
3806                                 err = 1;
3807                         if (key.objectid == BTRFS_TREE_RELOC_OBJECTID)
3808                                 btrfs_free_fs_root(tmp_root);
3809                 } else if (key.type == BTRFS_ROOT_REF_KEY ||
3810                            key.type == BTRFS_ROOT_BACKREF_KEY) {
3811                         process_root_ref(leaf, path.slots[0], &key,
3812                                          root_cache);
3813                 }
3814 next:
3815                 path.slots[0]++;
3816         }
3817 out:
3818         btrfs_release_path(&path);
3819         if (err)
3820                 free_extent_cache_tree(&wc.shared);
3821         if (!cache_tree_empty(&wc.shared))
3822                 fprintf(stderr, "warning line %d\n", __LINE__);
3823
3824         task_stop(ctx.info);
3825
3826         return err;
3827 }
3828
3829 static int all_backpointers_checked(struct extent_record *rec, int print_errs)
3830 {
3831         struct list_head *cur = rec->backrefs.next;
3832         struct extent_backref *back;
3833         struct tree_backref *tback;
3834         struct data_backref *dback;
3835         u64 found = 0;
3836         int err = 0;
3837
3838         while(cur != &rec->backrefs) {
3839                 back = to_extent_backref(cur);
3840                 cur = cur->next;
3841                 if (!back->found_extent_tree) {
3842                         err = 1;
3843                         if (!print_errs)
3844                                 goto out;
3845                         if (back->is_data) {
3846                                 dback = to_data_backref(back);
3847                                 fprintf(stderr, "Backref %llu %s %llu"
3848                                         " owner %llu offset %llu num_refs %lu"
3849                                         " not found in extent tree\n",
3850                                         (unsigned long long)rec->start,
3851                                         back->full_backref ?
3852                                         "parent" : "root",
3853                                         back->full_backref ?
3854                                         (unsigned long long)dback->parent:
3855                                         (unsigned long long)dback->root,
3856                                         (unsigned long long)dback->owner,
3857                                         (unsigned long long)dback->offset,
3858                                         (unsigned long)dback->num_refs);
3859                         } else {
3860                                 tback = to_tree_backref(back);
3861                                 fprintf(stderr, "Backref %llu parent %llu"
3862                                         " root %llu not found in extent tree\n",
3863                                         (unsigned long long)rec->start,
3864                                         (unsigned long long)tback->parent,
3865                                         (unsigned long long)tback->root);
3866                         }
3867                 }
3868                 if (!back->is_data && !back->found_ref) {
3869                         err = 1;
3870                         if (!print_errs)
3871                                 goto out;
3872                         tback = to_tree_backref(back);
3873                         fprintf(stderr, "Backref %llu %s %llu not referenced back %p\n",
3874                                 (unsigned long long)rec->start,
3875                                 back->full_backref ? "parent" : "root",
3876                                 back->full_backref ?
3877                                 (unsigned long long)tback->parent :
3878                                 (unsigned long long)tback->root, back);
3879                 }
3880                 if (back->is_data) {
3881                         dback = to_data_backref(back);
3882                         if (dback->found_ref != dback->num_refs) {
3883                                 err = 1;
3884                                 if (!print_errs)
3885                                         goto out;
3886                                 fprintf(stderr, "Incorrect local backref count"
3887                                         " on %llu %s %llu owner %llu"
3888                                         " offset %llu found %u wanted %u back %p\n",
3889                                         (unsigned long long)rec->start,
3890                                         back->full_backref ?
3891                                         "parent" : "root",
3892                                         back->full_backref ?
3893                                         (unsigned long long)dback->parent:
3894                                         (unsigned long long)dback->root,
3895                                         (unsigned long long)dback->owner,
3896                                         (unsigned long long)dback->offset,
3897                                         dback->found_ref, dback->num_refs, back);
3898                         }
3899                         if (dback->disk_bytenr != rec->start) {
3900                                 err = 1;
3901                                 if (!print_errs)
3902                                         goto out;
3903                                 fprintf(stderr, "Backref disk bytenr does not"
3904                                         " match extent record, bytenr=%llu, "
3905                                         "ref bytenr=%llu\n",
3906                                         (unsigned long long)rec->start,
3907                                         (unsigned long long)dback->disk_bytenr);
3908                         }
3909
3910                         if (dback->bytes != rec->nr) {
3911                                 err = 1;
3912                                 if (!print_errs)
3913                                         goto out;
3914                                 fprintf(stderr, "Backref bytes do not match "
3915                                         "extent backref, bytenr=%llu, ref "
3916                                         "bytes=%llu, backref bytes=%llu\n",
3917                                         (unsigned long long)rec->start,
3918                                         (unsigned long long)rec->nr,
3919                                         (unsigned long long)dback->bytes);
3920                         }
3921                 }
3922                 if (!back->is_data) {
3923                         found += 1;
3924                 } else {
3925                         dback = to_data_backref(back);
3926                         found += dback->found_ref;
3927                 }
3928         }
3929         if (found != rec->refs) {
3930                 err = 1;
3931                 if (!print_errs)
3932                         goto out;
3933                 fprintf(stderr, "Incorrect global backref count "
3934                         "on %llu found %llu wanted %llu\n",
3935                         (unsigned long long)rec->start,
3936                         (unsigned long long)found,
3937                         (unsigned long long)rec->refs);
3938         }
3939 out:
3940         return err;
3941 }
3942
3943 static int free_all_extent_backrefs(struct extent_record *rec)
3944 {
3945         struct extent_backref *back;
3946         struct list_head *cur;
3947         while (!list_empty(&rec->backrefs)) {
3948                 cur = rec->backrefs.next;
3949                 back = to_extent_backref(cur);
3950                 list_del(cur);
3951                 free(back);
3952         }
3953         return 0;
3954 }
3955
3956 static void free_extent_record_cache(struct btrfs_fs_info *fs_info,
3957                                      struct cache_tree *extent_cache)
3958 {
3959         struct cache_extent *cache;
3960         struct extent_record *rec;
3961
3962         while (1) {
3963                 cache = first_cache_extent(extent_cache);
3964                 if (!cache)
3965                         break;
3966                 rec = container_of(cache, struct extent_record, cache);
3967                 remove_cache_extent(extent_cache, cache);
3968                 free_all_extent_backrefs(rec);
3969                 free(rec);
3970         }
3971 }
3972
3973 static int maybe_free_extent_rec(struct cache_tree *extent_cache,
3974                                  struct extent_record *rec)
3975 {
3976         if (rec->content_checked && rec->owner_ref_checked &&
3977             rec->extent_item_refs == rec->refs && rec->refs > 0 &&
3978             rec->num_duplicates == 0 && !all_backpointers_checked(rec, 0) &&
3979             !rec->bad_full_backref && !rec->crossing_stripes &&
3980             !rec->wrong_chunk_type) {
3981                 remove_cache_extent(extent_cache, &rec->cache);
3982                 free_all_extent_backrefs(rec);
3983                 list_del_init(&rec->list);
3984                 free(rec);
3985         }
3986         return 0;
3987 }
3988
3989 static int check_owner_ref(struct btrfs_root *root,
3990                             struct extent_record *rec,
3991                             struct extent_buffer *buf)
3992 {
3993         struct extent_backref *node;
3994         struct tree_backref *back;
3995         struct btrfs_root *ref_root;
3996         struct btrfs_key key;
3997         struct btrfs_path path;
3998         struct extent_buffer *parent;
3999         int level;
4000         int found = 0;
4001         int ret;
4002
4003         list_for_each_entry(node, &rec->backrefs, list) {
4004                 if (node->is_data)
4005                         continue;
4006                 if (!node->found_ref)
4007                         continue;
4008                 if (node->full_backref)
4009                         continue;
4010                 back = to_tree_backref(node);
4011                 if (btrfs_header_owner(buf) == back->root)
4012                         return 0;
4013         }
4014         BUG_ON(rec->is_root);
4015
4016         /* try to find the block by search corresponding fs tree */
4017         key.objectid = btrfs_header_owner(buf);
4018         key.type = BTRFS_ROOT_ITEM_KEY;
4019         key.offset = (u64)-1;
4020
4021         ref_root = btrfs_read_fs_root(root->fs_info, &key);
4022         if (IS_ERR(ref_root))
4023                 return 1;
4024
4025         level = btrfs_header_level(buf);
4026         if (level == 0)
4027                 btrfs_item_key_to_cpu(buf, &key, 0);
4028         else
4029                 btrfs_node_key_to_cpu(buf, &key, 0);
4030
4031         btrfs_init_path(&path);
4032         path.lowest_level = level + 1;
4033         ret = btrfs_search_slot(NULL, ref_root, &key, &path, 0, 0);
4034         if (ret < 0)
4035                 return 0;
4036
4037         parent = path.nodes[level + 1];
4038         if (parent && buf->start == btrfs_node_blockptr(parent,
4039                                                         path.slots[level + 1]))
4040                 found = 1;
4041
4042         btrfs_release_path(&path);
4043         return found ? 0 : 1;
4044 }
4045
4046 static int is_extent_tree_record(struct extent_record *rec)
4047 {
4048         struct list_head *cur = rec->backrefs.next;
4049         struct extent_backref *node;
4050         struct tree_backref *back;
4051         int is_extent = 0;
4052
4053         while(cur != &rec->backrefs) {
4054                 node = to_extent_backref(cur);
4055                 cur = cur->next;
4056                 if (node->is_data)
4057                         return 0;
4058                 back = to_tree_backref(node);
4059                 if (node->full_backref)
4060                         return 0;
4061                 if (back->root == BTRFS_EXTENT_TREE_OBJECTID)
4062                         is_extent = 1;
4063         }
4064         return is_extent;
4065 }
4066
4067
4068 static int record_bad_block_io(struct btrfs_fs_info *info,
4069                                struct cache_tree *extent_cache,
4070                                u64 start, u64 len)
4071 {
4072         struct extent_record *rec;
4073         struct cache_extent *cache;
4074         struct btrfs_key key;
4075
4076         cache = lookup_cache_extent(extent_cache, start, len);
4077         if (!cache)
4078                 return 0;
4079
4080         rec = container_of(cache, struct extent_record, cache);
4081         if (!is_extent_tree_record(rec))
4082                 return 0;
4083
4084         btrfs_disk_key_to_cpu(&key, &rec->parent_key);
4085         return btrfs_add_corrupt_extent_record(info, &key, start, len, 0);
4086 }
4087
4088 static int swap_values(struct btrfs_root *root, struct btrfs_path *path,
4089                        struct extent_buffer *buf, int slot)
4090 {
4091         if (btrfs_header_level(buf)) {
4092                 struct btrfs_key_ptr ptr1, ptr2;
4093
4094                 read_extent_buffer(buf, &ptr1, btrfs_node_key_ptr_offset(slot),
4095                                    sizeof(struct btrfs_key_ptr));
4096                 read_extent_buffer(buf, &ptr2,
4097                                    btrfs_node_key_ptr_offset(slot + 1),
4098                                    sizeof(struct btrfs_key_ptr));
4099                 write_extent_buffer(buf, &ptr1,
4100                                     btrfs_node_key_ptr_offset(slot + 1),
4101                                     sizeof(struct btrfs_key_ptr));
4102                 write_extent_buffer(buf, &ptr2,
4103                                     btrfs_node_key_ptr_offset(slot),
4104                                     sizeof(struct btrfs_key_ptr));
4105                 if (slot == 0) {
4106                         struct btrfs_disk_key key;
4107                         btrfs_node_key(buf, &key, 0);
4108                         btrfs_fixup_low_keys(root, path, &key,
4109                                              btrfs_header_level(buf) + 1);
4110                 }
4111         } else {
4112                 struct btrfs_item *item1, *item2;
4113                 struct btrfs_key k1, k2;
4114                 char *item1_data, *item2_data;
4115                 u32 item1_offset, item2_offset, item1_size, item2_size;
4116
4117                 item1 = btrfs_item_nr(slot);
4118                 item2 = btrfs_item_nr(slot + 1);
4119                 btrfs_item_key_to_cpu(buf, &k1, slot);
4120                 btrfs_item_key_to_cpu(buf, &k2, slot + 1);
4121                 item1_offset = btrfs_item_offset(buf, item1);
4122                 item2_offset = btrfs_item_offset(buf, item2);
4123                 item1_size = btrfs_item_size(buf, item1);
4124                 item2_size = btrfs_item_size(buf, item2);
4125
4126                 item1_data = malloc(item1_size);
4127                 if (!item1_data)
4128                         return -ENOMEM;
4129                 item2_data = malloc(item2_size);
4130                 if (!item2_data) {
4131                         free(item1_data);
4132                         return -ENOMEM;
4133                 }
4134
4135                 read_extent_buffer(buf, item1_data, item1_offset, item1_size);
4136                 read_extent_buffer(buf, item2_data, item2_offset, item2_size);
4137
4138                 write_extent_buffer(buf, item1_data, item2_offset, item2_size);
4139                 write_extent_buffer(buf, item2_data, item1_offset, item1_size);
4140                 free(item1_data);
4141                 free(item2_data);
4142
4143                 btrfs_set_item_offset(buf, item1, item2_offset);
4144                 btrfs_set_item_offset(buf, item2, item1_offset);
4145                 btrfs_set_item_size(buf, item1, item2_size);
4146                 btrfs_set_item_size(buf, item2, item1_size);
4147
4148                 path->slots[0] = slot;
4149                 btrfs_set_item_key_unsafe(root, path, &k2);
4150                 path->slots[0] = slot + 1;
4151                 btrfs_set_item_key_unsafe(root, path, &k1);
4152         }
4153         return 0;
4154 }
4155
4156 static int fix_key_order(struct btrfs_trans_handle *trans,
4157                          struct btrfs_root *root,
4158                          struct btrfs_path *path)
4159 {
4160         struct extent_buffer *buf;
4161         struct btrfs_key k1, k2;
4162         int i;
4163         int level = path->lowest_level;
4164         int ret = -EIO;
4165
4166         buf = path->nodes[level];
4167         for (i = 0; i < btrfs_header_nritems(buf) - 1; i++) {
4168                 if (level) {
4169                         btrfs_node_key_to_cpu(buf, &k1, i);
4170                         btrfs_node_key_to_cpu(buf, &k2, i + 1);
4171                 } else {
4172                         btrfs_item_key_to_cpu(buf, &k1, i);
4173                         btrfs_item_key_to_cpu(buf, &k2, i + 1);
4174                 }
4175                 if (btrfs_comp_cpu_keys(&k1, &k2) < 0)
4176                         continue;
4177                 ret = swap_values(root, path, buf, i);
4178                 if (ret)
4179                         break;
4180                 btrfs_mark_buffer_dirty(buf);
4181                 i = 0;
4182         }
4183         return ret;
4184 }
4185
4186 static int delete_bogus_item(struct btrfs_trans_handle *trans,
4187                              struct btrfs_root *root,
4188                              struct btrfs_path *path,
4189                              struct extent_buffer *buf, int slot)
4190 {
4191         struct btrfs_key key;
4192         int nritems = btrfs_header_nritems(buf);
4193
4194         btrfs_item_key_to_cpu(buf, &key, slot);
4195
4196         /* These are all the keys we can deal with missing. */
4197         if (key.type != BTRFS_DIR_INDEX_KEY &&
4198             key.type != BTRFS_EXTENT_ITEM_KEY &&
4199             key.type != BTRFS_METADATA_ITEM_KEY &&
4200             key.type != BTRFS_TREE_BLOCK_REF_KEY &&
4201             key.type != BTRFS_EXTENT_DATA_REF_KEY)
4202                 return -1;
4203
4204         printf("Deleting bogus item [%llu,%u,%llu] at slot %d on block %llu\n",
4205                (unsigned long long)key.objectid, key.type,
4206                (unsigned long long)key.offset, slot, buf->start);
4207         memmove_extent_buffer(buf, btrfs_item_nr_offset(slot),
4208                               btrfs_item_nr_offset(slot + 1),
4209                               sizeof(struct btrfs_item) *
4210                               (nritems - slot - 1));
4211         btrfs_set_header_nritems(buf, nritems - 1);
4212         if (slot == 0) {
4213                 struct btrfs_disk_key disk_key;
4214
4215                 btrfs_item_key(buf, &disk_key, 0);
4216                 btrfs_fixup_low_keys(root, path, &disk_key, 1);
4217         }
4218         btrfs_mark_buffer_dirty(buf);
4219         return 0;
4220 }
4221
4222 static int fix_item_offset(struct btrfs_trans_handle *trans,
4223                            struct btrfs_root *root,
4224                            struct btrfs_path *path)
4225 {
4226         struct extent_buffer *buf;
4227         int i;
4228         int ret = 0;
4229
4230         /* We should only get this for leaves */
4231         BUG_ON(path->lowest_level);
4232         buf = path->nodes[0];
4233 again:
4234         for (i = 0; i < btrfs_header_nritems(buf); i++) {
4235                 unsigned int shift = 0, offset;
4236
4237                 if (i == 0 && btrfs_item_end_nr(buf, i) !=
4238                     BTRFS_LEAF_DATA_SIZE(root)) {
4239                         if (btrfs_item_end_nr(buf, i) >
4240                             BTRFS_LEAF_DATA_SIZE(root)) {
4241                                 ret = delete_bogus_item(trans, root, path,
4242                                                         buf, i);
4243                                 if (!ret)
4244                                         goto again;
4245                                 fprintf(stderr, "item is off the end of the "
4246                                         "leaf, can't fix\n");
4247                                 ret = -EIO;
4248                                 break;
4249                         }
4250                         shift = BTRFS_LEAF_DATA_SIZE(root) -
4251                                 btrfs_item_end_nr(buf, i);
4252                 } else if (i > 0 && btrfs_item_end_nr(buf, i) !=
4253                            btrfs_item_offset_nr(buf, i - 1)) {
4254                         if (btrfs_item_end_nr(buf, i) >
4255                             btrfs_item_offset_nr(buf, i - 1)) {
4256                                 ret = delete_bogus_item(trans, root, path,
4257                                                         buf, i);
4258                                 if (!ret)
4259                                         goto again;
4260                                 fprintf(stderr, "items overlap, can't fix\n");
4261                                 ret = -EIO;
4262                                 break;
4263                         }
4264                         shift = btrfs_item_offset_nr(buf, i - 1) -
4265                                 btrfs_item_end_nr(buf, i);
4266                 }
4267                 if (!shift)
4268                         continue;
4269
4270                 printf("Shifting item nr %d by %u bytes in block %llu\n",
4271                        i, shift, (unsigned long long)buf->start);
4272                 offset = btrfs_item_offset_nr(buf, i);
4273                 memmove_extent_buffer(buf,
4274                                       btrfs_leaf_data(buf) + offset + shift,
4275                                       btrfs_leaf_data(buf) + offset,
4276                                       btrfs_item_size_nr(buf, i));
4277                 btrfs_set_item_offset(buf, btrfs_item_nr(i),
4278                                       offset + shift);
4279                 btrfs_mark_buffer_dirty(buf);
4280         }
4281
4282         /*
4283          * We may have moved things, in which case we want to exit so we don't
4284          * write those changes out.  Once we have proper abort functionality in
4285          * progs this can be changed to something nicer.
4286          */
4287         BUG_ON(ret);
4288         return ret;
4289 }
4290
4291 /*
4292  * Attempt to fix basic block failures.  If we can't fix it for whatever reason
4293  * then just return -EIO.
4294  */
4295 static int try_to_fix_bad_block(struct btrfs_root *root,
4296                                 struct extent_buffer *buf,
4297                                 enum btrfs_tree_block_status status)
4298 {
4299         struct btrfs_trans_handle *trans;
4300         struct ulist *roots;
4301         struct ulist_node *node;
4302         struct btrfs_root *search_root;
4303         struct btrfs_path path;
4304         struct ulist_iterator iter;
4305         struct btrfs_key root_key, key;
4306         int ret;
4307
4308         if (status != BTRFS_TREE_BLOCK_BAD_KEY_ORDER &&
4309             status != BTRFS_TREE_BLOCK_INVALID_OFFSETS)
4310                 return -EIO;
4311
4312         ret = btrfs_find_all_roots(NULL, root->fs_info, buf->start, 0, &roots);
4313         if (ret)
4314                 return -EIO;
4315
4316         btrfs_init_path(&path);
4317         ULIST_ITER_INIT(&iter);
4318         while ((node = ulist_next(roots, &iter))) {
4319                 root_key.objectid = node->val;
4320                 root_key.type = BTRFS_ROOT_ITEM_KEY;
4321                 root_key.offset = (u64)-1;
4322
4323                 search_root = btrfs_read_fs_root(root->fs_info, &root_key);
4324                 if (IS_ERR(root)) {
4325                         ret = -EIO;
4326                         break;
4327                 }
4328
4329
4330                 trans = btrfs_start_transaction(search_root, 0);
4331                 if (IS_ERR(trans)) {
4332                         ret = PTR_ERR(trans);
4333                         break;
4334                 }
4335
4336                 path.lowest_level = btrfs_header_level(buf);
4337                 path.skip_check_block = 1;
4338                 if (path.lowest_level)
4339                         btrfs_node_key_to_cpu(buf, &key, 0);
4340                 else
4341                         btrfs_item_key_to_cpu(buf, &key, 0);
4342                 ret = btrfs_search_slot(trans, search_root, &key, &path, 0, 1);
4343                 if (ret) {
4344                         ret = -EIO;
4345                         btrfs_commit_transaction(trans, search_root);
4346                         break;
4347                 }
4348                 if (status == BTRFS_TREE_BLOCK_BAD_KEY_ORDER)
4349                         ret = fix_key_order(trans, search_root, &path);
4350                 else if (status == BTRFS_TREE_BLOCK_INVALID_OFFSETS)
4351                         ret = fix_item_offset(trans, search_root, &path);
4352                 if (ret) {
4353                         btrfs_commit_transaction(trans, search_root);
4354                         break;
4355                 }
4356                 btrfs_release_path(&path);
4357                 btrfs_commit_transaction(trans, search_root);
4358         }
4359         ulist_free(roots);
4360         btrfs_release_path(&path);
4361         return ret;
4362 }
4363
4364 static int check_block(struct btrfs_root *root,
4365                        struct cache_tree *extent_cache,
4366                        struct extent_buffer *buf, u64 flags)
4367 {
4368         struct extent_record *rec;
4369         struct cache_extent *cache;
4370         struct btrfs_key key;
4371         enum btrfs_tree_block_status status;
4372         int ret = 0;
4373         int level;
4374
4375         cache = lookup_cache_extent(extent_cache, buf->start, buf->len);
4376         if (!cache)
4377                 return 1;
4378         rec = container_of(cache, struct extent_record, cache);
4379         rec->generation = btrfs_header_generation(buf);
4380
4381         level = btrfs_header_level(buf);
4382         if (btrfs_header_nritems(buf) > 0) {
4383
4384                 if (level == 0)
4385                         btrfs_item_key_to_cpu(buf, &key, 0);
4386                 else
4387                         btrfs_node_key_to_cpu(buf, &key, 0);
4388
4389                 rec->info_objectid = key.objectid;
4390         }
4391         rec->info_level = level;
4392
4393         if (btrfs_is_leaf(buf))
4394                 status = btrfs_check_leaf(root, &rec->parent_key, buf);
4395         else
4396                 status = btrfs_check_node(root, &rec->parent_key, buf);
4397
4398         if (status != BTRFS_TREE_BLOCK_CLEAN) {
4399                 if (repair)
4400                         status = try_to_fix_bad_block(root, buf, status);
4401                 if (status != BTRFS_TREE_BLOCK_CLEAN) {
4402                         ret = -EIO;
4403                         fprintf(stderr, "bad block %llu\n",
4404                                 (unsigned long long)buf->start);
4405                 } else {
4406                         /*
4407                          * Signal to callers we need to start the scan over
4408                          * again since we'll have cowed blocks.
4409                          */
4410                         ret = -EAGAIN;
4411                 }
4412         } else {
4413                 rec->content_checked = 1;
4414                 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
4415                         rec->owner_ref_checked = 1;
4416                 else {
4417                         ret = check_owner_ref(root, rec, buf);
4418                         if (!ret)
4419                                 rec->owner_ref_checked = 1;
4420                 }
4421         }
4422         if (!ret)
4423                 maybe_free_extent_rec(extent_cache, rec);
4424         return ret;
4425 }
4426
4427 static struct tree_backref *find_tree_backref(struct extent_record *rec,
4428                                                 u64 parent, u64 root)
4429 {
4430         struct list_head *cur = rec->backrefs.next;
4431         struct extent_backref *node;
4432         struct tree_backref *back;
4433
4434         while(cur != &rec->backrefs) {
4435                 node = to_extent_backref(cur);
4436                 cur = cur->next;
4437                 if (node->is_data)
4438                         continue;
4439                 back = to_tree_backref(node);
4440                 if (parent > 0) {
4441                         if (!node->full_backref)
4442                                 continue;
4443                         if (parent == back->parent)
4444                                 return back;
4445                 } else {
4446                         if (node->full_backref)
4447                                 continue;
4448                         if (back->root == root)
4449                                 return back;
4450                 }
4451         }
4452         return NULL;
4453 }
4454
4455 static struct tree_backref *alloc_tree_backref(struct extent_record *rec,
4456                                                 u64 parent, u64 root)
4457 {
4458         struct tree_backref *ref = malloc(sizeof(*ref));
4459
4460         if (!ref)
4461                 return NULL;
4462         memset(&ref->node, 0, sizeof(ref->node));
4463         if (parent > 0) {
4464                 ref->parent = parent;
4465                 ref->node.full_backref = 1;
4466         } else {
4467                 ref->root = root;
4468                 ref->node.full_backref = 0;
4469         }
4470         list_add_tail(&ref->node.list, &rec->backrefs);
4471
4472         return ref;
4473 }
4474
4475 static struct data_backref *find_data_backref(struct extent_record *rec,
4476                                                 u64 parent, u64 root,
4477                                                 u64 owner, u64 offset,
4478                                                 int found_ref,
4479                                                 u64 disk_bytenr, u64 bytes)
4480 {
4481         struct list_head *cur = rec->backrefs.next;
4482         struct extent_backref *node;
4483         struct data_backref *back;
4484
4485         while(cur != &rec->backrefs) {
4486                 node = to_extent_backref(cur);
4487                 cur = cur->next;
4488                 if (!node->is_data)
4489                         continue;
4490                 back = to_data_backref(node);
4491                 if (parent > 0) {
4492                         if (!node->full_backref)
4493                                 continue;
4494                         if (parent == back->parent)
4495                                 return back;
4496                 } else {
4497                         if (node->full_backref)
4498                                 continue;
4499                         if (back->root == root && back->owner == owner &&
4500                             back->offset == offset) {
4501                                 if (found_ref && node->found_ref &&
4502                                     (back->bytes != bytes ||
4503                                     back->disk_bytenr != disk_bytenr))
4504                                         continue;
4505                                 return back;
4506                         }
4507                 }
4508         }
4509         return NULL;
4510 }
4511
4512 static struct data_backref *alloc_data_backref(struct extent_record *rec,
4513                                                 u64 parent, u64 root,
4514                                                 u64 owner, u64 offset,
4515                                                 u64 max_size)
4516 {
4517         struct data_backref *ref = malloc(sizeof(*ref));
4518
4519         if (!ref)
4520                 return NULL;
4521         memset(&ref->node, 0, sizeof(ref->node));
4522         ref->node.is_data = 1;
4523
4524         if (parent > 0) {
4525                 ref->parent = parent;
4526                 ref->owner = 0;
4527                 ref->offset = 0;
4528                 ref->node.full_backref = 1;
4529         } else {
4530                 ref->root = root;
4531                 ref->owner = owner;
4532                 ref->offset = offset;
4533                 ref->node.full_backref = 0;
4534         }
4535         ref->bytes = max_size;
4536         ref->found_ref = 0;
4537         ref->num_refs = 0;
4538         list_add_tail(&ref->node.list, &rec->backrefs);
4539         if (max_size > rec->max_size)
4540                 rec->max_size = max_size;
4541         return ref;
4542 }
4543
4544 /* Check if the type of extent matches with its chunk */
4545 static void check_extent_type(struct extent_record *rec)
4546 {
4547         struct btrfs_block_group_cache *bg_cache;
4548
4549         bg_cache = btrfs_lookup_first_block_group(global_info, rec->start);
4550         if (!bg_cache)
4551                 return;
4552
4553         /* data extent, check chunk directly*/
4554         if (!rec->metadata) {
4555                 if (!(bg_cache->flags & BTRFS_BLOCK_GROUP_DATA))
4556                         rec->wrong_chunk_type = 1;
4557                 return;
4558         }
4559
4560         /* metadata extent, check the obvious case first */
4561         if (!(bg_cache->flags & (BTRFS_BLOCK_GROUP_SYSTEM |
4562                                  BTRFS_BLOCK_GROUP_METADATA))) {
4563                 rec->wrong_chunk_type = 1;
4564                 return;
4565         }
4566
4567         /*
4568          * Check SYSTEM extent, as it's also marked as metadata, we can only
4569          * make sure it's a SYSTEM extent by its backref
4570          */
4571         if (!list_empty(&rec->backrefs)) {
4572                 struct extent_backref *node;
4573                 struct tree_backref *tback;
4574                 u64 bg_type;
4575
4576                 node = to_extent_backref(rec->backrefs.next);
4577                 if (node->is_data) {
4578                         /* tree block shouldn't have data backref */
4579                         rec->wrong_chunk_type = 1;
4580                         return;
4581                 }
4582                 tback = container_of(node, struct tree_backref, node);
4583
4584                 if (tback->root == BTRFS_CHUNK_TREE_OBJECTID)
4585                         bg_type = BTRFS_BLOCK_GROUP_SYSTEM;
4586                 else
4587                         bg_type = BTRFS_BLOCK_GROUP_METADATA;
4588                 if (!(bg_cache->flags & bg_type))
4589                         rec->wrong_chunk_type = 1;
4590         }
4591 }
4592
4593 /*
4594  * Allocate a new extent record, fill default values from @tmpl and insert int
4595  * @extent_cache. Caller is supposed to make sure the [start,nr) is not in
4596  * the cache, otherwise it fails.
4597  */
4598 static int add_extent_rec_nolookup(struct cache_tree *extent_cache,
4599                 struct extent_record *tmpl)
4600 {
4601         struct extent_record *rec;
4602         int ret = 0;
4603
4604         rec = malloc(sizeof(*rec));
4605         if (!rec)
4606                 return -ENOMEM;
4607         rec->start = tmpl->start;
4608         rec->max_size = tmpl->max_size;
4609         rec->nr = max(tmpl->nr, tmpl->max_size);
4610         rec->found_rec = tmpl->found_rec;
4611         rec->content_checked = tmpl->content_checked;
4612         rec->owner_ref_checked = tmpl->owner_ref_checked;
4613         rec->num_duplicates = 0;
4614         rec->metadata = tmpl->metadata;
4615         rec->flag_block_full_backref = FLAG_UNSET;
4616         rec->bad_full_backref = 0;
4617         rec->crossing_stripes = 0;
4618         rec->wrong_chunk_type = 0;
4619         rec->is_root = tmpl->is_root;
4620         rec->refs = tmpl->refs;
4621         rec->extent_item_refs = tmpl->extent_item_refs;
4622         rec->parent_generation = tmpl->parent_generation;
4623         INIT_LIST_HEAD(&rec->backrefs);
4624         INIT_LIST_HEAD(&rec->dups);
4625         INIT_LIST_HEAD(&rec->list);
4626         memcpy(&rec->parent_key, &tmpl->parent_key, sizeof(tmpl->parent_key));
4627         rec->cache.start = tmpl->start;
4628         rec->cache.size = tmpl->nr;
4629         ret = insert_cache_extent(extent_cache, &rec->cache);
4630         if (ret) {
4631                 free(rec);
4632                 return ret;
4633         }
4634         bytes_used += rec->nr;
4635
4636         if (tmpl->metadata)
4637                 rec->crossing_stripes = check_crossing_stripes(global_info,
4638                                 rec->start, global_info->tree_root->nodesize);
4639         check_extent_type(rec);
4640         return ret;
4641 }
4642
4643 /*
4644  * Lookup and modify an extent, some values of @tmpl are interpreted verbatim,
4645  * some are hints:
4646  * - refs              - if found, increase refs
4647  * - is_root           - if found, set
4648  * - content_checked   - if found, set
4649  * - owner_ref_checked - if found, set
4650  *
4651  * If not found, create a new one, initialize and insert.
4652  */
4653 static int add_extent_rec(struct cache_tree *extent_cache,
4654                 struct extent_record *tmpl)
4655 {
4656         struct extent_record *rec;
4657         struct cache_extent *cache;
4658         int ret = 0;
4659         int dup = 0;
4660
4661         cache = lookup_cache_extent(extent_cache, tmpl->start, tmpl->nr);
4662         if (cache) {
4663                 rec = container_of(cache, struct extent_record, cache);
4664                 if (tmpl->refs)
4665                         rec->refs++;
4666                 if (rec->nr == 1)
4667                         rec->nr = max(tmpl->nr, tmpl->max_size);
4668
4669                 /*
4670                  * We need to make sure to reset nr to whatever the extent
4671                  * record says was the real size, this way we can compare it to
4672                  * the backrefs.
4673                  */
4674                 if (tmpl->found_rec) {
4675                         if (tmpl->start != rec->start || rec->found_rec) {
4676                                 struct extent_record *tmp;
4677
4678                                 dup = 1;
4679                                 if (list_empty(&rec->list))
4680                                         list_add_tail(&rec->list,
4681                                                       &duplicate_extents);
4682
4683                                 /*
4684                                  * We have to do this song and dance in case we
4685                                  * find an extent record that falls inside of
4686                                  * our current extent record but does not have
4687                                  * the same objectid.
4688                                  */
4689                                 tmp = malloc(sizeof(*tmp));
4690                                 if (!tmp)
4691                                         return -ENOMEM;
4692                                 tmp->start = tmpl->start;
4693                                 tmp->max_size = tmpl->max_size;
4694                                 tmp->nr = tmpl->nr;
4695                                 tmp->found_rec = 1;
4696                                 tmp->metadata = tmpl->metadata;
4697                                 tmp->extent_item_refs = tmpl->extent_item_refs;
4698                                 INIT_LIST_HEAD(&tmp->list);
4699                                 list_add_tail(&tmp->list, &rec->dups);
4700                                 rec->num_duplicates++;
4701                         } else {
4702                                 rec->nr = tmpl->nr;
4703                                 rec->found_rec = 1;
4704                         }
4705                 }
4706
4707                 if (tmpl->extent_item_refs && !dup) {
4708                         if (rec->extent_item_refs) {
4709                                 fprintf(stderr, "block %llu rec "
4710                                         "extent_item_refs %llu, passed %llu\n",
4711                                         (unsigned long long)tmpl->start,
4712                                         (unsigned long long)
4713                                                         rec->extent_item_refs,
4714                                         (unsigned long long)tmpl->extent_item_refs);
4715                         }
4716                         rec->extent_item_refs = tmpl->extent_item_refs;
4717                 }
4718                 if (tmpl->is_root)
4719                         rec->is_root = 1;
4720                 if (tmpl->content_checked)
4721                         rec->content_checked = 1;
4722                 if (tmpl->owner_ref_checked)
4723                         rec->owner_ref_checked = 1;
4724                 memcpy(&rec->parent_key, &tmpl->parent_key,
4725                                 sizeof(tmpl->parent_key));
4726                 if (tmpl->parent_generation)
4727                         rec->parent_generation = tmpl->parent_generation;
4728                 if (rec->max_size < tmpl->max_size)
4729                         rec->max_size = tmpl->max_size;
4730
4731                 /*
4732                  * A metadata extent can't cross stripe_len boundary, otherwise
4733                  * kernel scrub won't be able to handle it.
4734                  * As now stripe_len is fixed to BTRFS_STRIPE_LEN, just check
4735                  * it.
4736                  */
4737                 if (tmpl->metadata)
4738                         rec->crossing_stripes = check_crossing_stripes(
4739                                         global_info, rec->start,
4740                                         global_info->tree_root->nodesize);
4741                 check_extent_type(rec);
4742                 maybe_free_extent_rec(extent_cache, rec);
4743                 return ret;
4744         }
4745
4746         ret = add_extent_rec_nolookup(extent_cache, tmpl);
4747
4748         return ret;
4749 }
4750
4751 static int add_tree_backref(struct cache_tree *extent_cache, u64 bytenr,
4752                             u64 parent, u64 root, int found_ref)
4753 {
4754         struct extent_record *rec;
4755         struct tree_backref *back;
4756         struct cache_extent *cache;
4757         int ret;
4758
4759         cache = lookup_cache_extent(extent_cache, bytenr, 1);
4760         if (!cache) {
4761                 struct extent_record tmpl;
4762
4763                 memset(&tmpl, 0, sizeof(tmpl));
4764                 tmpl.start = bytenr;
4765                 tmpl.nr = 1;
4766                 tmpl.metadata = 1;
4767
4768                 ret = add_extent_rec_nolookup(extent_cache, &tmpl);
4769                 if (ret)
4770                         return ret;
4771
4772                 /* really a bug in cache_extent implement now */
4773                 cache = lookup_cache_extent(extent_cache, bytenr, 1);
4774                 if (!cache)
4775                         return -ENOENT;
4776         }
4777
4778         rec = container_of(cache, struct extent_record, cache);
4779         if (rec->start != bytenr) {
4780                 /*
4781                  * Several cause, from unaligned bytenr to over lapping extents
4782                  */
4783                 return -EEXIST;
4784         }
4785
4786         back = find_tree_backref(rec, parent, root);
4787         if (!back) {
4788                 back = alloc_tree_backref(rec, parent, root);
4789                 if (!back)
4790                         return -ENOMEM;
4791         }
4792
4793         if (found_ref) {
4794                 if (back->node.found_ref) {
4795                         fprintf(stderr, "Extent back ref already exists "
4796                                 "for %llu parent %llu root %llu \n",
4797                                 (unsigned long long)bytenr,
4798                                 (unsigned long long)parent,
4799                                 (unsigned long long)root);
4800                 }
4801                 back->node.found_ref = 1;
4802         } else {
4803                 if (back->node.found_extent_tree) {
4804                         fprintf(stderr, "Extent back ref already exists "
4805                                 "for %llu parent %llu root %llu \n",
4806                                 (unsigned long long)bytenr,
4807                                 (unsigned long long)parent,
4808                                 (unsigned long long)root);
4809                 }
4810                 back->node.found_extent_tree = 1;
4811         }
4812         check_extent_type(rec);
4813         maybe_free_extent_rec(extent_cache, rec);
4814         return 0;
4815 }
4816
4817 static int add_data_backref(struct cache_tree *extent_cache, u64 bytenr,
4818                             u64 parent, u64 root, u64 owner, u64 offset,
4819                             u32 num_refs, int found_ref, u64 max_size)
4820 {
4821         struct extent_record *rec;
4822         struct data_backref *back;
4823         struct cache_extent *cache;
4824         int ret;
4825
4826         cache = lookup_cache_extent(extent_cache, bytenr, 1);
4827         if (!cache) {
4828                 struct extent_record tmpl;
4829
4830                 memset(&tmpl, 0, sizeof(tmpl));
4831                 tmpl.start = bytenr;
4832                 tmpl.nr = 1;
4833                 tmpl.max_size = max_size;
4834
4835                 ret = add_extent_rec_nolookup(extent_cache, &tmpl);
4836                 if (ret)
4837                         return ret;
4838
4839                 cache = lookup_cache_extent(extent_cache, bytenr, 1);
4840                 if (!cache)
4841                         abort();
4842         }
4843
4844         rec = container_of(cache, struct extent_record, cache);
4845         if (rec->max_size < max_size)
4846                 rec->max_size = max_size;
4847
4848         /*
4849          * If found_ref is set then max_size is the real size and must match the
4850          * existing refs.  So if we have already found a ref then we need to
4851          * make sure that this ref matches the existing one, otherwise we need
4852          * to add a new backref so we can notice that the backrefs don't match
4853          * and we need to figure out who is telling the truth.  This is to
4854          * account for that awful fsync bug I introduced where we'd end up with
4855          * a btrfs_file_extent_item that would have its length include multiple
4856          * prealloc extents or point inside of a prealloc extent.
4857          */
4858         back = find_data_backref(rec, parent, root, owner, offset, found_ref,
4859                                  bytenr, max_size);
4860         if (!back) {
4861                 back = alloc_data_backref(rec, parent, root, owner, offset,
4862                                           max_size);
4863                 BUG_ON(!back);
4864         }
4865
4866         if (found_ref) {
4867                 BUG_ON(num_refs != 1);
4868                 if (back->node.found_ref)
4869                         BUG_ON(back->bytes != max_size);
4870                 back->node.found_ref = 1;
4871                 back->found_ref += 1;
4872                 back->bytes = max_size;
4873                 back->disk_bytenr = bytenr;
4874                 rec->refs += 1;
4875                 rec->content_checked = 1;
4876                 rec->owner_ref_checked = 1;
4877         } else {
4878                 if (back->node.found_extent_tree) {
4879                         fprintf(stderr, "Extent back ref already exists "
4880                                 "for %llu parent %llu root %llu "
4881                                 "owner %llu offset %llu num_refs %lu\n",
4882                                 (unsigned long long)bytenr,
4883                                 (unsigned long long)parent,
4884                                 (unsigned long long)root,
4885                                 (unsigned long long)owner,
4886                                 (unsigned long long)offset,
4887                                 (unsigned long)num_refs);
4888                 }
4889                 back->num_refs = num_refs;
4890                 back->node.found_extent_tree = 1;
4891         }
4892         maybe_free_extent_rec(extent_cache, rec);
4893         return 0;
4894 }
4895
4896 static int add_pending(struct cache_tree *pending,
4897                        struct cache_tree *seen, u64 bytenr, u32 size)
4898 {
4899         int ret;
4900         ret = add_cache_extent(seen, bytenr, size);
4901         if (ret)
4902                 return ret;
4903         add_cache_extent(pending, bytenr, size);
4904         return 0;
4905 }
4906
4907 static int pick_next_pending(struct cache_tree *pending,
4908                         struct cache_tree *reada,
4909                         struct cache_tree *nodes,
4910                         u64 last, struct block_info *bits, int bits_nr,
4911                         int *reada_bits)
4912 {
4913         unsigned long node_start = last;
4914         struct cache_extent *cache;
4915         int ret;
4916
4917         cache = search_cache_extent(reada, 0);
4918         if (cache) {
4919                 bits[0].start = cache->start;
4920                 bits[0].size = cache->size;
4921                 *reada_bits = 1;
4922                 return 1;
4923         }
4924         *reada_bits = 0;
4925         if (node_start > 32768)
4926                 node_start -= 32768;
4927
4928         cache = search_cache_extent(nodes, node_start);
4929         if (!cache)
4930                 cache = search_cache_extent(nodes, 0);
4931
4932         if (!cache) {
4933                  cache = search_cache_extent(pending, 0);
4934                  if (!cache)
4935                          return 0;
4936                  ret = 0;
4937                  do {
4938                          bits[ret].start = cache->start;
4939                          bits[ret].size = cache->size;
4940                          cache = next_cache_extent(cache);
4941                          ret++;
4942                  } while (cache && ret < bits_nr);
4943                  return ret;
4944         }
4945
4946         ret = 0;
4947         do {
4948                 bits[ret].start = cache->start;
4949                 bits[ret].size = cache->size;
4950                 cache = next_cache_extent(cache);
4951                 ret++;
4952         } while (cache && ret < bits_nr);
4953
4954         if (bits_nr - ret > 8) {
4955                 u64 lookup = bits[0].start + bits[0].size;
4956                 struct cache_extent *next;
4957                 next = search_cache_extent(pending, lookup);
4958                 while(next) {
4959                         if (next->start - lookup > 32768)
4960                                 break;
4961                         bits[ret].start = next->start;
4962                         bits[ret].size = next->size;
4963                         lookup = next->start + next->size;
4964                         ret++;
4965                         if (ret == bits_nr)
4966                                 break;
4967                         next = next_cache_extent(next);
4968                         if (!next)
4969                                 break;
4970                 }
4971         }
4972         return ret;
4973 }
4974
4975 static void free_chunk_record(struct cache_extent *cache)
4976 {
4977         struct chunk_record *rec;
4978
4979         rec = container_of(cache, struct chunk_record, cache);
4980         list_del_init(&rec->list);
4981         list_del_init(&rec->dextents);
4982         free(rec);
4983 }
4984
4985 void free_chunk_cache_tree(struct cache_tree *chunk_cache)
4986 {
4987         cache_tree_free_extents(chunk_cache, free_chunk_record);
4988 }
4989
4990 static void free_device_record(struct rb_node *node)
4991 {
4992         struct device_record *rec;
4993
4994         rec = container_of(node, struct device_record, node);
4995         free(rec);
4996 }
4997
4998 FREE_RB_BASED_TREE(device_cache, free_device_record);
4999
5000 int insert_block_group_record(struct block_group_tree *tree,
5001                               struct block_group_record *bg_rec)
5002 {
5003         int ret;
5004
5005         ret = insert_cache_extent(&tree->tree, &bg_rec->cache);
5006         if (ret)
5007                 return ret;
5008
5009         list_add_tail(&bg_rec->list, &tree->block_groups);
5010         return 0;
5011 }
5012
5013 static void free_block_group_record(struct cache_extent *cache)
5014 {
5015         struct block_group_record *rec;
5016
5017         rec = container_of(cache, struct block_group_record, cache);
5018         list_del_init(&rec->list);
5019         free(rec);
5020 }
5021
5022 void free_block_group_tree(struct block_group_tree *tree)
5023 {
5024         cache_tree_free_extents(&tree->tree, free_block_group_record);
5025 }
5026
5027 int insert_device_extent_record(struct device_extent_tree *tree,
5028                                 struct device_extent_record *de_rec)
5029 {
5030         int ret;
5031
5032         /*
5033          * Device extent is a bit different from the other extents, because
5034          * the extents which belong to the different devices may have the
5035          * same start and size, so we need use the special extent cache
5036          * search/insert functions.
5037          */
5038         ret = insert_cache_extent2(&tree->tree, &de_rec->cache);
5039         if (ret)
5040                 return ret;
5041
5042         list_add_tail(&de_rec->chunk_list, &tree->no_chunk_orphans);
5043         list_add_tail(&de_rec->device_list, &tree->no_device_orphans);
5044         return 0;
5045 }
5046
5047 static void free_device_extent_record(struct cache_extent *cache)
5048 {
5049         struct device_extent_record *rec;
5050
5051         rec = container_of(cache, struct device_extent_record, cache);
5052         if (!list_empty(&rec->chunk_list))
5053                 list_del_init(&rec->chunk_list);
5054         if (!list_empty(&rec->device_list))
5055                 list_del_init(&rec->device_list);
5056         free(rec);
5057 }
5058
5059 void free_device_extent_tree(struct device_extent_tree *tree)
5060 {
5061         cache_tree_free_extents(&tree->tree, free_device_extent_record);
5062 }
5063
5064 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5065 static int process_extent_ref_v0(struct cache_tree *extent_cache,
5066                                  struct extent_buffer *leaf, int slot)
5067 {
5068         struct btrfs_extent_ref_v0 *ref0;
5069         struct btrfs_key key;
5070         int ret;
5071
5072         btrfs_item_key_to_cpu(leaf, &key, slot);
5073         ref0 = btrfs_item_ptr(leaf, slot, struct btrfs_extent_ref_v0);
5074         if (btrfs_ref_objectid_v0(leaf, ref0) < BTRFS_FIRST_FREE_OBJECTID) {
5075                 ret = add_tree_backref(extent_cache, key.objectid, key.offset,
5076                                 0, 0);
5077         } else {
5078                 ret = add_data_backref(extent_cache, key.objectid, key.offset,
5079                                 0, 0, 0, btrfs_ref_count_v0(leaf, ref0), 0, 0);
5080         }
5081         return ret;
5082 }
5083 #endif
5084
5085 struct chunk_record *btrfs_new_chunk_record(struct extent_buffer *leaf,
5086                                             struct btrfs_key *key,
5087                                             int slot)
5088 {
5089         struct btrfs_chunk *ptr;
5090         struct chunk_record *rec;
5091         int num_stripes, i;
5092
5093         ptr = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
5094         num_stripes = btrfs_chunk_num_stripes(leaf, ptr);
5095
5096         rec = calloc(1, btrfs_chunk_record_size(num_stripes));
5097         if (!rec) {
5098                 fprintf(stderr, "memory allocation failed\n");
5099                 exit(-1);
5100         }
5101
5102         INIT_LIST_HEAD(&rec->list);
5103         INIT_LIST_HEAD(&rec->dextents);
5104         rec->bg_rec = NULL;
5105
5106         rec->cache.start = key->offset;
5107         rec->cache.size = btrfs_chunk_length(leaf, ptr);
5108
5109         rec->generation = btrfs_header_generation(leaf);
5110
5111         rec->objectid = key->objectid;
5112         rec->type = key->type;
5113         rec->offset = key->offset;
5114
5115         rec->length = rec->cache.size;
5116         rec->owner = btrfs_chunk_owner(leaf, ptr);
5117         rec->stripe_len = btrfs_chunk_stripe_len(leaf, ptr);
5118         rec->type_flags = btrfs_chunk_type(leaf, ptr);
5119         rec->io_width = btrfs_chunk_io_width(leaf, ptr);
5120         rec->io_align = btrfs_chunk_io_align(leaf, ptr);
5121         rec->sector_size = btrfs_chunk_sector_size(leaf, ptr);
5122         rec->num_stripes = num_stripes;
5123         rec->sub_stripes = btrfs_chunk_sub_stripes(leaf, ptr);
5124
5125         for (i = 0; i < rec->num_stripes; ++i) {
5126                 rec->stripes[i].devid =
5127                         btrfs_stripe_devid_nr(leaf, ptr, i);
5128                 rec->stripes[i].offset =
5129                         btrfs_stripe_offset_nr(leaf, ptr, i);
5130                 read_extent_buffer(leaf, rec->stripes[i].dev_uuid,
5131                                 (unsigned long)btrfs_stripe_dev_uuid_nr(ptr, i),
5132                                 BTRFS_UUID_SIZE);
5133         }
5134
5135         return rec;
5136 }
5137
5138 static int process_chunk_item(struct cache_tree *chunk_cache,
5139                               struct btrfs_key *key, struct extent_buffer *eb,
5140                               int slot)
5141 {
5142         struct chunk_record *rec;
5143         struct btrfs_chunk *chunk;
5144         int ret = 0;
5145
5146         chunk = btrfs_item_ptr(eb, slot, struct btrfs_chunk);
5147         /*
5148          * Do extra check for this chunk item,
5149          *
5150          * It's still possible one can craft a leaf with CHUNK_ITEM, with
5151          * wrong onwer(3) out of chunk tree, to pass both chunk tree check
5152          * and owner<->key_type check.
5153          */
5154         ret = btrfs_check_chunk_valid(global_info->tree_root, eb, chunk, slot,
5155                                       key->offset);
5156         if (ret < 0) {
5157                 error("chunk(%llu, %llu) is not valid, ignore it",
5158                       key->offset, btrfs_chunk_length(eb, chunk));
5159                 return 0;
5160         }
5161         rec = btrfs_new_chunk_record(eb, key, slot);
5162         ret = insert_cache_extent(chunk_cache, &rec->cache);
5163         if (ret) {
5164                 fprintf(stderr, "Chunk[%llu, %llu] existed.\n",
5165                         rec->offset, rec->length);
5166                 free(rec);
5167         }
5168
5169         return ret;
5170 }
5171
5172 static int process_device_item(struct rb_root *dev_cache,
5173                 struct btrfs_key *key, struct extent_buffer *eb, int slot)
5174 {
5175         struct btrfs_dev_item *ptr;
5176         struct device_record *rec;
5177         int ret = 0;
5178
5179         ptr = btrfs_item_ptr(eb,
5180                 slot, struct btrfs_dev_item);
5181
5182         rec = malloc(sizeof(*rec));
5183         if (!rec) {
5184                 fprintf(stderr, "memory allocation failed\n");
5185                 return -ENOMEM;
5186         }
5187
5188         rec->devid = key->offset;
5189         rec->generation = btrfs_header_generation(eb);
5190
5191         rec->objectid = key->objectid;
5192         rec->type = key->type;
5193         rec->offset = key->offset;
5194
5195         rec->devid = btrfs_device_id(eb, ptr);
5196         rec->total_byte = btrfs_device_total_bytes(eb, ptr);
5197         rec->byte_used = btrfs_device_bytes_used(eb, ptr);
5198
5199         ret = rb_insert(dev_cache, &rec->node, device_record_compare);
5200         if (ret) {
5201                 fprintf(stderr, "Device[%llu] existed.\n", rec->devid);
5202                 free(rec);
5203         }
5204
5205         return ret;
5206 }
5207
5208 struct block_group_record *
5209 btrfs_new_block_group_record(struct extent_buffer *leaf, struct btrfs_key *key,
5210                              int slot)
5211 {
5212         struct btrfs_block_group_item *ptr;
5213         struct block_group_record *rec;
5214
5215         rec = calloc(1, sizeof(*rec));
5216         if (!rec) {
5217                 fprintf(stderr, "memory allocation failed\n");
5218                 exit(-1);
5219         }
5220
5221         rec->cache.start = key->objectid;
5222         rec->cache.size = key->offset;
5223
5224         rec->generation = btrfs_header_generation(leaf);
5225
5226         rec->objectid = key->objectid;
5227         rec->type = key->type;
5228         rec->offset = key->offset;
5229
5230         ptr = btrfs_item_ptr(leaf, slot, struct btrfs_block_group_item);
5231         rec->flags = btrfs_disk_block_group_flags(leaf, ptr);
5232
5233         INIT_LIST_HEAD(&rec->list);
5234
5235         return rec;
5236 }
5237
5238 static int process_block_group_item(struct block_group_tree *block_group_cache,
5239                                     struct btrfs_key *key,
5240                                     struct extent_buffer *eb, int slot)
5241 {
5242         struct block_group_record *rec;
5243         int ret = 0;
5244
5245         rec = btrfs_new_block_group_record(eb, key, slot);
5246         ret = insert_block_group_record(block_group_cache, rec);
5247         if (ret) {
5248                 fprintf(stderr, "Block Group[%llu, %llu] existed.\n",
5249                         rec->objectid, rec->offset);
5250                 free(rec);
5251         }
5252
5253         return ret;
5254 }
5255
5256 struct device_extent_record *
5257 btrfs_new_device_extent_record(struct extent_buffer *leaf,
5258                                struct btrfs_key *key, int slot)
5259 {
5260         struct device_extent_record *rec;
5261         struct btrfs_dev_extent *ptr;
5262
5263         rec = calloc(1, sizeof(*rec));
5264         if (!rec) {
5265                 fprintf(stderr, "memory allocation failed\n");
5266                 exit(-1);
5267         }
5268
5269         rec->cache.objectid = key->objectid;
5270         rec->cache.start = key->offset;
5271
5272         rec->generation = btrfs_header_generation(leaf);
5273
5274         rec->objectid = key->objectid;
5275         rec->type = key->type;
5276         rec->offset = key->offset;
5277
5278         ptr = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
5279         rec->chunk_objecteid =
5280                 btrfs_dev_extent_chunk_objectid(leaf, ptr);
5281         rec->chunk_offset =
5282                 btrfs_dev_extent_chunk_offset(leaf, ptr);
5283         rec->length = btrfs_dev_extent_length(leaf, ptr);
5284         rec->cache.size = rec->length;
5285
5286         INIT_LIST_HEAD(&rec->chunk_list);
5287         INIT_LIST_HEAD(&rec->device_list);
5288
5289         return rec;
5290 }
5291
5292 static int
5293 process_device_extent_item(struct device_extent_tree *dev_extent_cache,
5294                            struct btrfs_key *key, struct extent_buffer *eb,
5295                            int slot)
5296 {
5297         struct device_extent_record *rec;
5298         int ret;
5299
5300         rec = btrfs_new_device_extent_record(eb, key, slot);
5301         ret = insert_device_extent_record(dev_extent_cache, rec);
5302         if (ret) {
5303                 fprintf(stderr,
5304                         "Device extent[%llu, %llu, %llu] existed.\n",
5305                         rec->objectid, rec->offset, rec->length);
5306                 free(rec);
5307         }
5308
5309         return ret;
5310 }
5311
5312 static int process_extent_item(struct btrfs_root *root,
5313                                struct cache_tree *extent_cache,
5314                                struct extent_buffer *eb, int slot)
5315 {
5316         struct btrfs_extent_item *ei;
5317         struct btrfs_extent_inline_ref *iref;
5318         struct btrfs_extent_data_ref *dref;
5319         struct btrfs_shared_data_ref *sref;
5320         struct btrfs_key key;
5321         struct extent_record tmpl;
5322         unsigned long end;
5323         unsigned long ptr;
5324         int ret;
5325         int type;
5326         u32 item_size = btrfs_item_size_nr(eb, slot);
5327         u64 refs = 0;
5328         u64 offset;
5329         u64 num_bytes;
5330         int metadata = 0;
5331
5332         btrfs_item_key_to_cpu(eb, &key, slot);
5333
5334         if (key.type == BTRFS_METADATA_ITEM_KEY) {
5335                 metadata = 1;
5336                 num_bytes = root->nodesize;
5337         } else {
5338                 num_bytes = key.offset;
5339         }
5340
5341         if (!IS_ALIGNED(key.objectid, root->sectorsize)) {
5342                 error("ignoring invalid extent, bytenr %llu is not aligned to %u",
5343                       key.objectid, root->sectorsize);
5344                 return -EIO;
5345         }
5346         if (item_size < sizeof(*ei)) {
5347 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5348                 struct btrfs_extent_item_v0 *ei0;
5349                 BUG_ON(item_size != sizeof(*ei0));
5350                 ei0 = btrfs_item_ptr(eb, slot, struct btrfs_extent_item_v0);
5351                 refs = btrfs_extent_refs_v0(eb, ei0);
5352 #else
5353                 BUG();
5354 #endif
5355                 memset(&tmpl, 0, sizeof(tmpl));
5356                 tmpl.start = key.objectid;
5357                 tmpl.nr = num_bytes;
5358                 tmpl.extent_item_refs = refs;
5359                 tmpl.metadata = metadata;
5360                 tmpl.found_rec = 1;
5361                 tmpl.max_size = num_bytes;
5362
5363                 return add_extent_rec(extent_cache, &tmpl);
5364         }
5365
5366         ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
5367         refs = btrfs_extent_refs(eb, ei);
5368         if (btrfs_extent_flags(eb, ei) & BTRFS_EXTENT_FLAG_TREE_BLOCK)
5369                 metadata = 1;
5370         else
5371                 metadata = 0;
5372         if (metadata && num_bytes != root->nodesize) {
5373                 error("ignore invalid metadata extent, length %llu does not equal to %u",
5374                       num_bytes, root->nodesize);
5375                 return -EIO;
5376         }
5377         if (!metadata && !IS_ALIGNED(num_bytes, root->sectorsize)) {
5378                 error("ignore invalid data extent, length %llu is not aligned to %u",
5379                       num_bytes, root->sectorsize);
5380                 return -EIO;
5381         }
5382
5383         memset(&tmpl, 0, sizeof(tmpl));
5384         tmpl.start = key.objectid;
5385         tmpl.nr = num_bytes;
5386         tmpl.extent_item_refs = refs;
5387         tmpl.metadata = metadata;
5388         tmpl.found_rec = 1;
5389         tmpl.max_size = num_bytes;
5390         add_extent_rec(extent_cache, &tmpl);
5391
5392         ptr = (unsigned long)(ei + 1);
5393         if (btrfs_extent_flags(eb, ei) & BTRFS_EXTENT_FLAG_TREE_BLOCK &&
5394             key.type == BTRFS_EXTENT_ITEM_KEY)
5395                 ptr += sizeof(struct btrfs_tree_block_info);
5396
5397         end = (unsigned long)ei + item_size;
5398         while (ptr < end) {
5399                 iref = (struct btrfs_extent_inline_ref *)ptr;
5400                 type = btrfs_extent_inline_ref_type(eb, iref);
5401                 offset = btrfs_extent_inline_ref_offset(eb, iref);
5402                 switch (type) {
5403                 case BTRFS_TREE_BLOCK_REF_KEY:
5404                         ret = add_tree_backref(extent_cache, key.objectid,
5405                                         0, offset, 0);
5406                         if (ret < 0)
5407                                 error("add_tree_backref failed: %s",
5408                                       strerror(-ret));
5409                         break;
5410                 case BTRFS_SHARED_BLOCK_REF_KEY:
5411                         ret = add_tree_backref(extent_cache, key.objectid,
5412                                         offset, 0, 0);
5413                         if (ret < 0)
5414                                 error("add_tree_backref failed: %s",
5415                                       strerror(-ret));
5416                         break;
5417                 case BTRFS_EXTENT_DATA_REF_KEY:
5418                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
5419                         add_data_backref(extent_cache, key.objectid, 0,
5420                                         btrfs_extent_data_ref_root(eb, dref),
5421                                         btrfs_extent_data_ref_objectid(eb,
5422                                                                        dref),
5423                                         btrfs_extent_data_ref_offset(eb, dref),
5424                                         btrfs_extent_data_ref_count(eb, dref),
5425                                         0, num_bytes);
5426                         break;
5427                 case BTRFS_SHARED_DATA_REF_KEY:
5428                         sref = (struct btrfs_shared_data_ref *)(iref + 1);
5429                         add_data_backref(extent_cache, key.objectid, offset,
5430                                         0, 0, 0,
5431                                         btrfs_shared_data_ref_count(eb, sref),
5432                                         0, num_bytes);
5433                         break;
5434                 default:
5435                         fprintf(stderr, "corrupt extent record: key %Lu %u %Lu\n",
5436                                 key.objectid, key.type, num_bytes);
5437                         goto out;
5438                 }
5439                 ptr += btrfs_extent_inline_ref_size(type);
5440         }
5441         WARN_ON(ptr > end);
5442 out:
5443         return 0;
5444 }
5445
5446 static int check_cache_range(struct btrfs_root *root,
5447                              struct btrfs_block_group_cache *cache,
5448                              u64 offset, u64 bytes)
5449 {
5450         struct btrfs_free_space *entry;
5451         u64 *logical;
5452         u64 bytenr;
5453         int stripe_len;
5454         int i, nr, ret;
5455
5456         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
5457                 bytenr = btrfs_sb_offset(i);
5458                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
5459                                        cache->key.objectid, bytenr, 0,
5460                                        &logical, &nr, &stripe_len);
5461                 if (ret)
5462                         return ret;
5463
5464                 while (nr--) {
5465                         if (logical[nr] + stripe_len <= offset)
5466                                 continue;
5467                         if (offset + bytes <= logical[nr])
5468                                 continue;
5469                         if (logical[nr] == offset) {
5470                                 if (stripe_len >= bytes) {
5471                                         free(logical);
5472                                         return 0;
5473                                 }
5474                                 bytes -= stripe_len;
5475                                 offset += stripe_len;
5476                         } else if (logical[nr] < offset) {
5477                                 if (logical[nr] + stripe_len >=
5478                                     offset + bytes) {
5479                                         free(logical);
5480                                         return 0;
5481                                 }
5482                                 bytes = (offset + bytes) -
5483                                         (logical[nr] + stripe_len);
5484                                 offset = logical[nr] + stripe_len;
5485                         } else {
5486                                 /*
5487                                  * Could be tricky, the super may land in the
5488                                  * middle of the area we're checking.  First
5489                                  * check the easiest case, it's at the end.
5490                                  */
5491                                 if (logical[nr] + stripe_len >=
5492                                     bytes + offset) {
5493                                         bytes = logical[nr] - offset;
5494                                         continue;
5495                                 }
5496
5497                                 /* Check the left side */
5498                                 ret = check_cache_range(root, cache,
5499                                                         offset,
5500                                                         logical[nr] - offset);
5501                                 if (ret) {
5502                                         free(logical);
5503                                         return ret;
5504                                 }
5505
5506                                 /* Now we continue with the right side */
5507                                 bytes = (offset + bytes) -
5508                                         (logical[nr] + stripe_len);
5509                                 offset = logical[nr] + stripe_len;
5510                         }
5511                 }
5512
5513                 free(logical);
5514         }
5515
5516         entry = btrfs_find_free_space(cache->free_space_ctl, offset, bytes);
5517         if (!entry) {
5518                 fprintf(stderr, "There is no free space entry for %Lu-%Lu\n",
5519                         offset, offset+bytes);
5520                 return -EINVAL;
5521         }
5522
5523         if (entry->offset != offset) {
5524                 fprintf(stderr, "Wanted offset %Lu, found %Lu\n", offset,
5525                         entry->offset);
5526                 return -EINVAL;
5527         }
5528
5529         if (entry->bytes != bytes) {
5530                 fprintf(stderr, "Wanted bytes %Lu, found %Lu for off %Lu\n",
5531                         bytes, entry->bytes, offset);
5532                 return -EINVAL;
5533         }
5534
5535         unlink_free_space(cache->free_space_ctl, entry);
5536         free(entry);
5537         return 0;
5538 }
5539
5540 static int verify_space_cache(struct btrfs_root *root,
5541                               struct btrfs_block_group_cache *cache)
5542 {
5543         struct btrfs_path path;
5544         struct extent_buffer *leaf;
5545         struct btrfs_key key;
5546         u64 last;
5547         int ret = 0;
5548
5549         root = root->fs_info->extent_root;
5550
5551         last = max_t(u64, cache->key.objectid, BTRFS_SUPER_INFO_OFFSET);
5552
5553         btrfs_init_path(&path);
5554         key.objectid = last;
5555         key.offset = 0;
5556         key.type = BTRFS_EXTENT_ITEM_KEY;
5557         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
5558         if (ret < 0)
5559                 goto out;
5560         ret = 0;
5561         while (1) {
5562                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
5563                         ret = btrfs_next_leaf(root, &path);
5564                         if (ret < 0)
5565                                 goto out;
5566                         if (ret > 0) {
5567                                 ret = 0;
5568                                 break;
5569                         }
5570                 }
5571                 leaf = path.nodes[0];
5572                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
5573                 if (key.objectid >= cache->key.offset + cache->key.objectid)
5574                         break;
5575                 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
5576                     key.type != BTRFS_METADATA_ITEM_KEY) {
5577                         path.slots[0]++;
5578                         continue;
5579                 }
5580
5581                 if (last == key.objectid) {
5582                         if (key.type == BTRFS_EXTENT_ITEM_KEY)
5583                                 last = key.objectid + key.offset;
5584                         else
5585                                 last = key.objectid + root->nodesize;
5586                         path.slots[0]++;
5587                         continue;
5588                 }
5589
5590                 ret = check_cache_range(root, cache, last,
5591                                         key.objectid - last);
5592                 if (ret)
5593                         break;
5594                 if (key.type == BTRFS_EXTENT_ITEM_KEY)
5595                         last = key.objectid + key.offset;
5596                 else
5597                         last = key.objectid + root->nodesize;
5598                 path.slots[0]++;
5599         }
5600
5601         if (last < cache->key.objectid + cache->key.offset)
5602                 ret = check_cache_range(root, cache, last,
5603                                         cache->key.objectid +
5604                                         cache->key.offset - last);
5605
5606 out:
5607         btrfs_release_path(&path);
5608
5609         if (!ret &&
5610             !RB_EMPTY_ROOT(&cache->free_space_ctl->free_space_offset)) {
5611                 fprintf(stderr, "There are still entries left in the space "
5612                         "cache\n");
5613                 ret = -EINVAL;
5614         }
5615
5616         return ret;
5617 }
5618
5619 static int check_space_cache(struct btrfs_root *root)
5620 {
5621         struct btrfs_block_group_cache *cache;
5622         u64 start = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
5623         int ret;
5624         int error = 0;
5625
5626         if (btrfs_super_cache_generation(root->fs_info->super_copy) != -1ULL &&
5627             btrfs_super_generation(root->fs_info->super_copy) !=
5628             btrfs_super_cache_generation(root->fs_info->super_copy)) {
5629                 printf("cache and super generation don't match, space cache "
5630                        "will be invalidated\n");
5631                 return 0;
5632         }
5633
5634         if (ctx.progress_enabled) {
5635                 ctx.tp = TASK_FREE_SPACE;
5636                 task_start(ctx.info);
5637         }
5638
5639         while (1) {
5640                 cache = btrfs_lookup_first_block_group(root->fs_info, start);
5641                 if (!cache)
5642                         break;
5643
5644                 start = cache->key.objectid + cache->key.offset;
5645                 if (!cache->free_space_ctl) {
5646                         if (btrfs_init_free_space_ctl(cache,
5647                                                       root->sectorsize)) {
5648                                 ret = -ENOMEM;
5649                                 break;
5650                         }
5651                 } else {
5652                         btrfs_remove_free_space_cache(cache);
5653                 }
5654
5655                 if (btrfs_fs_compat_ro(root->fs_info,
5656                                        BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE)) {
5657                         ret = exclude_super_stripes(root, cache);
5658                         if (ret) {
5659                                 fprintf(stderr, "could not exclude super stripes: %s\n",
5660                                         strerror(-ret));
5661                                 error++;
5662                                 continue;
5663                         }
5664                         ret = load_free_space_tree(root->fs_info, cache);
5665                         free_excluded_extents(root, cache);
5666                         if (ret < 0) {
5667                                 fprintf(stderr, "could not load free space tree: %s\n",
5668                                         strerror(-ret));
5669                                 error++;
5670                                 continue;
5671                         }
5672                         error += ret;
5673                 } else {
5674                         ret = load_free_space_cache(root->fs_info, cache);
5675                         if (!ret)
5676                                 continue;
5677                 }
5678
5679                 ret = verify_space_cache(root, cache);
5680                 if (ret) {
5681                         fprintf(stderr, "cache appears valid but isn't %Lu\n",
5682                                 cache->key.objectid);
5683                         error++;
5684                 }
5685         }
5686
5687         task_stop(ctx.info);
5688
5689         return error ? -EINVAL : 0;
5690 }
5691
5692 static int check_extent_csums(struct btrfs_root *root, u64 bytenr,
5693                         u64 num_bytes, unsigned long leaf_offset,
5694                         struct extent_buffer *eb) {
5695
5696         u64 offset = 0;
5697         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
5698         char *data;
5699         unsigned long csum_offset;
5700         u32 csum;
5701         u32 csum_expected;
5702         u64 read_len;
5703         u64 data_checked = 0;
5704         u64 tmp;
5705         int ret = 0;
5706         int mirror;
5707         int num_copies;
5708
5709         if (num_bytes % root->sectorsize)
5710                 return -EINVAL;
5711
5712         data = malloc(num_bytes);
5713         if (!data)
5714                 return -ENOMEM;
5715
5716         while (offset < num_bytes) {
5717                 mirror = 0;
5718 again:
5719                 read_len = num_bytes - offset;
5720                 /* read as much space once a time */
5721                 ret = read_extent_data(root, data + offset,
5722                                 bytenr + offset, &read_len, mirror);
5723                 if (ret)
5724                         goto out;
5725                 data_checked = 0;
5726                 /* verify every 4k data's checksum */
5727                 while (data_checked < read_len) {
5728                         csum = ~(u32)0;
5729                         tmp = offset + data_checked;
5730
5731                         csum = btrfs_csum_data(NULL, (char *)data + tmp,
5732                                                csum, root->sectorsize);
5733                         btrfs_csum_final(csum, (u8 *)&csum);
5734
5735                         csum_offset = leaf_offset +
5736                                  tmp / root->sectorsize * csum_size;
5737                         read_extent_buffer(eb, (char *)&csum_expected,
5738                                            csum_offset, csum_size);
5739                         /* try another mirror */
5740                         if (csum != csum_expected) {
5741                                 fprintf(stderr, "mirror %d bytenr %llu csum %u expected csum %u\n",
5742                                                 mirror, bytenr + tmp,
5743                                                 csum, csum_expected);
5744                                 num_copies = btrfs_num_copies(
5745                                                 &root->fs_info->mapping_tree,
5746                                                 bytenr, num_bytes);
5747                                 if (mirror < num_copies - 1) {
5748                                         mirror += 1;
5749                                         goto again;
5750                                 }
5751                         }
5752                         data_checked += root->sectorsize;
5753                 }
5754                 offset += read_len;
5755         }
5756 out:
5757         free(data);
5758         return ret;
5759 }
5760
5761 static int check_extent_exists(struct btrfs_root *root, u64 bytenr,
5762                                u64 num_bytes)
5763 {
5764         struct btrfs_path path;
5765         struct extent_buffer *leaf;
5766         struct btrfs_key key;
5767         int ret;
5768
5769         btrfs_init_path(&path);
5770         key.objectid = bytenr;
5771         key.type = BTRFS_EXTENT_ITEM_KEY;
5772         key.offset = (u64)-1;
5773
5774 again:
5775         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, &path,
5776                                 0, 0);
5777         if (ret < 0) {
5778                 fprintf(stderr, "Error looking up extent record %d\n", ret);
5779                 btrfs_release_path(&path);
5780                 return ret;
5781         } else if (ret) {
5782                 if (path.slots[0] > 0) {
5783                         path.slots[0]--;
5784                 } else {
5785                         ret = btrfs_prev_leaf(root, &path);
5786                         if (ret < 0) {
5787                                 goto out;
5788                         } else if (ret > 0) {
5789                                 ret = 0;
5790                                 goto out;
5791                         }
5792                 }
5793         }
5794
5795         btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
5796
5797         /*
5798          * Block group items come before extent items if they have the same
5799          * bytenr, so walk back one more just in case.  Dear future traveller,
5800          * first congrats on mastering time travel.  Now if it's not too much
5801          * trouble could you go back to 2006 and tell Chris to make the
5802          * BLOCK_GROUP_ITEM_KEY (and BTRFS_*_REF_KEY) lower than the
5803          * EXTENT_ITEM_KEY please?
5804          */
5805         while (key.type > BTRFS_EXTENT_ITEM_KEY) {
5806                 if (path.slots[0] > 0) {
5807                         path.slots[0]--;
5808                 } else {
5809                         ret = btrfs_prev_leaf(root, &path);
5810                         if (ret < 0) {
5811                                 goto out;
5812                         } else if (ret > 0) {
5813                                 ret = 0;
5814                                 goto out;
5815                         }
5816                 }
5817                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
5818         }
5819
5820         while (num_bytes) {
5821                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
5822                         ret = btrfs_next_leaf(root, &path);
5823                         if (ret < 0) {
5824                                 fprintf(stderr, "Error going to next leaf "
5825                                         "%d\n", ret);
5826                                 btrfs_release_path(&path);
5827                                 return ret;
5828                         } else if (ret) {
5829                                 break;
5830                         }
5831                 }
5832                 leaf = path.nodes[0];
5833                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
5834                 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
5835                         path.slots[0]++;
5836                         continue;
5837                 }
5838                 if (key.objectid + key.offset < bytenr) {
5839                         path.slots[0]++;
5840                         continue;
5841                 }
5842                 if (key.objectid > bytenr + num_bytes)
5843                         break;
5844
5845                 if (key.objectid == bytenr) {
5846                         if (key.offset >= num_bytes) {
5847                                 num_bytes = 0;
5848                                 break;
5849                         }
5850                         num_bytes -= key.offset;
5851                         bytenr += key.offset;
5852                 } else if (key.objectid < bytenr) {
5853                         if (key.objectid + key.offset >= bytenr + num_bytes) {
5854                                 num_bytes = 0;
5855                                 break;
5856                         }
5857                         num_bytes = (bytenr + num_bytes) -
5858                                 (key.objectid + key.offset);
5859                         bytenr = key.objectid + key.offset;
5860                 } else {
5861                         if (key.objectid + key.offset < bytenr + num_bytes) {
5862                                 u64 new_start = key.objectid + key.offset;
5863                                 u64 new_bytes = bytenr + num_bytes - new_start;
5864
5865                                 /*
5866                                  * Weird case, the extent is in the middle of
5867                                  * our range, we'll have to search one side
5868                                  * and then the other.  Not sure if this happens
5869                                  * in real life, but no harm in coding it up
5870                                  * anyway just in case.
5871                                  */
5872                                 btrfs_release_path(&path);
5873                                 ret = check_extent_exists(root, new_start,
5874                                                           new_bytes);
5875                                 if (ret) {
5876                                         fprintf(stderr, "Right section didn't "
5877                                                 "have a record\n");
5878                                         break;
5879                                 }
5880                                 num_bytes = key.objectid - bytenr;
5881                                 goto again;
5882                         }
5883                         num_bytes = key.objectid - bytenr;
5884                 }
5885                 path.slots[0]++;
5886         }
5887         ret = 0;
5888
5889 out:
5890         if (num_bytes && !ret) {
5891                 fprintf(stderr, "There are no extents for csum range "
5892                         "%Lu-%Lu\n", bytenr, bytenr+num_bytes);
5893                 ret = 1;
5894         }
5895
5896         btrfs_release_path(&path);
5897         return ret;
5898 }
5899
5900 static int check_csums(struct btrfs_root *root)
5901 {
5902         struct btrfs_path path;
5903         struct extent_buffer *leaf;
5904         struct btrfs_key key;
5905         u64 offset = 0, num_bytes = 0;
5906         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
5907         int errors = 0;
5908         int ret;
5909         u64 data_len;
5910         unsigned long leaf_offset;
5911
5912         root = root->fs_info->csum_root;
5913         if (!extent_buffer_uptodate(root->node)) {
5914                 fprintf(stderr, "No valid csum tree found\n");
5915                 return -ENOENT;
5916         }
5917
5918         btrfs_init_path(&path);
5919         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
5920         key.type = BTRFS_EXTENT_CSUM_KEY;
5921         key.offset = 0;
5922         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
5923         if (ret < 0) {
5924                 fprintf(stderr, "Error searching csum tree %d\n", ret);
5925                 btrfs_release_path(&path);
5926                 return ret;
5927         }
5928
5929         if (ret > 0 && path.slots[0])
5930                 path.slots[0]--;
5931         ret = 0;
5932
5933         while (1) {
5934                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
5935                         ret = btrfs_next_leaf(root, &path);
5936                         if (ret < 0) {
5937                                 fprintf(stderr, "Error going to next leaf "
5938                                         "%d\n", ret);
5939                                 break;
5940                         }
5941                         if (ret)
5942                                 break;
5943                 }
5944                 leaf = path.nodes[0];
5945
5946                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
5947                 if (key.type != BTRFS_EXTENT_CSUM_KEY) {
5948                         path.slots[0]++;
5949                         continue;
5950                 }
5951
5952                 data_len = (btrfs_item_size_nr(leaf, path.slots[0]) /
5953                               csum_size) * root->sectorsize;
5954                 if (!check_data_csum)
5955                         goto skip_csum_check;
5956                 leaf_offset = btrfs_item_ptr_offset(leaf, path.slots[0]);
5957                 ret = check_extent_csums(root, key.offset, data_len,
5958                                          leaf_offset, leaf);
5959                 if (ret)
5960                         break;
5961 skip_csum_check:
5962                 if (!num_bytes) {
5963                         offset = key.offset;
5964                 } else if (key.offset != offset + num_bytes) {
5965                         ret = check_extent_exists(root, offset, num_bytes);
5966                         if (ret) {
5967                                 fprintf(stderr, "Csum exists for %Lu-%Lu but "
5968                                         "there is no extent record\n",
5969                                         offset, offset+num_bytes);
5970                                 errors++;
5971                         }
5972                         offset = key.offset;
5973                         num_bytes = 0;
5974                 }
5975                 num_bytes += data_len;
5976                 path.slots[0]++;
5977         }
5978
5979         btrfs_release_path(&path);
5980         return errors;
5981 }
5982
5983 static int is_dropped_key(struct btrfs_key *key,
5984                           struct btrfs_key *drop_key) {
5985         if (key->objectid < drop_key->objectid)
5986                 return 1;
5987         else if (key->objectid == drop_key->objectid) {
5988                 if (key->type < drop_key->type)
5989                         return 1;
5990                 else if (key->type == drop_key->type) {
5991                         if (key->offset < drop_key->offset)
5992                                 return 1;
5993                 }
5994         }
5995         return 0;
5996 }
5997
5998 /*
5999  * Here are the rules for FULL_BACKREF.
6000  *
6001  * 1) If BTRFS_HEADER_FLAG_RELOC is set then we have FULL_BACKREF set.
6002  * 2) If btrfs_header_owner(buf) no longer points to buf then we have
6003  *      FULL_BACKREF set.
6004  * 3) We cowed the block walking down a reloc tree.  This is impossible to tell
6005  *    if it happened after the relocation occurred since we'll have dropped the
6006  *    reloc root, so it's entirely possible to have FULL_BACKREF set on buf and
6007  *    have no real way to know for sure.
6008  *
6009  * We process the blocks one root at a time, and we start from the lowest root
6010  * objectid and go to the highest.  So we can just lookup the owner backref for
6011  * the record and if we don't find it then we know it doesn't exist and we have
6012  * a FULL BACKREF.
6013  *
6014  * FIXME: if we ever start reclaiming root objectid's then we need to fix this
6015  * assumption and simply indicate that we _think_ that the FULL BACKREF needs to
6016  * be set or not and then we can check later once we've gathered all the refs.
6017  */
6018 static int calc_extent_flag(struct btrfs_root *root,
6019                            struct cache_tree *extent_cache,
6020                            struct extent_buffer *buf,
6021                            struct root_item_record *ri,
6022                            u64 *flags)
6023 {
6024         struct extent_record *rec;
6025         struct cache_extent *cache;
6026         struct tree_backref *tback;
6027         u64 owner = 0;
6028
6029         cache = lookup_cache_extent(extent_cache, buf->start, 1);
6030         /* we have added this extent before */
6031         if (!cache)
6032                 return -ENOENT;
6033
6034         rec = container_of(cache, struct extent_record, cache);
6035
6036         /*
6037          * Except file/reloc tree, we can not have
6038          * FULL BACKREF MODE
6039          */
6040         if (ri->objectid < BTRFS_FIRST_FREE_OBJECTID)
6041                 goto normal;
6042         /*
6043          * root node
6044          */
6045         if (buf->start == ri->bytenr)
6046                 goto normal;
6047
6048         if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
6049                 goto full_backref;
6050
6051         owner = btrfs_header_owner(buf);
6052         if (owner == ri->objectid)
6053                 goto normal;
6054
6055         tback = find_tree_backref(rec, 0, owner);
6056         if (!tback)
6057                 goto full_backref;
6058 normal:
6059         *flags = 0;
6060         if (rec->flag_block_full_backref != FLAG_UNSET &&
6061             rec->flag_block_full_backref != 0)
6062                 rec->bad_full_backref = 1;
6063         return 0;
6064 full_backref:
6065         *flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6066         if (rec->flag_block_full_backref != FLAG_UNSET &&
6067             rec->flag_block_full_backref != 1)
6068                 rec->bad_full_backref = 1;
6069         return 0;
6070 }
6071
6072 static void report_mismatch_key_root(u8 key_type, u64 rootid)
6073 {
6074         fprintf(stderr, "Invalid key type(");
6075         print_key_type(stderr, 0, key_type);
6076         fprintf(stderr, ") found in root(");
6077         print_objectid(stderr, rootid, 0);
6078         fprintf(stderr, ")\n");
6079 }
6080
6081 /*
6082  * Check if the key is valid with its extent buffer.
6083  *
6084  * This is a early check in case invalid key exists in a extent buffer
6085  * This is not comprehensive yet, but should prevent wrong key/item passed
6086  * further
6087  */
6088 static int check_type_with_root(u64 rootid, u8 key_type)
6089 {
6090         switch (key_type) {
6091         /* Only valid in chunk tree */
6092         case BTRFS_DEV_ITEM_KEY:
6093         case BTRFS_CHUNK_ITEM_KEY:
6094                 if (rootid != BTRFS_CHUNK_TREE_OBJECTID)
6095                         goto err;
6096                 break;
6097         /* valid in csum and log tree */
6098         case BTRFS_CSUM_TREE_OBJECTID:
6099                 if (!(rootid == BTRFS_TREE_LOG_OBJECTID ||
6100                       is_fstree(rootid)))
6101                         goto err;
6102                 break;
6103         case BTRFS_EXTENT_ITEM_KEY:
6104         case BTRFS_METADATA_ITEM_KEY:
6105         case BTRFS_BLOCK_GROUP_ITEM_KEY:
6106                 if (rootid != BTRFS_EXTENT_TREE_OBJECTID)
6107                         goto err;
6108                 break;
6109         case BTRFS_ROOT_ITEM_KEY:
6110                 if (rootid != BTRFS_ROOT_TREE_OBJECTID)
6111                         goto err;
6112                 break;
6113         case BTRFS_DEV_EXTENT_KEY:
6114                 if (rootid != BTRFS_DEV_TREE_OBJECTID)
6115                         goto err;
6116                 break;
6117         }
6118         return 0;
6119 err:
6120         report_mismatch_key_root(key_type, rootid);
6121         return -EINVAL;
6122 }
6123
6124 static int run_next_block(struct btrfs_root *root,
6125                           struct block_info *bits,
6126                           int bits_nr,
6127                           u64 *last,
6128                           struct cache_tree *pending,
6129                           struct cache_tree *seen,
6130                           struct cache_tree *reada,
6131                           struct cache_tree *nodes,
6132                           struct cache_tree *extent_cache,
6133                           struct cache_tree *chunk_cache,
6134                           struct rb_root *dev_cache,
6135                           struct block_group_tree *block_group_cache,
6136                           struct device_extent_tree *dev_extent_cache,
6137                           struct root_item_record *ri)
6138 {
6139         struct extent_buffer *buf;
6140         struct extent_record *rec = NULL;
6141         u64 bytenr;
6142         u32 size;
6143         u64 parent;
6144         u64 owner;
6145         u64 flags;
6146         u64 ptr;
6147         u64 gen = 0;
6148         int ret = 0;
6149         int i;
6150         int nritems;
6151         struct btrfs_key key;
6152         struct cache_extent *cache;
6153         int reada_bits;
6154
6155         nritems = pick_next_pending(pending, reada, nodes, *last, bits,
6156                                     bits_nr, &reada_bits);
6157         if (nritems == 0)
6158                 return 1;
6159
6160         if (!reada_bits) {
6161                 for(i = 0; i < nritems; i++) {
6162                         ret = add_cache_extent(reada, bits[i].start,
6163                                                bits[i].size);
6164                         if (ret == -EEXIST)
6165                                 continue;
6166
6167                         /* fixme, get the parent transid */
6168                         readahead_tree_block(root, bits[i].start,
6169                                              bits[i].size, 0);
6170                 }
6171         }
6172         *last = bits[0].start;
6173         bytenr = bits[0].start;
6174         size = bits[0].size;
6175
6176         cache = lookup_cache_extent(pending, bytenr, size);
6177         if (cache) {
6178                 remove_cache_extent(pending, cache);
6179                 free(cache);
6180         }
6181         cache = lookup_cache_extent(reada, bytenr, size);
6182         if (cache) {
6183                 remove_cache_extent(reada, cache);
6184                 free(cache);
6185         }
6186         cache = lookup_cache_extent(nodes, bytenr, size);
6187         if (cache) {
6188                 remove_cache_extent(nodes, cache);
6189                 free(cache);
6190         }
6191         cache = lookup_cache_extent(extent_cache, bytenr, size);
6192         if (cache) {
6193                 rec = container_of(cache, struct extent_record, cache);
6194                 gen = rec->parent_generation;
6195         }
6196
6197         /* fixme, get the real parent transid */
6198         buf = read_tree_block(root, bytenr, size, gen);
6199         if (!extent_buffer_uptodate(buf)) {
6200                 record_bad_block_io(root->fs_info,
6201                                     extent_cache, bytenr, size);
6202                 goto out;
6203         }
6204
6205         nritems = btrfs_header_nritems(buf);
6206
6207         flags = 0;
6208         if (!init_extent_tree) {
6209                 ret = btrfs_lookup_extent_info(NULL, root, bytenr,
6210                                        btrfs_header_level(buf), 1, NULL,
6211                                        &flags);
6212                 if (ret < 0) {
6213                         ret = calc_extent_flag(root, extent_cache, buf, ri, &flags);
6214                         if (ret < 0) {
6215                                 fprintf(stderr, "Couldn't calc extent flags\n");
6216                                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6217                         }
6218                 }
6219         } else {
6220                 flags = 0;
6221                 ret = calc_extent_flag(root, extent_cache, buf, ri, &flags);
6222                 if (ret < 0) {
6223                         fprintf(stderr, "Couldn't calc extent flags\n");
6224                         flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6225                 }
6226         }
6227
6228         if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6229                 if (ri != NULL &&
6230                     ri->objectid != BTRFS_TREE_RELOC_OBJECTID &&
6231                     ri->objectid == btrfs_header_owner(buf)) {
6232                         /*
6233                          * Ok we got to this block from it's original owner and
6234                          * we have FULL_BACKREF set.  Relocation can leave
6235                          * converted blocks over so this is altogether possible,
6236                          * however it's not possible if the generation > the
6237                          * last snapshot, so check for this case.
6238                          */
6239                         if (!btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC) &&
6240                             btrfs_header_generation(buf) > ri->last_snapshot) {
6241                                 flags &= ~BTRFS_BLOCK_FLAG_FULL_BACKREF;
6242                                 rec->bad_full_backref = 1;
6243                         }
6244                 }
6245         } else {
6246                 if (ri != NULL &&
6247                     (ri->objectid == BTRFS_TREE_RELOC_OBJECTID ||
6248                      btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) {
6249                         flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6250                         rec->bad_full_backref = 1;
6251                 }
6252         }
6253
6254         if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6255                 rec->flag_block_full_backref = 1;
6256                 parent = bytenr;
6257                 owner = 0;
6258         } else {
6259                 rec->flag_block_full_backref = 0;
6260                 parent = 0;
6261                 owner = btrfs_header_owner(buf);
6262         }
6263
6264         ret = check_block(root, extent_cache, buf, flags);
6265         if (ret)
6266                 goto out;
6267
6268         if (btrfs_is_leaf(buf)) {
6269                 btree_space_waste += btrfs_leaf_free_space(root, buf);
6270                 for (i = 0; i < nritems; i++) {
6271                         struct btrfs_file_extent_item *fi;
6272                         btrfs_item_key_to_cpu(buf, &key, i);
6273                         /*
6274                          * Check key type against the leaf owner.
6275                          * Could filter quite a lot of early error if
6276                          * owner is correct
6277                          */
6278                         if (check_type_with_root(btrfs_header_owner(buf),
6279                                                  key.type)) {
6280                                 fprintf(stderr, "ignoring invalid key\n");
6281                                 continue;
6282                         }
6283                         if (key.type == BTRFS_EXTENT_ITEM_KEY) {
6284                                 process_extent_item(root, extent_cache, buf,
6285                                                     i);
6286                                 continue;
6287                         }
6288                         if (key.type == BTRFS_METADATA_ITEM_KEY) {
6289                                 process_extent_item(root, extent_cache, buf,
6290                                                     i);
6291                                 continue;
6292                         }
6293                         if (key.type == BTRFS_EXTENT_CSUM_KEY) {
6294                                 total_csum_bytes +=
6295                                         btrfs_item_size_nr(buf, i);
6296                                 continue;
6297                         }
6298                         if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6299                                 process_chunk_item(chunk_cache, &key, buf, i);
6300                                 continue;
6301                         }
6302                         if (key.type == BTRFS_DEV_ITEM_KEY) {
6303                                 process_device_item(dev_cache, &key, buf, i);
6304                                 continue;
6305                         }
6306                         if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
6307                                 process_block_group_item(block_group_cache,
6308                                         &key, buf, i);
6309                                 continue;
6310                         }
6311                         if (key.type == BTRFS_DEV_EXTENT_KEY) {
6312                                 process_device_extent_item(dev_extent_cache,
6313                                         &key, buf, i);
6314                                 continue;
6315
6316                         }
6317                         if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
6318 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6319                                 process_extent_ref_v0(extent_cache, buf, i);
6320 #else
6321                                 BUG();
6322 #endif
6323                                 continue;
6324                         }
6325
6326                         if (key.type == BTRFS_TREE_BLOCK_REF_KEY) {
6327                                 ret = add_tree_backref(extent_cache,
6328                                                 key.objectid, 0, key.offset, 0);
6329                                 if (ret < 0)
6330                                         error("add_tree_backref failed: %s",
6331                                               strerror(-ret));
6332                                 continue;
6333                         }
6334                         if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
6335                                 ret = add_tree_backref(extent_cache,
6336                                                 key.objectid, key.offset, 0, 0);
6337                                 if (ret < 0)
6338                                         error("add_tree_backref failed: %s",
6339                                               strerror(-ret));
6340                                 continue;
6341                         }
6342                         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
6343                                 struct btrfs_extent_data_ref *ref;
6344                                 ref = btrfs_item_ptr(buf, i,
6345                                                 struct btrfs_extent_data_ref);
6346                                 add_data_backref(extent_cache,
6347                                         key.objectid, 0,
6348                                         btrfs_extent_data_ref_root(buf, ref),
6349                                         btrfs_extent_data_ref_objectid(buf,
6350                                                                        ref),
6351                                         btrfs_extent_data_ref_offset(buf, ref),
6352                                         btrfs_extent_data_ref_count(buf, ref),
6353                                         0, root->sectorsize);
6354                                 continue;
6355                         }
6356                         if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
6357                                 struct btrfs_shared_data_ref *ref;
6358                                 ref = btrfs_item_ptr(buf, i,
6359                                                 struct btrfs_shared_data_ref);
6360                                 add_data_backref(extent_cache,
6361                                         key.objectid, key.offset, 0, 0, 0,
6362                                         btrfs_shared_data_ref_count(buf, ref),
6363                                         0, root->sectorsize);
6364                                 continue;
6365                         }
6366                         if (key.type == BTRFS_ORPHAN_ITEM_KEY) {
6367                                 struct bad_item *bad;
6368
6369                                 if (key.objectid == BTRFS_ORPHAN_OBJECTID)
6370                                         continue;
6371                                 if (!owner)
6372                                         continue;
6373                                 bad = malloc(sizeof(struct bad_item));
6374                                 if (!bad)
6375                                         continue;
6376                                 INIT_LIST_HEAD(&bad->list);
6377                                 memcpy(&bad->key, &key,
6378                                        sizeof(struct btrfs_key));
6379                                 bad->root_id = owner;
6380                                 list_add_tail(&bad->list, &delete_items);
6381                                 continue;
6382                         }
6383                         if (key.type != BTRFS_EXTENT_DATA_KEY)
6384                                 continue;
6385                         fi = btrfs_item_ptr(buf, i,
6386                                             struct btrfs_file_extent_item);
6387                         if (btrfs_file_extent_type(buf, fi) ==
6388                             BTRFS_FILE_EXTENT_INLINE)
6389                                 continue;
6390                         if (btrfs_file_extent_disk_bytenr(buf, fi) == 0)
6391                                 continue;
6392
6393                         data_bytes_allocated +=
6394                                 btrfs_file_extent_disk_num_bytes(buf, fi);
6395                         if (data_bytes_allocated < root->sectorsize) {
6396                                 abort();
6397                         }
6398                         data_bytes_referenced +=
6399                                 btrfs_file_extent_num_bytes(buf, fi);
6400                         add_data_backref(extent_cache,
6401                                 btrfs_file_extent_disk_bytenr(buf, fi),
6402                                 parent, owner, key.objectid, key.offset -
6403                                 btrfs_file_extent_offset(buf, fi), 1, 1,
6404                                 btrfs_file_extent_disk_num_bytes(buf, fi));
6405                 }
6406         } else {
6407                 int level;
6408                 struct btrfs_key first_key;
6409
6410                 first_key.objectid = 0;
6411
6412                 if (nritems > 0)
6413                         btrfs_item_key_to_cpu(buf, &first_key, 0);
6414                 level = btrfs_header_level(buf);
6415                 for (i = 0; i < nritems; i++) {
6416                         struct extent_record tmpl;
6417
6418                         ptr = btrfs_node_blockptr(buf, i);
6419                         size = root->nodesize;
6420                         btrfs_node_key_to_cpu(buf, &key, i);
6421                         if (ri != NULL) {
6422                                 if ((level == ri->drop_level)
6423                                     && is_dropped_key(&key, &ri->drop_key)) {
6424                                         continue;
6425                                 }
6426                         }
6427
6428                         memset(&tmpl, 0, sizeof(tmpl));
6429                         btrfs_cpu_key_to_disk(&tmpl.parent_key, &key);
6430                         tmpl.parent_generation = btrfs_node_ptr_generation(buf, i);
6431                         tmpl.start = ptr;
6432                         tmpl.nr = size;
6433                         tmpl.refs = 1;
6434                         tmpl.metadata = 1;
6435                         tmpl.max_size = size;
6436                         ret = add_extent_rec(extent_cache, &tmpl);
6437                         if (ret < 0)
6438                                 goto out;
6439
6440                         ret = add_tree_backref(extent_cache, ptr, parent,
6441                                         owner, 1);
6442                         if (ret < 0) {
6443                                 error("add_tree_backref failed: %s",
6444                                       strerror(-ret));
6445                                 continue;
6446                         }
6447
6448                         if (level > 1) {
6449                                 add_pending(nodes, seen, ptr, size);
6450                         } else {
6451                                 add_pending(pending, seen, ptr, size);
6452                         }
6453                 }
6454                 btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) -
6455                                       nritems) * sizeof(struct btrfs_key_ptr);
6456         }
6457         total_btree_bytes += buf->len;
6458         if (fs_root_objectid(btrfs_header_owner(buf)))
6459                 total_fs_tree_bytes += buf->len;
6460         if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID)
6461                 total_extent_tree_bytes += buf->len;
6462         if (!found_old_backref &&
6463             btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID &&
6464             btrfs_header_backref_rev(buf) == BTRFS_MIXED_BACKREF_REV &&
6465             !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
6466                 found_old_backref = 1;
6467 out:
6468         free_extent_buffer(buf);
6469         return ret;
6470 }
6471
6472 static int add_root_to_pending(struct extent_buffer *buf,
6473                                struct cache_tree *extent_cache,
6474                                struct cache_tree *pending,
6475                                struct cache_tree *seen,
6476                                struct cache_tree *nodes,
6477                                u64 objectid)
6478 {
6479         struct extent_record tmpl;
6480         int ret;
6481
6482         if (btrfs_header_level(buf) > 0)
6483                 add_pending(nodes, seen, buf->start, buf->len);
6484         else
6485                 add_pending(pending, seen, buf->start, buf->len);
6486
6487         memset(&tmpl, 0, sizeof(tmpl));
6488         tmpl.start = buf->start;
6489         tmpl.nr = buf->len;
6490         tmpl.is_root = 1;
6491         tmpl.refs = 1;
6492         tmpl.metadata = 1;
6493         tmpl.max_size = buf->len;
6494         add_extent_rec(extent_cache, &tmpl);
6495
6496         if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
6497             btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
6498                 ret = add_tree_backref(extent_cache, buf->start, buf->start,
6499                                 0, 1);
6500         else
6501                 ret = add_tree_backref(extent_cache, buf->start, 0, objectid,
6502                                 1);
6503         return ret;
6504 }
6505
6506 /* as we fix the tree, we might be deleting blocks that
6507  * we're tracking for repair.  This hook makes sure we
6508  * remove any backrefs for blocks as we are fixing them.
6509  */
6510 static int free_extent_hook(struct btrfs_trans_handle *trans,
6511                             struct btrfs_root *root,
6512                             u64 bytenr, u64 num_bytes, u64 parent,
6513                             u64 root_objectid, u64 owner, u64 offset,
6514                             int refs_to_drop)
6515 {
6516         struct extent_record *rec;
6517         struct cache_extent *cache;
6518         int is_data;
6519         struct cache_tree *extent_cache = root->fs_info->fsck_extent_cache;
6520
6521         is_data = owner >= BTRFS_FIRST_FREE_OBJECTID;
6522         cache = lookup_cache_extent(extent_cache, bytenr, num_bytes);
6523         if (!cache)
6524                 return 0;
6525
6526         rec = container_of(cache, struct extent_record, cache);
6527         if (is_data) {
6528                 struct data_backref *back;
6529                 back = find_data_backref(rec, parent, root_objectid, owner,
6530                                          offset, 1, bytenr, num_bytes);
6531                 if (!back)
6532                         goto out;
6533                 if (back->node.found_ref) {
6534                         back->found_ref -= refs_to_drop;
6535                         if (rec->refs)
6536                                 rec->refs -= refs_to_drop;
6537                 }
6538                 if (back->node.found_extent_tree) {
6539                         back->num_refs -= refs_to_drop;
6540                         if (rec->extent_item_refs)
6541                                 rec->extent_item_refs -= refs_to_drop;
6542                 }
6543                 if (back->found_ref == 0)
6544                         back->node.found_ref = 0;
6545                 if (back->num_refs == 0)
6546                         back->node.found_extent_tree = 0;
6547
6548                 if (!back->node.found_extent_tree && back->node.found_ref) {
6549                         list_del(&back->node.list);
6550                         free(back);
6551                 }
6552         } else {
6553                 struct tree_backref *back;
6554                 back = find_tree_backref(rec, parent, root_objectid);
6555                 if (!back)
6556                         goto out;
6557                 if (back->node.found_ref) {
6558                         if (rec->refs)
6559                                 rec->refs--;
6560                         back->node.found_ref = 0;
6561                 }
6562                 if (back->node.found_extent_tree) {
6563                         if (rec->extent_item_refs)
6564                                 rec->extent_item_refs--;
6565                         back->node.found_extent_tree = 0;
6566                 }
6567                 if (!back->node.found_extent_tree && back->node.found_ref) {
6568                         list_del(&back->node.list);
6569                         free(back);
6570                 }
6571         }
6572         maybe_free_extent_rec(extent_cache, rec);
6573 out:
6574         return 0;
6575 }
6576
6577 static int delete_extent_records(struct btrfs_trans_handle *trans,
6578                                  struct btrfs_root *root,
6579                                  struct btrfs_path *path,
6580                                  u64 bytenr, u64 new_len)
6581 {
6582         struct btrfs_key key;
6583         struct btrfs_key found_key;
6584         struct extent_buffer *leaf;
6585         int ret;
6586         int slot;
6587
6588
6589         key.objectid = bytenr;
6590         key.type = (u8)-1;
6591         key.offset = (u64)-1;
6592
6593         while(1) {
6594                 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
6595                                         &key, path, 0, 1);
6596                 if (ret < 0)
6597                         break;
6598
6599                 if (ret > 0) {
6600                         ret = 0;
6601                         if (path->slots[0] == 0)
6602                                 break;
6603                         path->slots[0]--;
6604                 }
6605                 ret = 0;
6606
6607                 leaf = path->nodes[0];
6608                 slot = path->slots[0];
6609
6610                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6611                 if (found_key.objectid != bytenr)
6612                         break;
6613
6614                 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
6615                     found_key.type != BTRFS_METADATA_ITEM_KEY &&
6616                     found_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
6617                     found_key.type != BTRFS_EXTENT_DATA_REF_KEY &&
6618                     found_key.type != BTRFS_EXTENT_REF_V0_KEY &&
6619                     found_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
6620                     found_key.type != BTRFS_SHARED_DATA_REF_KEY) {
6621                         btrfs_release_path(path);
6622                         if (found_key.type == 0) {
6623                                 if (found_key.offset == 0)
6624                                         break;
6625                                 key.offset = found_key.offset - 1;
6626                                 key.type = found_key.type;
6627                         }
6628                         key.type = found_key.type - 1;
6629                         key.offset = (u64)-1;
6630                         continue;
6631                 }
6632
6633                 fprintf(stderr, "repair deleting extent record: key %Lu %u %Lu\n",
6634                         found_key.objectid, found_key.type, found_key.offset);
6635
6636                 ret = btrfs_del_item(trans, root->fs_info->extent_root, path);
6637                 if (ret)
6638                         break;
6639                 btrfs_release_path(path);
6640
6641                 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
6642                     found_key.type == BTRFS_METADATA_ITEM_KEY) {
6643                         u64 bytes = (found_key.type == BTRFS_EXTENT_ITEM_KEY) ?
6644                                 found_key.offset : root->nodesize;
6645
6646                         ret = btrfs_update_block_group(trans, root, bytenr,
6647                                                        bytes, 0, 0);
6648                         if (ret)
6649                                 break;
6650                 }
6651         }
6652
6653         btrfs_release_path(path);
6654         return ret;
6655 }
6656
6657 /*
6658  * for a single backref, this will allocate a new extent
6659  * and add the backref to it.
6660  */
6661 static int record_extent(struct btrfs_trans_handle *trans,
6662                          struct btrfs_fs_info *info,
6663                          struct btrfs_path *path,
6664                          struct extent_record *rec,
6665                          struct extent_backref *back,
6666                          int allocated, u64 flags)
6667 {
6668         int ret;
6669         struct btrfs_root *extent_root = info->extent_root;
6670         struct extent_buffer *leaf;
6671         struct btrfs_key ins_key;
6672         struct btrfs_extent_item *ei;
6673         struct data_backref *dback;
6674         struct btrfs_tree_block_info *bi;
6675
6676         if (!back->is_data)
6677                 rec->max_size = max_t(u64, rec->max_size,
6678                                     info->extent_root->nodesize);
6679
6680         if (!allocated) {
6681                 u32 item_size = sizeof(*ei);
6682
6683                 if (!back->is_data)
6684                         item_size += sizeof(*bi);
6685
6686                 ins_key.objectid = rec->start;
6687                 ins_key.offset = rec->max_size;
6688                 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
6689
6690                 ret = btrfs_insert_empty_item(trans, extent_root, path,
6691                                         &ins_key, item_size);
6692                 if (ret)
6693                         goto fail;
6694
6695                 leaf = path->nodes[0];
6696                 ei = btrfs_item_ptr(leaf, path->slots[0],
6697                                     struct btrfs_extent_item);
6698
6699                 btrfs_set_extent_refs(leaf, ei, 0);
6700                 btrfs_set_extent_generation(leaf, ei, rec->generation);
6701
6702                 if (back->is_data) {
6703                         btrfs_set_extent_flags(leaf, ei,
6704                                                BTRFS_EXTENT_FLAG_DATA);
6705                 } else {
6706                         struct btrfs_disk_key copy_key;;
6707
6708                         bi = (struct btrfs_tree_block_info *)(ei + 1);
6709                         memset_extent_buffer(leaf, 0, (unsigned long)bi,
6710                                              sizeof(*bi));
6711
6712                         btrfs_set_disk_key_objectid(&copy_key,
6713                                                     rec->info_objectid);
6714                         btrfs_set_disk_key_type(&copy_key, 0);
6715                         btrfs_set_disk_key_offset(&copy_key, 0);
6716
6717                         btrfs_set_tree_block_level(leaf, bi, rec->info_level);
6718                         btrfs_set_tree_block_key(leaf, bi, &copy_key);
6719
6720                         btrfs_set_extent_flags(leaf, ei,
6721                                                BTRFS_EXTENT_FLAG_TREE_BLOCK | flags);
6722                 }
6723
6724                 btrfs_mark_buffer_dirty(leaf);
6725                 ret = btrfs_update_block_group(trans, extent_root, rec->start,
6726                                                rec->max_size, 1, 0);
6727                 if (ret)
6728                         goto fail;
6729                 btrfs_release_path(path);
6730         }
6731
6732         if (back->is_data) {
6733                 u64 parent;
6734                 int i;
6735
6736                 dback = to_data_backref(back);
6737                 if (back->full_backref)
6738                         parent = dback->parent;
6739                 else
6740                         parent = 0;
6741
6742                 for (i = 0; i < dback->found_ref; i++) {
6743                         /* if parent != 0, we're doing a full backref
6744                          * passing BTRFS_FIRST_FREE_OBJECTID as the owner
6745                          * just makes the backref allocator create a data
6746                          * backref
6747                          */
6748                         ret = btrfs_inc_extent_ref(trans, info->extent_root,
6749                                                    rec->start, rec->max_size,
6750                                                    parent,
6751                                                    dback->root,
6752                                                    parent ?
6753                                                    BTRFS_FIRST_FREE_OBJECTID :
6754                                                    dback->owner,
6755                                                    dback->offset);
6756                         if (ret)
6757                                 break;
6758                 }
6759                 fprintf(stderr, "adding new data backref"
6760                                 " on %llu %s %llu owner %llu"
6761                                 " offset %llu found %d\n",
6762                                 (unsigned long long)rec->start,
6763                                 back->full_backref ?
6764                                 "parent" : "root",
6765                                 back->full_backref ?
6766                                 (unsigned long long)parent :
6767                                 (unsigned long long)dback->root,
6768                                 (unsigned long long)dback->owner,
6769                                 (unsigned long long)dback->offset,
6770                                 dback->found_ref);
6771         } else {
6772                 u64 parent;
6773                 struct tree_backref *tback;
6774
6775                 tback = to_tree_backref(back);
6776                 if (back->full_backref)
6777                         parent = tback->parent;
6778                 else
6779                         parent = 0;
6780
6781                 ret = btrfs_inc_extent_ref(trans, info->extent_root,
6782                                            rec->start, rec->max_size,
6783                                            parent, tback->root, 0, 0);
6784                 fprintf(stderr, "adding new tree backref on "
6785                         "start %llu len %llu parent %llu root %llu\n",
6786                         rec->start, rec->max_size, parent, tback->root);
6787         }
6788 fail:
6789         btrfs_release_path(path);
6790         return ret;
6791 }
6792
6793 static struct extent_entry *find_entry(struct list_head *entries,
6794                                        u64 bytenr, u64 bytes)
6795 {
6796         struct extent_entry *entry = NULL;
6797
6798         list_for_each_entry(entry, entries, list) {
6799                 if (entry->bytenr == bytenr && entry->bytes == bytes)
6800                         return entry;
6801         }
6802
6803         return NULL;
6804 }
6805
6806 static struct extent_entry *find_most_right_entry(struct list_head *entries)
6807 {
6808         struct extent_entry *entry, *best = NULL, *prev = NULL;
6809
6810         list_for_each_entry(entry, entries, list) {
6811                 if (!prev) {
6812                         prev = entry;
6813                         continue;
6814                 }
6815
6816                 /*
6817                  * If there are as many broken entries as entries then we know
6818                  * not to trust this particular entry.
6819                  */
6820                 if (entry->broken == entry->count)
6821                         continue;
6822
6823                 /*
6824                  * If our current entry == best then we can't be sure our best
6825                  * is really the best, so we need to keep searching.
6826                  */
6827                 if (best && best->count == entry->count) {
6828                         prev = entry;
6829                         best = NULL;
6830                         continue;
6831                 }
6832
6833                 /* Prev == entry, not good enough, have to keep searching */
6834                 if (!prev->broken && prev->count == entry->count)
6835                         continue;
6836
6837                 if (!best)
6838                         best = (prev->count > entry->count) ? prev : entry;
6839                 else if (best->count < entry->count)
6840                         best = entry;
6841                 prev = entry;
6842         }
6843
6844         return best;
6845 }
6846
6847 static int repair_ref(struct btrfs_fs_info *info, struct btrfs_path *path,
6848                       struct data_backref *dback, struct extent_entry *entry)
6849 {
6850         struct btrfs_trans_handle *trans;
6851         struct btrfs_root *root;
6852         struct btrfs_file_extent_item *fi;
6853         struct extent_buffer *leaf;
6854         struct btrfs_key key;
6855         u64 bytenr, bytes;
6856         int ret, err;
6857
6858         key.objectid = dback->root;
6859         key.type = BTRFS_ROOT_ITEM_KEY;
6860         key.offset = (u64)-1;
6861         root = btrfs_read_fs_root(info, &key);
6862         if (IS_ERR(root)) {
6863                 fprintf(stderr, "Couldn't find root for our ref\n");
6864                 return -EINVAL;
6865         }
6866
6867         /*
6868          * The backref points to the original offset of the extent if it was
6869          * split, so we need to search down to the offset we have and then walk
6870          * forward until we find the backref we're looking for.
6871          */
6872         key.objectid = dback->owner;
6873         key.type = BTRFS_EXTENT_DATA_KEY;
6874         key.offset = dback->offset;
6875         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6876         if (ret < 0) {
6877                 fprintf(stderr, "Error looking up ref %d\n", ret);
6878                 return ret;
6879         }
6880
6881         while (1) {
6882                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
6883                         ret = btrfs_next_leaf(root, path);
6884                         if (ret) {
6885                                 fprintf(stderr, "Couldn't find our ref, next\n");
6886                                 return -EINVAL;
6887                         }
6888                 }
6889                 leaf = path->nodes[0];
6890                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6891                 if (key.objectid != dback->owner ||
6892                     key.type != BTRFS_EXTENT_DATA_KEY) {
6893                         fprintf(stderr, "Couldn't find our ref, search\n");
6894                         return -EINVAL;
6895                 }
6896                 fi = btrfs_item_ptr(leaf, path->slots[0],
6897                                     struct btrfs_file_extent_item);
6898                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6899                 bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
6900
6901                 if (bytenr == dback->disk_bytenr && bytes == dback->bytes)
6902                         break;
6903                 path->slots[0]++;
6904         }
6905
6906         btrfs_release_path(path);
6907
6908         trans = btrfs_start_transaction(root, 1);
6909         if (IS_ERR(trans))
6910                 return PTR_ERR(trans);
6911
6912         /*
6913          * Ok we have the key of the file extent we want to fix, now we can cow
6914          * down to the thing and fix it.
6915          */
6916         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6917         if (ret < 0) {
6918                 fprintf(stderr, "Error cowing down to ref [%Lu, %u, %Lu]: %d\n",
6919                         key.objectid, key.type, key.offset, ret);
6920                 goto out;
6921         }
6922         if (ret > 0) {
6923                 fprintf(stderr, "Well that's odd, we just found this key "
6924                         "[%Lu, %u, %Lu]\n", key.objectid, key.type,
6925                         key.offset);
6926                 ret = -EINVAL;
6927                 goto out;
6928         }
6929         leaf = path->nodes[0];
6930         fi = btrfs_item_ptr(leaf, path->slots[0],
6931                             struct btrfs_file_extent_item);
6932
6933         if (btrfs_file_extent_compression(leaf, fi) &&
6934             dback->disk_bytenr != entry->bytenr) {
6935                 fprintf(stderr, "Ref doesn't match the record start and is "
6936                         "compressed, please take a btrfs-image of this file "
6937                         "system and send it to a btrfs developer so they can "
6938                         "complete this functionality for bytenr %Lu\n",
6939                         dback->disk_bytenr);
6940                 ret = -EINVAL;
6941                 goto out;
6942         }
6943
6944         if (dback->node.broken && dback->disk_bytenr != entry->bytenr) {
6945                 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
6946         } else if (dback->disk_bytenr > entry->bytenr) {
6947                 u64 off_diff, offset;
6948
6949                 off_diff = dback->disk_bytenr - entry->bytenr;
6950                 offset = btrfs_file_extent_offset(leaf, fi);
6951                 if (dback->disk_bytenr + offset +
6952                     btrfs_file_extent_num_bytes(leaf, fi) >
6953                     entry->bytenr + entry->bytes) {
6954                         fprintf(stderr, "Ref is past the entry end, please "
6955                                 "take a btrfs-image of this file system and "
6956                                 "send it to a btrfs developer, ref %Lu\n",
6957                                 dback->disk_bytenr);
6958                         ret = -EINVAL;
6959                         goto out;
6960                 }
6961                 offset += off_diff;
6962                 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
6963                 btrfs_set_file_extent_offset(leaf, fi, offset);
6964         } else if (dback->disk_bytenr < entry->bytenr) {
6965                 u64 offset;
6966
6967                 offset = btrfs_file_extent_offset(leaf, fi);
6968                 if (dback->disk_bytenr + offset < entry->bytenr) {
6969                         fprintf(stderr, "Ref is before the entry start, please"
6970                                 " take a btrfs-image of this file system and "
6971                                 "send it to a btrfs developer, ref %Lu\n",
6972                                 dback->disk_bytenr);
6973                         ret = -EINVAL;
6974                         goto out;
6975                 }
6976
6977                 offset += dback->disk_bytenr;
6978                 offset -= entry->bytenr;
6979                 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
6980                 btrfs_set_file_extent_offset(leaf, fi, offset);
6981         }
6982
6983         btrfs_set_file_extent_disk_num_bytes(leaf, fi, entry->bytes);
6984
6985         /*
6986          * Chances are if disk_num_bytes were wrong then so is ram_bytes, but
6987          * only do this if we aren't using compression, otherwise it's a
6988          * trickier case.
6989          */
6990         if (!btrfs_file_extent_compression(leaf, fi))
6991                 btrfs_set_file_extent_ram_bytes(leaf, fi, entry->bytes);
6992         else
6993                 printf("ram bytes may be wrong?\n");
6994         btrfs_mark_buffer_dirty(leaf);
6995 out:
6996         err = btrfs_commit_transaction(trans, root);
6997         btrfs_release_path(path);
6998         return ret ? ret : err;
6999 }
7000
7001 static int verify_backrefs(struct btrfs_fs_info *info, struct btrfs_path *path,
7002                            struct extent_record *rec)
7003 {
7004         struct extent_backref *back;
7005         struct data_backref *dback;
7006         struct extent_entry *entry, *best = NULL;
7007         LIST_HEAD(entries);
7008         int nr_entries = 0;
7009         int broken_entries = 0;
7010         int ret = 0;
7011         short mismatch = 0;
7012
7013         /*
7014          * Metadata is easy and the backrefs should always agree on bytenr and
7015          * size, if not we've got bigger issues.
7016          */
7017         if (rec->metadata)
7018                 return 0;
7019
7020         list_for_each_entry(back, &rec->backrefs, list) {
7021                 if (back->full_backref || !back->is_data)
7022                         continue;
7023
7024                 dback = to_data_backref(back);
7025
7026                 /*
7027                  * We only pay attention to backrefs that we found a real
7028                  * backref for.
7029                  */
7030                 if (dback->found_ref == 0)
7031                         continue;
7032
7033                 /*
7034                  * For now we only catch when the bytes don't match, not the
7035                  * bytenr.  We can easily do this at the same time, but I want
7036                  * to have a fs image to test on before we just add repair
7037                  * functionality willy-nilly so we know we won't screw up the
7038                  * repair.
7039                  */
7040
7041                 entry = find_entry(&entries, dback->disk_bytenr,
7042                                    dback->bytes);
7043                 if (!entry) {
7044                         entry = malloc(sizeof(struct extent_entry));
7045                         if (!entry) {
7046                                 ret = -ENOMEM;
7047                                 goto out;
7048                         }
7049                         memset(entry, 0, sizeof(*entry));
7050                         entry->bytenr = dback->disk_bytenr;
7051                         entry->bytes = dback->bytes;
7052                         list_add_tail(&entry->list, &entries);
7053                         nr_entries++;
7054                 }
7055
7056                 /*
7057                  * If we only have on entry we may think the entries agree when
7058                  * in reality they don't so we have to do some extra checking.
7059                  */
7060                 if (dback->disk_bytenr != rec->start ||
7061                     dback->bytes != rec->nr || back->broken)
7062                         mismatch = 1;
7063
7064                 if (back->broken) {
7065                         entry->broken++;
7066                         broken_entries++;
7067                 }
7068
7069                 entry->count++;
7070         }
7071
7072         /* Yay all the backrefs agree, carry on good sir */
7073         if (nr_entries <= 1 && !mismatch)
7074                 goto out;
7075
7076         fprintf(stderr, "attempting to repair backref discrepency for bytenr "
7077                 "%Lu\n", rec->start);
7078
7079         /*
7080          * First we want to see if the backrefs can agree amongst themselves who
7081          * is right, so figure out which one of the entries has the highest
7082          * count.
7083          */
7084         best = find_most_right_entry(&entries);
7085
7086         /*
7087          * Ok so we may have an even split between what the backrefs think, so
7088          * this is where we use the extent ref to see what it thinks.
7089          */
7090         if (!best) {
7091                 entry = find_entry(&entries, rec->start, rec->nr);
7092                 if (!entry && (!broken_entries || !rec->found_rec)) {
7093                         fprintf(stderr, "Backrefs don't agree with each other "
7094                                 "and extent record doesn't agree with anybody,"
7095                                 " so we can't fix bytenr %Lu bytes %Lu\n",
7096                                 rec->start, rec->nr);
7097                         ret = -EINVAL;
7098                         goto out;
7099                 } else if (!entry) {
7100                         /*
7101                          * Ok our backrefs were broken, we'll assume this is the
7102                          * correct value and add an entry for this range.
7103                          */
7104                         entry = malloc(sizeof(struct extent_entry));
7105                         if (!entry) {
7106                                 ret = -ENOMEM;
7107                                 goto out;
7108                         }
7109                         memset(entry, 0, sizeof(*entry));
7110                         entry->bytenr = rec->start;
7111                         entry->bytes = rec->nr;
7112                         list_add_tail(&entry->list, &entries);
7113                         nr_entries++;
7114                 }
7115                 entry->count++;
7116                 best = find_most_right_entry(&entries);
7117                 if (!best) {
7118                         fprintf(stderr, "Backrefs and extent record evenly "
7119                                 "split on who is right, this is going to "
7120                                 "require user input to fix bytenr %Lu bytes "
7121                                 "%Lu\n", rec->start, rec->nr);
7122                         ret = -EINVAL;
7123                         goto out;
7124                 }
7125         }
7126
7127         /*
7128          * I don't think this can happen currently as we'll abort() if we catch
7129          * this case higher up, but in case somebody removes that we still can't
7130          * deal with it properly here yet, so just bail out of that's the case.
7131          */
7132         if (best->bytenr != rec->start) {
7133                 fprintf(stderr, "Extent start and backref starts don't match, "
7134                         "please use btrfs-image on this file system and send "
7135                         "it to a btrfs developer so they can make fsck fix "
7136                         "this particular case.  bytenr is %Lu, bytes is %Lu\n",
7137                         rec->start, rec->nr);
7138                 ret = -EINVAL;
7139                 goto out;
7140         }
7141
7142         /*
7143          * Ok great we all agreed on an extent record, let's go find the real
7144          * references and fix up the ones that don't match.
7145          */
7146         list_for_each_entry(back, &rec->backrefs, list) {
7147                 if (back->full_backref || !back->is_data)
7148                         continue;
7149
7150                 dback = to_data_backref(back);
7151
7152                 /*
7153                  * Still ignoring backrefs that don't have a real ref attached
7154                  * to them.
7155                  */
7156                 if (dback->found_ref == 0)
7157                         continue;
7158
7159                 if (dback->bytes == best->bytes &&
7160                     dback->disk_bytenr == best->bytenr)
7161                         continue;
7162
7163                 ret = repair_ref(info, path, dback, best);
7164                 if (ret)
7165                         goto out;
7166         }
7167
7168         /*
7169          * Ok we messed with the actual refs, which means we need to drop our
7170          * entire cache and go back and rescan.  I know this is a huge pain and
7171          * adds a lot of extra work, but it's the only way to be safe.  Once all
7172          * the backrefs agree we may not need to do anything to the extent
7173          * record itself.
7174          */
7175         ret = -EAGAIN;
7176 out:
7177         while (!list_empty(&entries)) {
7178                 entry = list_entry(entries.next, struct extent_entry, list);
7179                 list_del_init(&entry->list);
7180                 free(entry);
7181         }
7182         return ret;
7183 }
7184
7185 static int process_duplicates(struct btrfs_root *root,
7186                               struct cache_tree *extent_cache,
7187                               struct extent_record *rec)
7188 {
7189         struct extent_record *good, *tmp;
7190         struct cache_extent *cache;
7191         int ret;
7192
7193         /*
7194          * If we found a extent record for this extent then return, or if we
7195          * have more than one duplicate we are likely going to need to delete
7196          * something.
7197          */
7198         if (rec->found_rec || rec->num_duplicates > 1)
7199                 return 0;
7200
7201         /* Shouldn't happen but just in case */
7202         BUG_ON(!rec->num_duplicates);
7203
7204         /*
7205          * So this happens if we end up with a backref that doesn't match the
7206          * actual extent entry.  So either the backref is bad or the extent
7207          * entry is bad.  Either way we want to have the extent_record actually
7208          * reflect what we found in the extent_tree, so we need to take the
7209          * duplicate out and use that as the extent_record since the only way we
7210          * get a duplicate is if we find a real life BTRFS_EXTENT_ITEM_KEY.
7211          */
7212         remove_cache_extent(extent_cache, &rec->cache);
7213
7214         good = to_extent_record(rec->dups.next);
7215         list_del_init(&good->list);
7216         INIT_LIST_HEAD(&good->backrefs);
7217         INIT_LIST_HEAD(&good->dups);
7218         good->cache.start = good->start;
7219         good->cache.size = good->nr;
7220         good->content_checked = 0;
7221         good->owner_ref_checked = 0;
7222         good->num_duplicates = 0;
7223         good->refs = rec->refs;
7224         list_splice_init(&rec->backrefs, &good->backrefs);
7225         while (1) {
7226                 cache = lookup_cache_extent(extent_cache, good->start,
7227                                             good->nr);
7228                 if (!cache)
7229                         break;
7230                 tmp = container_of(cache, struct extent_record, cache);
7231
7232                 /*
7233                  * If we find another overlapping extent and it's found_rec is
7234                  * set then it's a duplicate and we need to try and delete
7235                  * something.
7236                  */
7237                 if (tmp->found_rec || tmp->num_duplicates > 0) {
7238                         if (list_empty(&good->list))
7239                                 list_add_tail(&good->list,
7240                                               &duplicate_extents);
7241                         good->num_duplicates += tmp->num_duplicates + 1;
7242                         list_splice_init(&tmp->dups, &good->dups);
7243                         list_del_init(&tmp->list);
7244                         list_add_tail(&tmp->list, &good->dups);
7245                         remove_cache_extent(extent_cache, &tmp->cache);
7246                         continue;
7247                 }
7248
7249                 /*
7250                  * Ok we have another non extent item backed extent rec, so lets
7251                  * just add it to this extent and carry on like we did above.
7252                  */
7253                 good->refs += tmp->refs;
7254                 list_splice_init(&tmp->backrefs, &good->backrefs);
7255                 remove_cache_extent(extent_cache, &tmp->cache);
7256                 free(tmp);
7257         }
7258         ret = insert_cache_extent(extent_cache, &good->cache);
7259         BUG_ON(ret);
7260         free(rec);
7261         return good->num_duplicates ? 0 : 1;
7262 }
7263
7264 static int delete_duplicate_records(struct btrfs_root *root,
7265                                     struct extent_record *rec)
7266 {
7267         struct btrfs_trans_handle *trans;
7268         LIST_HEAD(delete_list);
7269         struct btrfs_path path;
7270         struct extent_record *tmp, *good, *n;
7271         int nr_del = 0;
7272         int ret = 0, err;
7273         struct btrfs_key key;
7274
7275         btrfs_init_path(&path);
7276
7277         good = rec;
7278         /* Find the record that covers all of the duplicates. */
7279         list_for_each_entry(tmp, &rec->dups, list) {
7280                 if (good->start < tmp->start)
7281                         continue;
7282                 if (good->nr > tmp->nr)
7283                         continue;
7284
7285                 if (tmp->start + tmp->nr < good->start + good->nr) {
7286                         fprintf(stderr, "Ok we have overlapping extents that "
7287                                 "aren't completely covered by each other, this "
7288                                 "is going to require more careful thought.  "
7289                                 "The extents are [%Lu-%Lu] and [%Lu-%Lu]\n",
7290                                 tmp->start, tmp->nr, good->start, good->nr);
7291                         abort();
7292                 }
7293                 good = tmp;
7294         }
7295
7296         if (good != rec)
7297                 list_add_tail(&rec->list, &delete_list);
7298
7299         list_for_each_entry_safe(tmp, n, &rec->dups, list) {
7300                 if (tmp == good)
7301                         continue;
7302                 list_move_tail(&tmp->list, &delete_list);
7303         }
7304
7305         root = root->fs_info->extent_root;
7306         trans = btrfs_start_transaction(root, 1);
7307         if (IS_ERR(trans)) {
7308                 ret = PTR_ERR(trans);
7309                 goto out;
7310         }
7311
7312         list_for_each_entry(tmp, &delete_list, list) {
7313                 if (tmp->found_rec == 0)
7314                         continue;
7315                 key.objectid = tmp->start;
7316                 key.type = BTRFS_EXTENT_ITEM_KEY;
7317                 key.offset = tmp->nr;
7318
7319                 /* Shouldn't happen but just in case */
7320                 if (tmp->metadata) {
7321                         fprintf(stderr, "Well this shouldn't happen, extent "
7322                                 "record overlaps but is metadata? "
7323                                 "[%Lu, %Lu]\n", tmp->start, tmp->nr);
7324                         abort();
7325                 }
7326
7327                 ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
7328                 if (ret) {
7329                         if (ret > 0)
7330                                 ret = -EINVAL;
7331                         break;
7332                 }
7333                 ret = btrfs_del_item(trans, root, &path);
7334                 if (ret)
7335                         break;
7336                 btrfs_release_path(&path);
7337                 nr_del++;
7338         }
7339         err = btrfs_commit_transaction(trans, root);
7340         if (err && !ret)
7341                 ret = err;
7342 out:
7343         while (!list_empty(&delete_list)) {
7344                 tmp = to_extent_record(delete_list.next);
7345                 list_del_init(&tmp->list);
7346                 if (tmp == rec)
7347                         continue;
7348                 free(tmp);
7349         }
7350
7351         while (!list_empty(&rec->dups)) {
7352                 tmp = to_extent_record(rec->dups.next);
7353                 list_del_init(&tmp->list);
7354                 free(tmp);
7355         }
7356
7357         btrfs_release_path(&path);
7358
7359         if (!ret && !nr_del)
7360                 rec->num_duplicates = 0;
7361
7362         return ret ? ret : nr_del;
7363 }
7364
7365 static int find_possible_backrefs(struct btrfs_fs_info *info,
7366                                   struct btrfs_path *path,
7367                                   struct cache_tree *extent_cache,
7368                                   struct extent_record *rec)
7369 {
7370         struct btrfs_root *root;
7371         struct extent_backref *back;
7372         struct data_backref *dback;
7373         struct cache_extent *cache;
7374         struct btrfs_file_extent_item *fi;
7375         struct btrfs_key key;
7376         u64 bytenr, bytes;
7377         int ret;
7378
7379         list_for_each_entry(back, &rec->backrefs, list) {
7380                 /* Don't care about full backrefs (poor unloved backrefs) */
7381                 if (back->full_backref || !back->is_data)
7382                         continue;
7383
7384                 dback = to_data_backref(back);
7385
7386                 /* We found this one, we don't need to do a lookup */
7387                 if (dback->found_ref)
7388                         continue;
7389
7390                 key.objectid = dback->root;
7391                 key.type = BTRFS_ROOT_ITEM_KEY;
7392                 key.offset = (u64)-1;
7393
7394                 root = btrfs_read_fs_root(info, &key);
7395
7396                 /* No root, definitely a bad ref, skip */
7397                 if (IS_ERR(root) && PTR_ERR(root) == -ENOENT)
7398                         continue;
7399                 /* Other err, exit */
7400                 if (IS_ERR(root))
7401                         return PTR_ERR(root);
7402
7403                 key.objectid = dback->owner;
7404                 key.type = BTRFS_EXTENT_DATA_KEY;
7405                 key.offset = dback->offset;
7406                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7407                 if (ret) {
7408                         btrfs_release_path(path);
7409                         if (ret < 0)
7410                                 return ret;
7411                         /* Didn't find it, we can carry on */
7412                         ret = 0;
7413                         continue;
7414                 }
7415
7416                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
7417                                     struct btrfs_file_extent_item);
7418                 bytenr = btrfs_file_extent_disk_bytenr(path->nodes[0], fi);
7419                 bytes = btrfs_file_extent_disk_num_bytes(path->nodes[0], fi);
7420                 btrfs_release_path(path);
7421                 cache = lookup_cache_extent(extent_cache, bytenr, 1);
7422                 if (cache) {
7423                         struct extent_record *tmp;
7424                         tmp = container_of(cache, struct extent_record, cache);
7425
7426                         /*
7427                          * If we found an extent record for the bytenr for this
7428                          * particular backref then we can't add it to our
7429                          * current extent record.  We only want to add backrefs
7430                          * that don't have a corresponding extent item in the
7431                          * extent tree since they likely belong to this record
7432                          * and we need to fix it if it doesn't match bytenrs.
7433                          */
7434                         if  (tmp->found_rec)
7435                                 continue;
7436                 }
7437
7438                 dback->found_ref += 1;
7439                 dback->disk_bytenr = bytenr;
7440                 dback->bytes = bytes;
7441
7442                 /*
7443                  * Set this so the verify backref code knows not to trust the
7444                  * values in this backref.
7445                  */
7446                 back->broken = 1;
7447         }
7448
7449         return 0;
7450 }
7451
7452 /*
7453  * Record orphan data ref into corresponding root.
7454  *
7455  * Return 0 if the extent item contains data ref and recorded.
7456  * Return 1 if the extent item contains no useful data ref
7457  *   On that case, it may contains only shared_dataref or metadata backref
7458  *   or the file extent exists(this should be handled by the extent bytenr
7459  *   recovery routine)
7460  * Return <0 if something goes wrong.
7461  */
7462 static int record_orphan_data_extents(struct btrfs_fs_info *fs_info,
7463                                       struct extent_record *rec)
7464 {
7465         struct btrfs_key key;
7466         struct btrfs_root *dest_root;
7467         struct extent_backref *back;
7468         struct data_backref *dback;
7469         struct orphan_data_extent *orphan;
7470         struct btrfs_path path;
7471         int recorded_data_ref = 0;
7472         int ret = 0;
7473
7474         if (rec->metadata)
7475                 return 1;
7476         btrfs_init_path(&path);
7477         list_for_each_entry(back, &rec->backrefs, list) {
7478                 if (back->full_backref || !back->is_data ||
7479                     !back->found_extent_tree)
7480                         continue;
7481                 dback = to_data_backref(back);
7482                 if (dback->found_ref)
7483                         continue;
7484                 key.objectid = dback->root;
7485                 key.type = BTRFS_ROOT_ITEM_KEY;
7486                 key.offset = (u64)-1;
7487
7488                 dest_root = btrfs_read_fs_root(fs_info, &key);
7489
7490                 /* For non-exist root we just skip it */
7491                 if (IS_ERR(dest_root) || !dest_root)
7492                         continue;
7493
7494                 key.objectid = dback->owner;
7495                 key.type = BTRFS_EXTENT_DATA_KEY;
7496                 key.offset = dback->offset;
7497
7498                 ret = btrfs_search_slot(NULL, dest_root, &key, &path, 0, 0);
7499                 btrfs_release_path(&path);
7500                 /*
7501                  * For ret < 0, it's OK since the fs-tree may be corrupted,
7502                  * we need to record it for inode/file extent rebuild.
7503                  * For ret > 0, we record it only for file extent rebuild.
7504                  * For ret == 0, the file extent exists but only bytenr
7505                  * mismatch, let the original bytenr fix routine to handle,
7506                  * don't record it.
7507                  */
7508                 if (ret == 0)
7509                         continue;
7510                 ret = 0;
7511                 orphan = malloc(sizeof(*orphan));
7512                 if (!orphan) {
7513                         ret = -ENOMEM;
7514                         goto out;
7515                 }
7516                 INIT_LIST_HEAD(&orphan->list);
7517                 orphan->root = dback->root;
7518                 orphan->objectid = dback->owner;
7519                 orphan->offset = dback->offset;
7520                 orphan->disk_bytenr = rec->cache.start;
7521                 orphan->disk_len = rec->cache.size;
7522                 list_add(&dest_root->orphan_data_extents, &orphan->list);
7523                 recorded_data_ref = 1;
7524         }
7525 out:
7526         btrfs_release_path(&path);
7527         if (!ret)
7528                 return !recorded_data_ref;
7529         else
7530                 return ret;
7531 }
7532
7533 /*
7534  * when an incorrect extent item is found, this will delete
7535  * all of the existing entries for it and recreate them
7536  * based on what the tree scan found.
7537  */
7538 static int fixup_extent_refs(struct btrfs_fs_info *info,
7539                              struct cache_tree *extent_cache,
7540                              struct extent_record *rec)
7541 {
7542         struct btrfs_trans_handle *trans = NULL;
7543         int ret;
7544         struct btrfs_path path;
7545         struct list_head *cur = rec->backrefs.next;
7546         struct cache_extent *cache;
7547         struct extent_backref *back;
7548         int allocated = 0;
7549         u64 flags = 0;
7550
7551         if (rec->flag_block_full_backref)
7552                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7553
7554         btrfs_init_path(&path);
7555         if (rec->refs != rec->extent_item_refs && !rec->metadata) {
7556                 /*
7557                  * Sometimes the backrefs themselves are so broken they don't
7558                  * get attached to any meaningful rec, so first go back and
7559                  * check any of our backrefs that we couldn't find and throw
7560                  * them into the list if we find the backref so that
7561                  * verify_backrefs can figure out what to do.
7562                  */
7563                 ret = find_possible_backrefs(info, &path, extent_cache, rec);
7564                 if (ret < 0)
7565                         goto out;
7566         }
7567
7568         /* step one, make sure all of the backrefs agree */
7569         ret = verify_backrefs(info, &path, rec);
7570         if (ret < 0)
7571                 goto out;
7572
7573         trans = btrfs_start_transaction(info->extent_root, 1);
7574         if (IS_ERR(trans)) {
7575                 ret = PTR_ERR(trans);
7576                 goto out;
7577         }
7578
7579         /* step two, delete all the existing records */
7580         ret = delete_extent_records(trans, info->extent_root, &path,
7581                                     rec->start, rec->max_size);
7582
7583         if (ret < 0)
7584                 goto out;
7585
7586         /* was this block corrupt?  If so, don't add references to it */
7587         cache = lookup_cache_extent(info->corrupt_blocks,
7588                                     rec->start, rec->max_size);
7589         if (cache) {
7590                 ret = 0;
7591                 goto out;
7592         }
7593
7594         /* step three, recreate all the refs we did find */
7595         while(cur != &rec->backrefs) {
7596                 back = to_extent_backref(cur);
7597                 cur = cur->next;
7598
7599                 /*
7600                  * if we didn't find any references, don't create a
7601                  * new extent record
7602                  */
7603                 if (!back->found_ref)
7604                         continue;
7605
7606                 rec->bad_full_backref = 0;
7607                 ret = record_extent(trans, info, &path, rec, back, allocated, flags);
7608                 allocated = 1;
7609
7610                 if (ret)
7611                         goto out;
7612         }
7613 out:
7614         if (trans) {
7615                 int err = btrfs_commit_transaction(trans, info->extent_root);
7616                 if (!ret)
7617                         ret = err;
7618         }
7619
7620         btrfs_release_path(&path);
7621         return ret;
7622 }
7623
7624 static int fixup_extent_flags(struct btrfs_fs_info *fs_info,
7625                               struct extent_record *rec)
7626 {
7627         struct btrfs_trans_handle *trans;
7628         struct btrfs_root *root = fs_info->extent_root;
7629         struct btrfs_path path;
7630         struct btrfs_extent_item *ei;
7631         struct btrfs_key key;
7632         u64 flags;
7633         int ret = 0;
7634
7635         key.objectid = rec->start;
7636         if (rec->metadata) {
7637                 key.type = BTRFS_METADATA_ITEM_KEY;
7638                 key.offset = rec->info_level;
7639         } else {
7640                 key.type = BTRFS_EXTENT_ITEM_KEY;
7641                 key.offset = rec->max_size;
7642         }
7643
7644         trans = btrfs_start_transaction(root, 0);
7645         if (IS_ERR(trans))
7646                 return PTR_ERR(trans);
7647
7648         btrfs_init_path(&path);
7649         ret = btrfs_search_slot(trans, root, &key, &path, 0, 1);
7650         if (ret < 0) {
7651                 btrfs_release_path(&path);
7652                 btrfs_commit_transaction(trans, root);
7653                 return ret;
7654         } else if (ret) {
7655                 fprintf(stderr, "Didn't find extent for %llu\n",
7656                         (unsigned long long)rec->start);
7657                 btrfs_release_path(&path);
7658                 btrfs_commit_transaction(trans, root);
7659                 return -ENOENT;
7660         }
7661
7662         ei = btrfs_item_ptr(path.nodes[0], path.slots[0],
7663                             struct btrfs_extent_item);
7664         flags = btrfs_extent_flags(path.nodes[0], ei);
7665         if (rec->flag_block_full_backref) {
7666                 fprintf(stderr, "setting full backref on %llu\n",
7667                         (unsigned long long)key.objectid);
7668                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7669         } else {
7670                 fprintf(stderr, "clearing full backref on %llu\n",
7671                         (unsigned long long)key.objectid);
7672                 flags &= ~BTRFS_BLOCK_FLAG_FULL_BACKREF;
7673         }
7674         btrfs_set_extent_flags(path.nodes[0], ei, flags);
7675         btrfs_mark_buffer_dirty(path.nodes[0]);
7676         btrfs_release_path(&path);
7677         return btrfs_commit_transaction(trans, root);
7678 }
7679
7680 /* right now we only prune from the extent allocation tree */
7681 static int prune_one_block(struct btrfs_trans_handle *trans,
7682                            struct btrfs_fs_info *info,
7683                            struct btrfs_corrupt_block *corrupt)
7684 {
7685         int ret;
7686         struct btrfs_path path;
7687         struct extent_buffer *eb;
7688         u64 found;
7689         int slot;
7690         int nritems;
7691         int level = corrupt->level + 1;
7692
7693         btrfs_init_path(&path);
7694 again:
7695         /* we want to stop at the parent to our busted block */
7696         path.lowest_level = level;
7697
7698         ret = btrfs_search_slot(trans, info->extent_root,
7699                                 &corrupt->key, &path, -1, 1);
7700
7701         if (ret < 0)
7702                 goto out;
7703
7704         eb = path.nodes[level];
7705         if (!eb) {
7706                 ret = -ENOENT;
7707                 goto out;
7708         }
7709
7710         /*
7711          * hopefully the search gave us the block we want to prune,
7712          * lets try that first
7713          */
7714         slot = path.slots[level];
7715         found =  btrfs_node_blockptr(eb, slot);
7716         if (found == corrupt->cache.start)
7717                 goto del_ptr;
7718
7719         nritems = btrfs_header_nritems(eb);
7720
7721         /* the search failed, lets scan this node and hope we find it */
7722         for (slot = 0; slot < nritems; slot++) {
7723                 found =  btrfs_node_blockptr(eb, slot);
7724                 if (found == corrupt->cache.start)
7725                         goto del_ptr;
7726         }
7727         /*
7728          * we couldn't find the bad block.  TODO, search all the nodes for pointers
7729          * to this block
7730          */
7731         if (eb == info->extent_root->node) {
7732                 ret = -ENOENT;
7733                 goto out;
7734         } else {
7735                 level++;
7736                 btrfs_release_path(&path);
7737                 goto again;
7738         }
7739
7740 del_ptr:
7741         printk("deleting pointer to block %Lu\n", corrupt->cache.start);
7742         ret = btrfs_del_ptr(trans, info->extent_root, &path, level, slot);
7743
7744 out:
7745         btrfs_release_path(&path);
7746         return ret;
7747 }
7748
7749 static int prune_corrupt_blocks(struct btrfs_fs_info *info)
7750 {
7751         struct btrfs_trans_handle *trans = NULL;
7752         struct cache_extent *cache;
7753         struct btrfs_corrupt_block *corrupt;
7754
7755         while (1) {
7756                 cache = search_cache_extent(info->corrupt_blocks, 0);
7757                 if (!cache)
7758                         break;
7759                 if (!trans) {
7760                         trans = btrfs_start_transaction(info->extent_root, 1);
7761                         if (IS_ERR(trans))
7762                                 return PTR_ERR(trans);
7763                 }
7764                 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
7765                 prune_one_block(trans, info, corrupt);
7766                 remove_cache_extent(info->corrupt_blocks, cache);
7767         }
7768         if (trans)
7769                 return btrfs_commit_transaction(trans, info->extent_root);
7770         return 0;
7771 }
7772
7773 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info)
7774 {
7775         struct btrfs_block_group_cache *cache;
7776         u64 start, end;
7777         int ret;
7778
7779         while (1) {
7780                 ret = find_first_extent_bit(&fs_info->free_space_cache, 0,
7781                                             &start, &end, EXTENT_DIRTY);
7782                 if (ret)
7783                         break;
7784                 clear_extent_dirty(&fs_info->free_space_cache, start, end,
7785                                    GFP_NOFS);
7786         }
7787
7788         start = 0;
7789         while (1) {
7790                 cache = btrfs_lookup_first_block_group(fs_info, start);
7791                 if (!cache)
7792                         break;
7793                 if (cache->cached)
7794                         cache->cached = 0;
7795                 start = cache->key.objectid + cache->key.offset;
7796         }
7797 }
7798
7799 static int check_extent_refs(struct btrfs_root *root,
7800                              struct cache_tree *extent_cache)
7801 {
7802         struct extent_record *rec;
7803         struct cache_extent *cache;
7804         int err = 0;
7805         int ret = 0;
7806         int fixed = 0;
7807         int had_dups = 0;
7808         int recorded = 0;
7809
7810         if (repair) {
7811                 /*
7812                  * if we're doing a repair, we have to make sure
7813                  * we don't allocate from the problem extents.
7814                  * In the worst case, this will be all the
7815                  * extents in the FS
7816                  */
7817                 cache = search_cache_extent(extent_cache, 0);
7818                 while(cache) {
7819                         rec = container_of(cache, struct extent_record, cache);
7820                         set_extent_dirty(root->fs_info->excluded_extents,
7821                                          rec->start,
7822                                          rec->start + rec->max_size - 1,
7823                                          GFP_NOFS);
7824                         cache = next_cache_extent(cache);
7825                 }
7826
7827                 /* pin down all the corrupted blocks too */
7828                 cache = search_cache_extent(root->fs_info->corrupt_blocks, 0);
7829                 while(cache) {
7830                         set_extent_dirty(root->fs_info->excluded_extents,
7831                                          cache->start,
7832                                          cache->start + cache->size - 1,
7833                                          GFP_NOFS);
7834                         cache = next_cache_extent(cache);
7835                 }
7836                 prune_corrupt_blocks(root->fs_info);
7837                 reset_cached_block_groups(root->fs_info);
7838         }
7839
7840         reset_cached_block_groups(root->fs_info);
7841
7842         /*
7843          * We need to delete any duplicate entries we find first otherwise we
7844          * could mess up the extent tree when we have backrefs that actually
7845          * belong to a different extent item and not the weird duplicate one.
7846          */
7847         while (repair && !list_empty(&duplicate_extents)) {
7848                 rec = to_extent_record(duplicate_extents.next);
7849                 list_del_init(&rec->list);
7850
7851                 /* Sometimes we can find a backref before we find an actual
7852                  * extent, so we need to process it a little bit to see if there
7853                  * truly are multiple EXTENT_ITEM_KEY's for the same range, or
7854                  * if this is a backref screwup.  If we need to delete stuff
7855                  * process_duplicates() will return 0, otherwise it will return
7856                  * 1 and we
7857                  */
7858                 if (process_duplicates(root, extent_cache, rec))
7859                         continue;
7860                 ret = delete_duplicate_records(root, rec);
7861                 if (ret < 0)
7862                         return ret;
7863                 /*
7864                  * delete_duplicate_records will return the number of entries
7865                  * deleted, so if it's greater than 0 then we know we actually
7866                  * did something and we need to remove.
7867                  */
7868                 if (ret)
7869                         had_dups = 1;
7870         }
7871
7872         if (had_dups)
7873                 return -EAGAIN;
7874
7875         while(1) {
7876                 int cur_err = 0;
7877
7878                 fixed = 0;
7879                 recorded = 0;
7880                 cache = search_cache_extent(extent_cache, 0);
7881                 if (!cache)
7882                         break;
7883                 rec = container_of(cache, struct extent_record, cache);
7884                 if (rec->num_duplicates) {
7885                         fprintf(stderr, "extent item %llu has multiple extent "
7886                                 "items\n", (unsigned long long)rec->start);
7887                         err = 1;
7888                         cur_err = 1;
7889                 }
7890
7891                 if (rec->refs != rec->extent_item_refs) {
7892                         fprintf(stderr, "ref mismatch on [%llu %llu] ",
7893                                 (unsigned long long)rec->start,
7894                                 (unsigned long long)rec->nr);
7895                         fprintf(stderr, "extent item %llu, found %llu\n",
7896                                 (unsigned long long)rec->extent_item_refs,
7897                                 (unsigned long long)rec->refs);
7898                         ret = record_orphan_data_extents(root->fs_info, rec);
7899                         if (ret < 0)
7900                                 goto repair_abort;
7901                         if (ret == 0) {
7902                                 recorded = 1;
7903                         } else {
7904                                 /*
7905                                  * we can't use the extent to repair file
7906                                  * extent, let the fallback method handle it.
7907                                  */
7908                                 if (!fixed && repair) {
7909                                         ret = fixup_extent_refs(
7910                                                         root->fs_info,
7911                                                         extent_cache, rec);
7912                                         if (ret)
7913                                                 goto repair_abort;
7914                                         fixed = 1;
7915                                 }
7916                         }
7917                         err = 1;
7918                         cur_err = 1;
7919                 }
7920                 if (all_backpointers_checked(rec, 1)) {
7921                         fprintf(stderr, "backpointer mismatch on [%llu %llu]\n",
7922                                 (unsigned long long)rec->start,
7923                                 (unsigned long long)rec->nr);
7924
7925                         if (!fixed && !recorded && repair) {
7926                                 ret = fixup_extent_refs(root->fs_info,
7927                                                         extent_cache, rec);
7928                                 if (ret)
7929                                         goto repair_abort;
7930                                 fixed = 1;
7931                         }
7932                         cur_err = 1;
7933                         err = 1;
7934                 }
7935                 if (!rec->owner_ref_checked) {
7936                         fprintf(stderr, "owner ref check failed [%llu %llu]\n",
7937                                 (unsigned long long)rec->start,
7938                                 (unsigned long long)rec->nr);
7939                         if (!fixed && !recorded && repair) {
7940                                 ret = fixup_extent_refs(root->fs_info,
7941                                                         extent_cache, rec);
7942                                 if (ret)
7943                                         goto repair_abort;
7944                                 fixed = 1;
7945                         }
7946                         err = 1;
7947                         cur_err = 1;
7948                 }
7949                 if (rec->bad_full_backref) {
7950                         fprintf(stderr, "bad full backref, on [%llu]\n",
7951                                 (unsigned long long)rec->start);
7952                         if (repair) {
7953                                 ret = fixup_extent_flags(root->fs_info, rec);
7954                                 if (ret)
7955                                         goto repair_abort;
7956                                 fixed = 1;
7957                         }
7958                         err = 1;
7959                         cur_err = 1;
7960                 }
7961                 /*
7962                  * Although it's not a extent ref's problem, we reuse this
7963                  * routine for error reporting.
7964                  * No repair function yet.
7965                  */
7966                 if (rec->crossing_stripes) {
7967                         fprintf(stderr,
7968                                 "bad metadata [%llu, %llu) crossing stripe boundary\n",
7969                                 rec->start, rec->start + rec->max_size);
7970                         err = 1;
7971                         cur_err = 1;
7972                 }
7973
7974                 if (rec->wrong_chunk_type) {
7975                         fprintf(stderr,
7976                                 "bad extent [%llu, %llu), type mismatch with chunk\n",
7977                                 rec->start, rec->start + rec->max_size);
7978                         err = 1;
7979                         cur_err = 1;
7980                 }
7981
7982                 remove_cache_extent(extent_cache, cache);
7983                 free_all_extent_backrefs(rec);
7984                 if (!init_extent_tree && repair && (!cur_err || fixed))
7985                         clear_extent_dirty(root->fs_info->excluded_extents,
7986                                            rec->start,
7987                                            rec->start + rec->max_size - 1,
7988                                            GFP_NOFS);
7989                 free(rec);
7990         }
7991 repair_abort:
7992         if (repair) {
7993                 if (ret && ret != -EAGAIN) {
7994                         fprintf(stderr, "failed to repair damaged filesystem, aborting\n");
7995                         exit(1);
7996                 } else if (!ret) {
7997                         struct btrfs_trans_handle *trans;
7998
7999                         root = root->fs_info->extent_root;
8000                         trans = btrfs_start_transaction(root, 1);
8001                         if (IS_ERR(trans)) {
8002                                 ret = PTR_ERR(trans);
8003                                 goto repair_abort;
8004                         }
8005
8006                         btrfs_fix_block_accounting(trans, root);
8007                         ret = btrfs_commit_transaction(trans, root);
8008                         if (ret)
8009                                 goto repair_abort;
8010                 }
8011                 if (err)
8012                         fprintf(stderr, "repaired damaged extent references\n");
8013                 return ret;
8014         }
8015         return err;
8016 }
8017
8018 u64 calc_stripe_length(u64 type, u64 length, int num_stripes)
8019 {
8020         u64 stripe_size;
8021
8022         if (type & BTRFS_BLOCK_GROUP_RAID0) {
8023                 stripe_size = length;
8024                 stripe_size /= num_stripes;
8025         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
8026                 stripe_size = length * 2;
8027                 stripe_size /= num_stripes;
8028         } else if (type & BTRFS_BLOCK_GROUP_RAID5) {
8029                 stripe_size = length;
8030                 stripe_size /= (num_stripes - 1);
8031         } else if (type & BTRFS_BLOCK_GROUP_RAID6) {
8032                 stripe_size = length;
8033                 stripe_size /= (num_stripes - 2);
8034         } else {
8035                 stripe_size = length;
8036         }
8037         return stripe_size;
8038 }
8039
8040 /*
8041  * Check the chunk with its block group/dev list ref:
8042  * Return 0 if all refs seems valid.
8043  * Return 1 if part of refs seems valid, need later check for rebuild ref
8044  * like missing block group and needs to search extent tree to rebuild them.
8045  * Return -1 if essential refs are missing and unable to rebuild.
8046  */
8047 static int check_chunk_refs(struct chunk_record *chunk_rec,
8048                             struct block_group_tree *block_group_cache,
8049                             struct device_extent_tree *dev_extent_cache,
8050                             int silent)
8051 {
8052         struct cache_extent *block_group_item;
8053         struct block_group_record *block_group_rec;
8054         struct cache_extent *dev_extent_item;
8055         struct device_extent_record *dev_extent_rec;
8056         u64 devid;
8057         u64 offset;
8058         u64 length;
8059         int metadump_v2 = 0;
8060         int i;
8061         int ret = 0;
8062
8063         block_group_item = lookup_cache_extent(&block_group_cache->tree,
8064                                                chunk_rec->offset,
8065                                                chunk_rec->length);
8066         if (block_group_item) {
8067                 block_group_rec = container_of(block_group_item,
8068                                                struct block_group_record,
8069                                                cache);
8070                 if (chunk_rec->length != block_group_rec->offset ||
8071                     chunk_rec->offset != block_group_rec->objectid ||
8072                     (!metadump_v2 &&
8073                      chunk_rec->type_flags != block_group_rec->flags)) {
8074                         if (!silent)
8075                                 fprintf(stderr,
8076                                         "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) mismatch with block group[%llu, %u, %llu]: offset(%llu), objectid(%llu), flags(%llu)\n",
8077                                         chunk_rec->objectid,
8078                                         chunk_rec->type,
8079                                         chunk_rec->offset,
8080                                         chunk_rec->length,
8081                                         chunk_rec->offset,
8082                                         chunk_rec->type_flags,
8083                                         block_group_rec->objectid,
8084                                         block_group_rec->type,
8085                                         block_group_rec->offset,
8086                                         block_group_rec->offset,
8087                                         block_group_rec->objectid,
8088                                         block_group_rec->flags);
8089                         ret = -1;
8090                 } else {
8091                         list_del_init(&block_group_rec->list);
8092                         chunk_rec->bg_rec = block_group_rec;
8093                 }
8094         } else {
8095                 if (!silent)
8096                         fprintf(stderr,
8097                                 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) is not found in block group\n",
8098                                 chunk_rec->objectid,
8099                                 chunk_rec->type,
8100                                 chunk_rec->offset,
8101                                 chunk_rec->length,
8102                                 chunk_rec->offset,
8103                                 chunk_rec->type_flags);
8104                 ret = 1;
8105         }
8106
8107         if (metadump_v2)
8108                 return ret;
8109
8110         length = calc_stripe_length(chunk_rec->type_flags, chunk_rec->length,
8111                                     chunk_rec->num_stripes);
8112         for (i = 0; i < chunk_rec->num_stripes; ++i) {
8113                 devid = chunk_rec->stripes[i].devid;
8114                 offset = chunk_rec->stripes[i].offset;
8115                 dev_extent_item = lookup_cache_extent2(&dev_extent_cache->tree,
8116                                                        devid, offset, length);
8117                 if (dev_extent_item) {
8118                         dev_extent_rec = container_of(dev_extent_item,
8119                                                 struct device_extent_record,
8120                                                 cache);
8121                         if (dev_extent_rec->objectid != devid ||
8122                             dev_extent_rec->offset != offset ||
8123                             dev_extent_rec->chunk_offset != chunk_rec->offset ||
8124                             dev_extent_rec->length != length) {
8125                                 if (!silent)
8126                                         fprintf(stderr,
8127                                                 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] dismatch dev extent[%llu, %llu, %llu]\n",
8128                                                 chunk_rec->objectid,
8129                                                 chunk_rec->type,
8130                                                 chunk_rec->offset,
8131                                                 chunk_rec->stripes[i].devid,
8132                                                 chunk_rec->stripes[i].offset,
8133                                                 dev_extent_rec->objectid,
8134                                                 dev_extent_rec->offset,
8135                                                 dev_extent_rec->length);
8136                                 ret = -1;
8137                         } else {
8138                                 list_move(&dev_extent_rec->chunk_list,
8139                                           &chunk_rec->dextents);
8140                         }
8141                 } else {
8142                         if (!silent)
8143                                 fprintf(stderr,
8144                                         "Chunk[%llu, %u, %llu] stripe[%llu, %llu] is not found in dev extent\n",
8145                                         chunk_rec->objectid,
8146                                         chunk_rec->type,
8147                                         chunk_rec->offset,
8148                                         chunk_rec->stripes[i].devid,
8149                                         chunk_rec->stripes[i].offset);
8150                         ret = -1;
8151                 }
8152         }
8153         return ret;
8154 }
8155
8156 /* check btrfs_chunk -> btrfs_dev_extent / btrfs_block_group_item */
8157 int check_chunks(struct cache_tree *chunk_cache,
8158                  struct block_group_tree *block_group_cache,
8159                  struct device_extent_tree *dev_extent_cache,
8160                  struct list_head *good, struct list_head *bad,
8161                  struct list_head *rebuild, int silent)
8162 {
8163         struct cache_extent *chunk_item;
8164         struct chunk_record *chunk_rec;
8165         struct block_group_record *bg_rec;
8166         struct device_extent_record *dext_rec;
8167         int err;
8168         int ret = 0;
8169
8170         chunk_item = first_cache_extent(chunk_cache);
8171         while (chunk_item) {
8172                 chunk_rec = container_of(chunk_item, struct chunk_record,
8173                                          cache);
8174                 err = check_chunk_refs(chunk_rec, block_group_cache,
8175                                        dev_extent_cache, silent);
8176                 if (err < 0)
8177                         ret = err;
8178                 if (err == 0 && good)
8179                         list_add_tail(&chunk_rec->list, good);
8180                 if (err > 0 && rebuild)
8181                         list_add_tail(&chunk_rec->list, rebuild);
8182                 if (err < 0 && bad)
8183                         list_add_tail(&chunk_rec->list, bad);
8184                 chunk_item = next_cache_extent(chunk_item);
8185         }
8186
8187         list_for_each_entry(bg_rec, &block_group_cache->block_groups, list) {
8188                 if (!silent)
8189                         fprintf(stderr,
8190                                 "Block group[%llu, %llu] (flags = %llu) didn't find the relative chunk.\n",
8191                                 bg_rec->objectid,
8192                                 bg_rec->offset,
8193                                 bg_rec->flags);
8194                 if (!ret)
8195                         ret = 1;
8196         }
8197
8198         list_for_each_entry(dext_rec, &dev_extent_cache->no_chunk_orphans,
8199                             chunk_list) {
8200                 if (!silent)
8201                         fprintf(stderr,
8202                                 "Device extent[%llu, %llu, %llu] didn't find the relative chunk.\n",
8203                                 dext_rec->objectid,
8204                                 dext_rec->offset,
8205                                 dext_rec->length);
8206                 if (!ret)
8207                         ret = 1;
8208         }
8209         return ret;
8210 }
8211
8212
8213 static int check_device_used(struct device_record *dev_rec,
8214                              struct device_extent_tree *dext_cache)
8215 {
8216         struct cache_extent *cache;
8217         struct device_extent_record *dev_extent_rec;
8218         u64 total_byte = 0;
8219
8220         cache = search_cache_extent2(&dext_cache->tree, dev_rec->devid, 0);
8221         while (cache) {
8222                 dev_extent_rec = container_of(cache,
8223                                               struct device_extent_record,
8224                                               cache);
8225                 if (dev_extent_rec->objectid != dev_rec->devid)
8226                         break;
8227
8228                 list_del_init(&dev_extent_rec->device_list);
8229                 total_byte += dev_extent_rec->length;
8230                 cache = next_cache_extent(cache);
8231         }
8232
8233         if (total_byte != dev_rec->byte_used) {
8234                 fprintf(stderr,
8235                         "Dev extent's total-byte(%llu) is not equal to byte-used(%llu) in dev[%llu, %u, %llu]\n",
8236                         total_byte, dev_rec->byte_used, dev_rec->objectid,
8237                         dev_rec->type, dev_rec->offset);
8238                 return -1;
8239         } else {
8240                 return 0;
8241         }
8242 }
8243
8244 /* check btrfs_dev_item -> btrfs_dev_extent */
8245 static int check_devices(struct rb_root *dev_cache,
8246                          struct device_extent_tree *dev_extent_cache)
8247 {
8248         struct rb_node *dev_node;
8249         struct device_record *dev_rec;
8250         struct device_extent_record *dext_rec;
8251         int err;
8252         int ret = 0;
8253
8254         dev_node = rb_first(dev_cache);
8255         while (dev_node) {
8256                 dev_rec = container_of(dev_node, struct device_record, node);
8257                 err = check_device_used(dev_rec, dev_extent_cache);
8258                 if (err)
8259                         ret = err;
8260
8261                 dev_node = rb_next(dev_node);
8262         }
8263         list_for_each_entry(dext_rec, &dev_extent_cache->no_device_orphans,
8264                             device_list) {
8265                 fprintf(stderr,
8266                         "Device extent[%llu, %llu, %llu] didn't find its device.\n",
8267                         dext_rec->objectid, dext_rec->offset, dext_rec->length);
8268                 if (!ret)
8269                         ret = 1;
8270         }
8271         return ret;
8272 }
8273
8274 static int add_root_item_to_list(struct list_head *head,
8275                                   u64 objectid, u64 bytenr, u64 last_snapshot,
8276                                   u8 level, u8 drop_level,
8277                                   int level_size, struct btrfs_key *drop_key)
8278 {
8279
8280         struct root_item_record *ri_rec;
8281         ri_rec = malloc(sizeof(*ri_rec));
8282         if (!ri_rec)
8283                 return -ENOMEM;
8284         ri_rec->bytenr = bytenr;
8285         ri_rec->objectid = objectid;
8286         ri_rec->level = level;
8287         ri_rec->level_size = level_size;
8288         ri_rec->drop_level = drop_level;
8289         ri_rec->last_snapshot = last_snapshot;
8290         if (drop_key)
8291                 memcpy(&ri_rec->drop_key, drop_key, sizeof(*drop_key));
8292         list_add_tail(&ri_rec->list, head);
8293
8294         return 0;
8295 }
8296
8297 static void free_root_item_list(struct list_head *list)
8298 {
8299         struct root_item_record *ri_rec;
8300
8301         while (!list_empty(list)) {
8302                 ri_rec = list_first_entry(list, struct root_item_record,
8303                                           list);
8304                 list_del_init(&ri_rec->list);
8305                 free(ri_rec);
8306         }
8307 }
8308
8309 static int deal_root_from_list(struct list_head *list,
8310                                struct btrfs_root *root,
8311                                struct block_info *bits,
8312                                int bits_nr,
8313                                struct cache_tree *pending,
8314                                struct cache_tree *seen,
8315                                struct cache_tree *reada,
8316                                struct cache_tree *nodes,
8317                                struct cache_tree *extent_cache,
8318                                struct cache_tree *chunk_cache,
8319                                struct rb_root *dev_cache,
8320                                struct block_group_tree *block_group_cache,
8321                                struct device_extent_tree *dev_extent_cache)
8322 {
8323         int ret = 0;
8324         u64 last;
8325
8326         while (!list_empty(list)) {
8327                 struct root_item_record *rec;
8328                 struct extent_buffer *buf;
8329                 rec = list_entry(list->next,
8330                                  struct root_item_record, list);
8331                 last = 0;
8332                 buf = read_tree_block(root->fs_info->tree_root,
8333                                       rec->bytenr, rec->level_size, 0);
8334                 if (!extent_buffer_uptodate(buf)) {
8335                         free_extent_buffer(buf);
8336                         ret = -EIO;
8337                         break;
8338                 }
8339                 ret = add_root_to_pending(buf, extent_cache, pending,
8340                                     seen, nodes, rec->objectid);
8341                 if (ret < 0)
8342                         break;
8343                 /*
8344                  * To rebuild extent tree, we need deal with snapshot
8345                  * one by one, otherwise we deal with node firstly which
8346                  * can maximize readahead.
8347                  */
8348                 while (1) {
8349                         ret = run_next_block(root, bits, bits_nr, &last,
8350                                              pending, seen, reada, nodes,
8351                                              extent_cache, chunk_cache,
8352                                              dev_cache, block_group_cache,
8353                                              dev_extent_cache, rec);
8354                         if (ret != 0)
8355                                 break;
8356                 }
8357                 free_extent_buffer(buf);
8358                 list_del(&rec->list);
8359                 free(rec);
8360                 if (ret < 0)
8361                         break;
8362         }
8363         while (ret >= 0) {
8364                 ret = run_next_block(root, bits, bits_nr, &last, pending, seen,
8365                                      reada, nodes, extent_cache, chunk_cache,
8366                                      dev_cache, block_group_cache,
8367                                      dev_extent_cache, NULL);
8368                 if (ret != 0) {
8369                         if (ret > 0)
8370                                 ret = 0;
8371                         break;
8372                 }
8373         }
8374         return ret;
8375 }
8376
8377 static int check_chunks_and_extents(struct btrfs_root *root)
8378 {
8379         struct rb_root dev_cache;
8380         struct cache_tree chunk_cache;
8381         struct block_group_tree block_group_cache;
8382         struct device_extent_tree dev_extent_cache;
8383         struct cache_tree extent_cache;
8384         struct cache_tree seen;
8385         struct cache_tree pending;
8386         struct cache_tree reada;
8387         struct cache_tree nodes;
8388         struct extent_io_tree excluded_extents;
8389         struct cache_tree corrupt_blocks;
8390         struct btrfs_path path;
8391         struct btrfs_key key;
8392         struct btrfs_key found_key;
8393         int ret, err = 0;
8394         struct block_info *bits;
8395         int bits_nr;
8396         struct extent_buffer *leaf;
8397         int slot;
8398         struct btrfs_root_item ri;
8399         struct list_head dropping_trees;
8400         struct list_head normal_trees;
8401         struct btrfs_root *root1;
8402         u64 objectid;
8403         u32 level_size;
8404         u8 level;
8405
8406         dev_cache = RB_ROOT;
8407         cache_tree_init(&chunk_cache);
8408         block_group_tree_init(&block_group_cache);
8409         device_extent_tree_init(&dev_extent_cache);
8410
8411         cache_tree_init(&extent_cache);
8412         cache_tree_init(&seen);
8413         cache_tree_init(&pending);
8414         cache_tree_init(&nodes);
8415         cache_tree_init(&reada);
8416         cache_tree_init(&corrupt_blocks);
8417         extent_io_tree_init(&excluded_extents);
8418         INIT_LIST_HEAD(&dropping_trees);
8419         INIT_LIST_HEAD(&normal_trees);
8420
8421         if (repair) {
8422                 root->fs_info->excluded_extents = &excluded_extents;
8423                 root->fs_info->fsck_extent_cache = &extent_cache;
8424                 root->fs_info->free_extent_hook = free_extent_hook;
8425                 root->fs_info->corrupt_blocks = &corrupt_blocks;
8426         }
8427
8428         bits_nr = 1024;
8429         bits = malloc(bits_nr * sizeof(struct block_info));
8430         if (!bits) {
8431                 perror("malloc");
8432                 exit(1);
8433         }
8434
8435         if (ctx.progress_enabled) {
8436                 ctx.tp = TASK_EXTENTS;
8437                 task_start(ctx.info);
8438         }
8439
8440 again:
8441         root1 = root->fs_info->tree_root;
8442         level = btrfs_header_level(root1->node);
8443         ret = add_root_item_to_list(&normal_trees, root1->root_key.objectid,
8444                                     root1->node->start, 0, level, 0,
8445                                     root1->nodesize, NULL);
8446         if (ret < 0)
8447                 goto out;
8448         root1 = root->fs_info->chunk_root;
8449         level = btrfs_header_level(root1->node);
8450         ret = add_root_item_to_list(&normal_trees, root1->root_key.objectid,
8451                                     root1->node->start, 0, level, 0,
8452                                     root1->nodesize, NULL);
8453         if (ret < 0)
8454                 goto out;
8455         btrfs_init_path(&path);
8456         key.offset = 0;
8457         key.objectid = 0;
8458         key.type = BTRFS_ROOT_ITEM_KEY;
8459         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
8460                                         &key, &path, 0, 0);
8461         if (ret < 0)
8462                 goto out;
8463         while(1) {
8464                 leaf = path.nodes[0];
8465                 slot = path.slots[0];
8466                 if (slot >= btrfs_header_nritems(path.nodes[0])) {
8467                         ret = btrfs_next_leaf(root, &path);
8468                         if (ret != 0)
8469                                 break;
8470                         leaf = path.nodes[0];
8471                         slot = path.slots[0];
8472                 }
8473                 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
8474                 if (found_key.type == BTRFS_ROOT_ITEM_KEY) {
8475                         unsigned long offset;
8476                         u64 last_snapshot;
8477
8478                         offset = btrfs_item_ptr_offset(leaf, path.slots[0]);
8479                         read_extent_buffer(leaf, &ri, offset, sizeof(ri));
8480                         last_snapshot = btrfs_root_last_snapshot(&ri);
8481                         if (btrfs_disk_key_objectid(&ri.drop_progress) == 0) {
8482                                 level = btrfs_root_level(&ri);
8483                                 level_size = root->nodesize;
8484                                 ret = add_root_item_to_list(&normal_trees,
8485                                                 found_key.objectid,
8486                                                 btrfs_root_bytenr(&ri),
8487                                                 last_snapshot, level,
8488                                                 0, level_size, NULL);
8489                                 if (ret < 0)
8490                                         goto out;
8491                         } else {
8492                                 level = btrfs_root_level(&ri);
8493                                 level_size = root->nodesize;
8494                                 objectid = found_key.objectid;
8495                                 btrfs_disk_key_to_cpu(&found_key,
8496                                                       &ri.drop_progress);
8497                                 ret = add_root_item_to_list(&dropping_trees,
8498                                                 objectid,
8499                                                 btrfs_root_bytenr(&ri),
8500                                                 last_snapshot, level,
8501                                                 ri.drop_level,
8502                                                 level_size, &found_key);
8503                                 if (ret < 0)
8504                                         goto out;
8505                         }
8506                 }
8507                 path.slots[0]++;
8508         }
8509         btrfs_release_path(&path);
8510
8511         /*
8512          * check_block can return -EAGAIN if it fixes something, please keep
8513          * this in mind when dealing with return values from these functions, if
8514          * we get -EAGAIN we want to fall through and restart the loop.
8515          */
8516         ret = deal_root_from_list(&normal_trees, root, bits, bits_nr, &pending,
8517                                   &seen, &reada, &nodes, &extent_cache,
8518                                   &chunk_cache, &dev_cache, &block_group_cache,
8519                                   &dev_extent_cache);
8520         if (ret < 0) {
8521                 if (ret == -EAGAIN)
8522                         goto loop;
8523                 goto out;
8524         }
8525         ret = deal_root_from_list(&dropping_trees, root, bits, bits_nr,
8526                                   &pending, &seen, &reada, &nodes,
8527                                   &extent_cache, &chunk_cache, &dev_cache,
8528                                   &block_group_cache, &dev_extent_cache);
8529         if (ret < 0) {
8530                 if (ret == -EAGAIN)
8531                         goto loop;
8532                 goto out;
8533         }
8534
8535         ret = check_chunks(&chunk_cache, &block_group_cache,
8536                            &dev_extent_cache, NULL, NULL, NULL, 0);
8537         if (ret) {
8538                 if (ret == -EAGAIN)
8539                         goto loop;
8540                 err = ret;
8541         }
8542
8543         ret = check_extent_refs(root, &extent_cache);
8544         if (ret < 0) {
8545                 if (ret == -EAGAIN)
8546                         goto loop;
8547                 goto out;
8548         }
8549
8550         ret = check_devices(&dev_cache, &dev_extent_cache);
8551         if (ret && err)
8552                 ret = err;
8553
8554 out:
8555         task_stop(ctx.info);
8556         if (repair) {
8557                 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
8558                 extent_io_tree_cleanup(&excluded_extents);
8559                 root->fs_info->fsck_extent_cache = NULL;
8560                 root->fs_info->free_extent_hook = NULL;
8561                 root->fs_info->corrupt_blocks = NULL;
8562                 root->fs_info->excluded_extents = NULL;
8563         }
8564         free(bits);
8565         free_chunk_cache_tree(&chunk_cache);
8566         free_device_cache_tree(&dev_cache);
8567         free_block_group_tree(&block_group_cache);
8568         free_device_extent_tree(&dev_extent_cache);
8569         free_extent_cache_tree(&seen);
8570         free_extent_cache_tree(&pending);
8571         free_extent_cache_tree(&reada);
8572         free_extent_cache_tree(&nodes);
8573         return ret;
8574 loop:
8575         free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
8576         free_extent_cache_tree(&seen);
8577         free_extent_cache_tree(&pending);
8578         free_extent_cache_tree(&reada);
8579         free_extent_cache_tree(&nodes);
8580         free_chunk_cache_tree(&chunk_cache);
8581         free_block_group_tree(&block_group_cache);
8582         free_device_cache_tree(&dev_cache);
8583         free_device_extent_tree(&dev_extent_cache);
8584         free_extent_record_cache(root->fs_info, &extent_cache);
8585         free_root_item_list(&normal_trees);
8586         free_root_item_list(&dropping_trees);
8587         extent_io_tree_cleanup(&excluded_extents);
8588         goto again;
8589 }
8590
8591 /*
8592  * Check backrefs of a tree block given by @bytenr or @eb.
8593  *
8594  * @root:       the root containing the @bytenr or @eb
8595  * @eb:         tree block extent buffer, can be NULL
8596  * @bytenr:     bytenr of the tree block to search
8597  * @level:      tree level of the tree block
8598  * @owner:      owner of the tree block
8599  *
8600  * Return >0 for any error found and output error message
8601  * Return 0 for no error found
8602  */
8603 static int check_tree_block_ref(struct btrfs_root *root,
8604                                 struct extent_buffer *eb, u64 bytenr,
8605                                 int level, u64 owner)
8606 {
8607         struct btrfs_key key;
8608         struct btrfs_root *extent_root = root->fs_info->extent_root;
8609         struct btrfs_path path;
8610         struct btrfs_extent_item *ei;
8611         struct btrfs_extent_inline_ref *iref;
8612         struct extent_buffer *leaf;
8613         unsigned long end;
8614         unsigned long ptr;
8615         int slot;
8616         int skinny_level;
8617         int type;
8618         u32 nodesize = root->nodesize;
8619         u32 item_size;
8620         u64 offset;
8621         int found_ref = 0;
8622         int err = 0;
8623         int ret;
8624
8625         btrfs_init_path(&path);
8626         key.objectid = bytenr;
8627         if (btrfs_fs_incompat(root->fs_info,
8628                               BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA))
8629                 key.type = BTRFS_METADATA_ITEM_KEY;
8630         else
8631                 key.type = BTRFS_EXTENT_ITEM_KEY;
8632         key.offset = (u64)-1;
8633
8634         /* Search for the backref in extent tree */
8635         ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
8636         if (ret < 0) {
8637                 err |= BACKREF_MISSING;
8638                 goto out;
8639         }
8640         ret = btrfs_previous_extent_item(extent_root, &path, bytenr);
8641         if (ret) {
8642                 err |= BACKREF_MISSING;
8643                 goto out;
8644         }
8645
8646         leaf = path.nodes[0];
8647         slot = path.slots[0];
8648         btrfs_item_key_to_cpu(leaf, &key, slot);
8649
8650         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
8651
8652         if (key.type == BTRFS_METADATA_ITEM_KEY) {
8653                 skinny_level = (int)key.offset;
8654                 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
8655         } else {
8656                 struct btrfs_tree_block_info *info;
8657
8658                 info = (struct btrfs_tree_block_info *)(ei + 1);
8659                 skinny_level = btrfs_tree_block_level(leaf, info);
8660                 iref = (struct btrfs_extent_inline_ref *)(info + 1);
8661         }
8662
8663         if (eb) {
8664                 u64 header_gen;
8665                 u64 extent_gen;
8666
8667                 if (!(btrfs_extent_flags(leaf, ei) &
8668                       BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
8669                         error(
8670                 "extent[%llu %u] backref type mismatch, missing bit: %llx",
8671                                 key.objectid, nodesize,
8672                                 BTRFS_EXTENT_FLAG_TREE_BLOCK);
8673                         err = BACKREF_MISMATCH;
8674                 }
8675                 header_gen = btrfs_header_generation(eb);
8676                 extent_gen = btrfs_extent_generation(leaf, ei);
8677                 if (header_gen != extent_gen) {
8678                         error(
8679         "extent[%llu %u] backref generation mismatch, wanted: %llu, have: %llu",
8680                                 key.objectid, nodesize, header_gen,
8681                                 extent_gen);
8682                         err = BACKREF_MISMATCH;
8683                 }
8684                 if (level != skinny_level) {
8685                         error(
8686                         "extent[%llu %u] level mismatch, wanted: %u, have: %u",
8687                                 key.objectid, nodesize, level, skinny_level);
8688                         err = BACKREF_MISMATCH;
8689                 }
8690                 if (!is_fstree(owner) && btrfs_extent_refs(leaf, ei) != 1) {
8691                         error(
8692                         "extent[%llu %u] is referred by other roots than %llu",
8693                                 key.objectid, nodesize, root->objectid);
8694                         err = BACKREF_MISMATCH;
8695                 }
8696         }
8697
8698         /*
8699          * Iterate the extent/metadata item to find the exact backref
8700          */
8701         item_size = btrfs_item_size_nr(leaf, slot);
8702         ptr = (unsigned long)iref;
8703         end = (unsigned long)ei + item_size;
8704         while (ptr < end) {
8705                 iref = (struct btrfs_extent_inline_ref *)ptr;
8706                 type = btrfs_extent_inline_ref_type(leaf, iref);
8707                 offset = btrfs_extent_inline_ref_offset(leaf, iref);
8708
8709                 if (type == BTRFS_TREE_BLOCK_REF_KEY &&
8710                         (offset == root->objectid || offset == owner)) {
8711                         found_ref = 1;
8712                 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
8713                         /* Check if the backref points to valid referencer */
8714                         found_ref = !check_tree_block_ref(root, NULL, offset,
8715                                                           level + 1, owner);
8716                 }
8717
8718                 if (found_ref)
8719                         break;
8720                 ptr += btrfs_extent_inline_ref_size(type);
8721         }
8722
8723         /*
8724          * Inlined extent item doesn't have what we need, check
8725          * TREE_BLOCK_REF_KEY
8726          */
8727         if (!found_ref) {
8728                 btrfs_release_path(&path);
8729                 key.objectid = bytenr;
8730                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
8731                 key.offset = root->objectid;
8732
8733                 ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
8734                 if (!ret)
8735                         found_ref = 1;
8736         }
8737         if (!found_ref)
8738                 err |= BACKREF_MISSING;
8739 out:
8740         btrfs_release_path(&path);
8741         if (eb && (err & BACKREF_MISSING))
8742                 error("extent[%llu %u] backref lost (owner: %llu, level: %u)",
8743                         bytenr, nodesize, owner, level);
8744         return err;
8745 }
8746
8747 /*
8748  * Check EXTENT_DATA item, mainly for its dbackref in extent tree
8749  *
8750  * Return >0 any error found and output error message
8751  * Return 0 for no error found
8752  */
8753 static int check_extent_data_item(struct btrfs_root *root,
8754                                   struct extent_buffer *eb, int slot)
8755 {
8756         struct btrfs_file_extent_item *fi;
8757         struct btrfs_path path;
8758         struct btrfs_root *extent_root = root->fs_info->extent_root;
8759         struct btrfs_key fi_key;
8760         struct btrfs_key dbref_key;
8761         struct extent_buffer *leaf;
8762         struct btrfs_extent_item *ei;
8763         struct btrfs_extent_inline_ref *iref;
8764         struct btrfs_extent_data_ref *dref;
8765         u64 owner;
8766         u64 file_extent_gen;
8767         u64 disk_bytenr;
8768         u64 disk_num_bytes;
8769         u64 extent_num_bytes;
8770         u64 extent_flags;
8771         u64 extent_gen;
8772         u32 item_size;
8773         unsigned long end;
8774         unsigned long ptr;
8775         int type;
8776         u64 ref_root;
8777         int found_dbackref = 0;
8778         int err = 0;
8779         int ret;
8780
8781         btrfs_item_key_to_cpu(eb, &fi_key, slot);
8782         fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
8783         file_extent_gen = btrfs_file_extent_generation(eb, fi);
8784
8785         /* Nothing to check for hole and inline data extents */
8786         if (btrfs_file_extent_type(eb, fi) == BTRFS_FILE_EXTENT_INLINE ||
8787             btrfs_file_extent_disk_bytenr(eb, fi) == 0)
8788                 return 0;
8789
8790         disk_bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
8791         disk_num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
8792         extent_num_bytes = btrfs_file_extent_num_bytes(eb, fi);
8793
8794         /* Check unaligned disk_num_bytes and num_bytes */
8795         if (!IS_ALIGNED(disk_num_bytes, root->sectorsize)) {
8796                 error(
8797 "file extent [%llu, %llu] has unaligned disk num bytes: %llu, should be aligned to %u",
8798                         fi_key.objectid, fi_key.offset, disk_num_bytes,
8799                         root->sectorsize);
8800                 err |= BYTES_UNALIGNED;
8801         } else {
8802                 data_bytes_allocated += disk_num_bytes;
8803         }
8804         if (!IS_ALIGNED(extent_num_bytes, root->sectorsize)) {
8805                 error(
8806 "file extent [%llu, %llu] has unaligned num bytes: %llu, should be aligned to %u",
8807                         fi_key.objectid, fi_key.offset, extent_num_bytes,
8808                         root->sectorsize);
8809                 err |= BYTES_UNALIGNED;
8810         } else {
8811                 data_bytes_referenced += extent_num_bytes;
8812         }
8813         owner = btrfs_header_owner(eb);
8814
8815         /* Check the extent item of the file extent in extent tree */
8816         btrfs_init_path(&path);
8817         dbref_key.objectid = btrfs_file_extent_disk_bytenr(eb, fi);
8818         dbref_key.type = BTRFS_EXTENT_ITEM_KEY;
8819         dbref_key.offset = btrfs_file_extent_disk_num_bytes(eb, fi);
8820
8821         ret = btrfs_search_slot(NULL, extent_root, &dbref_key, &path, 0, 0);
8822         if (ret) {
8823                 err |= BACKREF_MISSING;
8824                 goto error;
8825         }
8826
8827         leaf = path.nodes[0];
8828         slot = path.slots[0];
8829         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
8830
8831         extent_flags = btrfs_extent_flags(leaf, ei);
8832         extent_gen = btrfs_extent_generation(leaf, ei);
8833
8834         if (!(extent_flags & BTRFS_EXTENT_FLAG_DATA)) {
8835                 error(
8836                     "extent[%llu %llu] backref type mismatch, wanted bit: %llx",
8837                     disk_bytenr, disk_num_bytes,
8838                     BTRFS_EXTENT_FLAG_DATA);
8839                 err |= BACKREF_MISMATCH;
8840         }
8841
8842         if (file_extent_gen < extent_gen) {
8843                 error(
8844 "extent[%llu %llu] backref generation mismatch, wanted: <=%llu, have: %llu",
8845                         disk_bytenr, disk_num_bytes, file_extent_gen,
8846                         extent_gen);
8847                 err |= BACKREF_MISMATCH;
8848         }
8849
8850         /* Check data backref inside that extent item */
8851         item_size = btrfs_item_size_nr(leaf, path.slots[0]);
8852         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
8853         ptr = (unsigned long)iref;
8854         end = (unsigned long)ei + item_size;
8855         while (ptr < end) {
8856                 iref = (struct btrfs_extent_inline_ref *)ptr;
8857                 type = btrfs_extent_inline_ref_type(leaf, iref);
8858                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
8859
8860                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
8861                         ref_root = btrfs_extent_data_ref_root(leaf, dref);
8862                         if (ref_root == owner || ref_root == root->objectid)
8863                                 found_dbackref = 1;
8864                 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
8865                         found_dbackref = !check_tree_block_ref(root, NULL,
8866                                 btrfs_extent_inline_ref_offset(leaf, iref),
8867                                 0, owner);
8868                 }
8869
8870                 if (found_dbackref)
8871                         break;
8872                 ptr += btrfs_extent_inline_ref_size(type);
8873         }
8874
8875         /* Didn't found inlined data backref, try EXTENT_DATA_REF_KEY */
8876         if (!found_dbackref) {
8877                 btrfs_release_path(&path);
8878
8879                 btrfs_init_path(&path);
8880                 dbref_key.objectid = btrfs_file_extent_disk_bytenr(eb, fi);
8881                 dbref_key.type = BTRFS_EXTENT_DATA_REF_KEY;
8882                 dbref_key.offset = hash_extent_data_ref(root->objectid,
8883                                 fi_key.objectid, fi_key.offset);
8884
8885                 ret = btrfs_search_slot(NULL, root->fs_info->extent_root,
8886                                         &dbref_key, &path, 0, 0);
8887                 if (!ret)
8888                         found_dbackref = 1;
8889         }
8890
8891         if (!found_dbackref)
8892                 err |= BACKREF_MISSING;
8893 error:
8894         btrfs_release_path(&path);
8895         if (err & BACKREF_MISSING) {
8896                 error("data extent[%llu %llu] backref lost",
8897                       disk_bytenr, disk_num_bytes);
8898         }
8899         return err;
8900 }
8901
8902 /*
8903  * Get real tree block level for the case like shared block
8904  * Return >= 0 as tree level
8905  * Return <0 for error
8906  */
8907 static int query_tree_block_level(struct btrfs_fs_info *fs_info, u64 bytenr)
8908 {
8909         struct extent_buffer *eb;
8910         struct btrfs_path path;
8911         struct btrfs_key key;
8912         struct btrfs_extent_item *ei;
8913         u64 flags;
8914         u64 transid;
8915         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
8916         u8 backref_level;
8917         u8 header_level;
8918         int ret;
8919
8920         /* Search extent tree for extent generation and level */
8921         key.objectid = bytenr;
8922         key.type = BTRFS_METADATA_ITEM_KEY;
8923         key.offset = (u64)-1;
8924
8925         btrfs_init_path(&path);
8926         ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, &path, 0, 0);
8927         if (ret < 0)
8928                 goto release_out;
8929         ret = btrfs_previous_extent_item(fs_info->extent_root, &path, bytenr);
8930         if (ret < 0)
8931                 goto release_out;
8932         if (ret > 0) {
8933                 ret = -ENOENT;
8934                 goto release_out;
8935         }
8936
8937         btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
8938         ei = btrfs_item_ptr(path.nodes[0], path.slots[0],
8939                             struct btrfs_extent_item);
8940         flags = btrfs_extent_flags(path.nodes[0], ei);
8941         if (!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
8942                 ret = -ENOENT;
8943                 goto release_out;
8944         }
8945
8946         /* Get transid for later read_tree_block() check */
8947         transid = btrfs_extent_generation(path.nodes[0], ei);
8948
8949         /* Get backref level as one source */
8950         if (key.type == BTRFS_METADATA_ITEM_KEY) {
8951                 backref_level = key.offset;
8952         } else {
8953                 struct btrfs_tree_block_info *info;
8954
8955                 info = (struct btrfs_tree_block_info *)(ei + 1);
8956                 backref_level = btrfs_tree_block_level(path.nodes[0], info);
8957         }
8958         btrfs_release_path(&path);
8959
8960         /* Get level from tree block as an alternative source */
8961         eb = read_tree_block_fs_info(fs_info, bytenr, nodesize, transid);
8962         if (!extent_buffer_uptodate(eb)) {
8963                 free_extent_buffer(eb);
8964                 return -EIO;
8965         }
8966         header_level = btrfs_header_level(eb);
8967         free_extent_buffer(eb);
8968
8969         if (header_level != backref_level)
8970                 return -EIO;
8971         return header_level;
8972
8973 release_out:
8974         btrfs_release_path(&path);
8975         return ret;
8976 }
8977
8978 /*
8979  * Check if a tree block backref is valid (points to a valid tree block)
8980  * if level == -1, level will be resolved
8981  * Return >0 for any error found and print error message
8982  */
8983 static int check_tree_block_backref(struct btrfs_fs_info *fs_info, u64 root_id,
8984                                     u64 bytenr, int level)
8985 {
8986         struct btrfs_root *root;
8987         struct btrfs_key key;
8988         struct btrfs_path path;
8989         struct extent_buffer *eb;
8990         struct extent_buffer *node;
8991         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
8992         int err = 0;
8993         int ret;
8994
8995         /* Query level for level == -1 special case */
8996         if (level == -1)
8997                 level = query_tree_block_level(fs_info, bytenr);
8998         if (level < 0) {
8999                 err |= REFERENCER_MISSING;
9000                 goto out;
9001         }
9002
9003         key.objectid = root_id;
9004         key.type = BTRFS_ROOT_ITEM_KEY;
9005         key.offset = (u64)-1;
9006
9007         root = btrfs_read_fs_root(fs_info, &key);
9008         if (IS_ERR(root)) {
9009                 err |= REFERENCER_MISSING;
9010                 goto out;
9011         }
9012
9013         /* Read out the tree block to get item/node key */
9014         eb = read_tree_block(root, bytenr, root->nodesize, 0);
9015         if (!extent_buffer_uptodate(eb)) {
9016                 err |= REFERENCER_MISSING;
9017                 free_extent_buffer(eb);
9018                 goto out;
9019         }
9020
9021         /* Empty tree, no need to check key */
9022         if (!btrfs_header_nritems(eb) && !level) {
9023                 free_extent_buffer(eb);
9024                 goto out;
9025         }
9026
9027         if (level)
9028                 btrfs_node_key_to_cpu(eb, &key, 0);
9029         else
9030                 btrfs_item_key_to_cpu(eb, &key, 0);
9031
9032         free_extent_buffer(eb);
9033
9034         btrfs_init_path(&path);
9035         path.lowest_level = level;
9036         /* Search with the first key, to ensure we can reach it */
9037         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
9038         if (ret < 0) {
9039                 err |= REFERENCER_MISSING;
9040                 goto release_out;
9041         }
9042
9043         node = path.nodes[level];
9044         if (btrfs_header_bytenr(node) != bytenr) {
9045                 error(
9046         "extent [%llu %d] referencer bytenr mismatch, wanted: %llu, have: %llu",
9047                         bytenr, nodesize, bytenr,
9048                         btrfs_header_bytenr(node));
9049                 err |= REFERENCER_MISMATCH;
9050         }
9051         if (btrfs_header_level(node) != level) {
9052                 error(
9053         "extent [%llu %d] referencer level mismatch, wanted: %d, have: %d",
9054                         bytenr, nodesize, level,
9055                         btrfs_header_level(node));
9056                 err |= REFERENCER_MISMATCH;
9057         }
9058
9059 release_out:
9060         btrfs_release_path(&path);
9061 out:
9062         if (err & REFERENCER_MISSING) {
9063                 if (level < 0)
9064                         error("extent [%llu %d] lost referencer (owner: %llu)",
9065                                 bytenr, nodesize, root_id);
9066                 else
9067                         error(
9068                 "extent [%llu %d] lost referencer (owner: %llu, level: %u)",
9069                                 bytenr, nodesize, root_id, level);
9070         }
9071
9072         return err;
9073 }
9074
9075 /*
9076  * Check referencer for shared block backref
9077  * If level == -1, this function will resolve the level.
9078  */
9079 static int check_shared_block_backref(struct btrfs_fs_info *fs_info,
9080                                      u64 parent, u64 bytenr, int level)
9081 {
9082         struct extent_buffer *eb;
9083         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
9084         u32 nr;
9085         int found_parent = 0;
9086         int i;
9087
9088         eb = read_tree_block_fs_info(fs_info, parent, nodesize, 0);
9089         if (!extent_buffer_uptodate(eb))
9090                 goto out;
9091
9092         if (level == -1)
9093                 level = query_tree_block_level(fs_info, bytenr);
9094         if (level < 0)
9095                 goto out;
9096
9097         if (level + 1 != btrfs_header_level(eb))
9098                 goto out;
9099
9100         nr = btrfs_header_nritems(eb);
9101         for (i = 0; i < nr; i++) {
9102                 if (bytenr == btrfs_node_blockptr(eb, i)) {
9103                         found_parent = 1;
9104                         break;
9105                 }
9106         }
9107 out:
9108         free_extent_buffer(eb);
9109         if (!found_parent) {
9110                 error(
9111         "shared extent[%llu %u] lost its parent (parent: %llu, level: %u)",
9112                         bytenr, nodesize, parent, level);
9113                 return REFERENCER_MISSING;
9114         }
9115         return 0;
9116 }
9117
9118 /*
9119  * Check referencer for normal (inlined) data ref
9120  * If len == 0, it will be resolved by searching in extent tree
9121  */
9122 static int check_extent_data_backref(struct btrfs_fs_info *fs_info,
9123                                      u64 root_id, u64 objectid, u64 offset,
9124                                      u64 bytenr, u64 len, u32 count)
9125 {
9126         struct btrfs_root *root;
9127         struct btrfs_root *extent_root = fs_info->extent_root;
9128         struct btrfs_key key;
9129         struct btrfs_path path;
9130         struct extent_buffer *leaf;
9131         struct btrfs_file_extent_item *fi;
9132         u32 found_count = 0;
9133         int slot;
9134         int ret = 0;
9135
9136         if (!len) {
9137                 key.objectid = bytenr;
9138                 key.type = BTRFS_EXTENT_ITEM_KEY;
9139                 key.offset = (u64)-1;
9140
9141                 btrfs_init_path(&path);
9142                 ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
9143                 if (ret < 0)
9144                         goto out;
9145                 ret = btrfs_previous_extent_item(extent_root, &path, bytenr);
9146                 if (ret)
9147                         goto out;
9148                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
9149                 if (key.objectid != bytenr ||
9150                     key.type != BTRFS_EXTENT_ITEM_KEY)
9151                         goto out;
9152                 len = key.offset;
9153                 btrfs_release_path(&path);
9154         }
9155         key.objectid = root_id;
9156         key.type = BTRFS_ROOT_ITEM_KEY;
9157         key.offset = (u64)-1;
9158         btrfs_init_path(&path);
9159
9160         root = btrfs_read_fs_root(fs_info, &key);
9161         if (IS_ERR(root))
9162                 goto out;
9163
9164         key.objectid = objectid;
9165         key.type = BTRFS_EXTENT_DATA_KEY;
9166         /*
9167          * It can be nasty as data backref offset is
9168          * file offset - file extent offset, which is smaller or
9169          * equal to original backref offset.  The only special case is
9170          * overflow.  So we need to special check and do further search.
9171          */
9172         key.offset = offset & (1ULL << 63) ? 0 : offset;
9173
9174         ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
9175         if (ret < 0)
9176                 goto out;
9177
9178         /*
9179          * Search afterwards to get correct one
9180          * NOTE: As we must do a comprehensive check on the data backref to
9181          * make sure the dref count also matches, we must iterate all file
9182          * extents for that inode.
9183          */
9184         while (1) {
9185                 leaf = path.nodes[0];
9186                 slot = path.slots[0];
9187
9188                 btrfs_item_key_to_cpu(leaf, &key, slot);
9189                 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
9190                         break;
9191                 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
9192                 /*
9193                  * Except normal disk bytenr and disk num bytes, we still
9194                  * need to do extra check on dbackref offset as
9195                  * dbackref offset = file_offset - file_extent_offset
9196                  */
9197                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == bytenr &&
9198                     btrfs_file_extent_disk_num_bytes(leaf, fi) == len &&
9199                     (u64)(key.offset - btrfs_file_extent_offset(leaf, fi)) ==
9200                     offset)
9201                         found_count++;
9202
9203                 ret = btrfs_next_item(root, &path);
9204                 if (ret)
9205                         break;
9206         }
9207 out:
9208         btrfs_release_path(&path);
9209         if (found_count != count) {
9210                 error(
9211 "extent[%llu, %llu] referencer count mismatch (root: %llu, owner: %llu, offset: %llu) wanted: %u, have: %u",
9212                         bytenr, len, root_id, objectid, offset, count, found_count);
9213                 return REFERENCER_MISSING;
9214         }
9215         return 0;
9216 }
9217
9218 /*
9219  * Check if the referencer of a shared data backref exists
9220  */
9221 static int check_shared_data_backref(struct btrfs_fs_info *fs_info,
9222                                      u64 parent, u64 bytenr)
9223 {
9224         struct extent_buffer *eb;
9225         struct btrfs_key key;
9226         struct btrfs_file_extent_item *fi;
9227         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
9228         u32 nr;
9229         int found_parent = 0;
9230         int i;
9231
9232         eb = read_tree_block_fs_info(fs_info, parent, nodesize, 0);
9233         if (!extent_buffer_uptodate(eb))
9234                 goto out;
9235
9236         nr = btrfs_header_nritems(eb);
9237         for (i = 0; i < nr; i++) {
9238                 btrfs_item_key_to_cpu(eb, &key, i);
9239                 if (key.type != BTRFS_EXTENT_DATA_KEY)
9240                         continue;
9241
9242                 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
9243                 if (btrfs_file_extent_type(eb, fi) == BTRFS_FILE_EXTENT_INLINE)
9244                         continue;
9245
9246                 if (btrfs_file_extent_disk_bytenr(eb, fi) == bytenr) {
9247                         found_parent = 1;
9248                         break;
9249                 }
9250         }
9251
9252 out:
9253         free_extent_buffer(eb);
9254         if (!found_parent) {
9255                 error("shared extent %llu referencer lost (parent: %llu)",
9256                         bytenr, parent);
9257                 return REFERENCER_MISSING;
9258         }
9259         return 0;
9260 }
9261
9262 /*
9263  * This function will check a given extent item, including its backref and
9264  * itself (like crossing stripe boundary and type)
9265  *
9266  * Since we don't use extent_record anymore, introduce new error bit
9267  */
9268 static int check_extent_item(struct btrfs_fs_info *fs_info,
9269                              struct extent_buffer *eb, int slot)
9270 {
9271         struct btrfs_extent_item *ei;
9272         struct btrfs_extent_inline_ref *iref;
9273         struct btrfs_extent_data_ref *dref;
9274         unsigned long end;
9275         unsigned long ptr;
9276         int type;
9277         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
9278         u32 item_size = btrfs_item_size_nr(eb, slot);
9279         u64 flags;
9280         u64 offset;
9281         int metadata = 0;
9282         int level;
9283         struct btrfs_key key;
9284         int ret;
9285         int err = 0;
9286
9287         btrfs_item_key_to_cpu(eb, &key, slot);
9288         if (key.type == BTRFS_EXTENT_ITEM_KEY)
9289                 bytes_used += key.offset;
9290         else
9291                 bytes_used += nodesize;
9292
9293         if (item_size < sizeof(*ei)) {
9294                 /*
9295                  * COMPAT_EXTENT_TREE_V0 case, but it's already a super
9296                  * old thing when on disk format is still un-determined.
9297                  * No need to care about it anymore
9298                  */
9299                 error("unsupported COMPAT_EXTENT_TREE_V0 detected");
9300                 return -ENOTTY;
9301         }
9302
9303         ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
9304         flags = btrfs_extent_flags(eb, ei);
9305
9306         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
9307                 metadata = 1;
9308         if (metadata && check_crossing_stripes(global_info, key.objectid,
9309                                                eb->len)) {
9310                 error("bad metadata [%llu, %llu) crossing stripe boundary",
9311                       key.objectid, key.objectid + nodesize);
9312                 err |= CROSSING_STRIPE_BOUNDARY;
9313         }
9314
9315         ptr = (unsigned long)(ei + 1);
9316
9317         if (metadata && key.type == BTRFS_EXTENT_ITEM_KEY) {
9318                 /* Old EXTENT_ITEM metadata */
9319                 struct btrfs_tree_block_info *info;
9320
9321                 info = (struct btrfs_tree_block_info *)ptr;
9322                 level = btrfs_tree_block_level(eb, info);
9323                 ptr += sizeof(struct btrfs_tree_block_info);
9324         } else {
9325                 /* New METADATA_ITEM */
9326                 level = key.offset;
9327         }
9328         end = (unsigned long)ei + item_size;
9329
9330         if (ptr >= end) {
9331                 err |= ITEM_SIZE_MISMATCH;
9332                 goto out;
9333         }
9334
9335         /* Now check every backref in this extent item */
9336 next:
9337         iref = (struct btrfs_extent_inline_ref *)ptr;
9338         type = btrfs_extent_inline_ref_type(eb, iref);
9339         offset = btrfs_extent_inline_ref_offset(eb, iref);
9340         switch (type) {
9341         case BTRFS_TREE_BLOCK_REF_KEY:
9342                 ret = check_tree_block_backref(fs_info, offset, key.objectid,
9343                                                level);
9344                 err |= ret;
9345                 break;
9346         case BTRFS_SHARED_BLOCK_REF_KEY:
9347                 ret = check_shared_block_backref(fs_info, offset, key.objectid,
9348                                                  level);
9349                 err |= ret;
9350                 break;
9351         case BTRFS_EXTENT_DATA_REF_KEY:
9352                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
9353                 ret = check_extent_data_backref(fs_info,
9354                                 btrfs_extent_data_ref_root(eb, dref),
9355                                 btrfs_extent_data_ref_objectid(eb, dref),
9356                                 btrfs_extent_data_ref_offset(eb, dref),
9357                                 key.objectid, key.offset,
9358                                 btrfs_extent_data_ref_count(eb, dref));
9359                 err |= ret;
9360                 break;
9361         case BTRFS_SHARED_DATA_REF_KEY:
9362                 ret = check_shared_data_backref(fs_info, offset, key.objectid);
9363                 err |= ret;
9364                 break;
9365         default:
9366                 error("extent[%llu %d %llu] has unknown ref type: %d",
9367                         key.objectid, key.type, key.offset, type);
9368                 err |= UNKNOWN_TYPE;
9369                 goto out;
9370         }
9371
9372         ptr += btrfs_extent_inline_ref_size(type);
9373         if (ptr < end)
9374                 goto next;
9375
9376 out:
9377         return err;
9378 }
9379
9380 /*
9381  * Check if a dev extent item is referred correctly by its chunk
9382  */
9383 static int check_dev_extent_item(struct btrfs_fs_info *fs_info,
9384                                  struct extent_buffer *eb, int slot)
9385 {
9386         struct btrfs_root *chunk_root = fs_info->chunk_root;
9387         struct btrfs_dev_extent *ptr;
9388         struct btrfs_path path;
9389         struct btrfs_key chunk_key;
9390         struct btrfs_key devext_key;
9391         struct btrfs_chunk *chunk;
9392         struct extent_buffer *l;
9393         int num_stripes;
9394         u64 length;
9395         int i;
9396         int found_chunk = 0;
9397         int ret;
9398
9399         btrfs_item_key_to_cpu(eb, &devext_key, slot);
9400         ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_extent);
9401         length = btrfs_dev_extent_length(eb, ptr);
9402
9403         chunk_key.objectid = btrfs_dev_extent_chunk_objectid(eb, ptr);
9404         chunk_key.type = BTRFS_CHUNK_ITEM_KEY;
9405         chunk_key.offset = btrfs_dev_extent_chunk_offset(eb, ptr);
9406
9407         btrfs_init_path(&path);
9408         ret = btrfs_search_slot(NULL, chunk_root, &chunk_key, &path, 0, 0);
9409         if (ret)
9410                 goto out;
9411
9412         l = path.nodes[0];
9413         chunk = btrfs_item_ptr(l, path.slots[0], struct btrfs_chunk);
9414         if (btrfs_chunk_length(l, chunk) != length)
9415                 goto out;
9416
9417         num_stripes = btrfs_chunk_num_stripes(l, chunk);
9418         for (i = 0; i < num_stripes; i++) {
9419                 u64 devid = btrfs_stripe_devid_nr(l, chunk, i);
9420                 u64 offset = btrfs_stripe_offset_nr(l, chunk, i);
9421
9422                 if (devid == devext_key.objectid &&
9423                     offset == devext_key.offset) {
9424                         found_chunk = 1;
9425                         break;
9426                 }
9427         }
9428 out:
9429         btrfs_release_path(&path);
9430         if (!found_chunk) {
9431                 error(
9432                 "device extent[%llu, %llu, %llu] did not find the related chunk",
9433                         devext_key.objectid, devext_key.offset, length);
9434                 return REFERENCER_MISSING;
9435         }
9436         return 0;
9437 }
9438
9439 /*
9440  * Check if the used space is correct with the dev item
9441  */
9442 static int check_dev_item(struct btrfs_fs_info *fs_info,
9443                           struct extent_buffer *eb, int slot)
9444 {
9445         struct btrfs_root *dev_root = fs_info->dev_root;
9446         struct btrfs_dev_item *dev_item;
9447         struct btrfs_path path;
9448         struct btrfs_key key;
9449         struct btrfs_dev_extent *ptr;
9450         u64 dev_id;
9451         u64 used;
9452         u64 total = 0;
9453         int ret;
9454
9455         dev_item = btrfs_item_ptr(eb, slot, struct btrfs_dev_item);
9456         dev_id = btrfs_device_id(eb, dev_item);
9457         used = btrfs_device_bytes_used(eb, dev_item);
9458
9459         key.objectid = dev_id;
9460         key.type = BTRFS_DEV_EXTENT_KEY;
9461         key.offset = 0;
9462
9463         btrfs_init_path(&path);
9464         ret = btrfs_search_slot(NULL, dev_root, &key, &path, 0, 0);
9465         if (ret < 0) {
9466                 btrfs_item_key_to_cpu(eb, &key, slot);
9467                 error("cannot find any related dev extent for dev[%llu, %u, %llu]",
9468                         key.objectid, key.type, key.offset);
9469                 btrfs_release_path(&path);
9470                 return REFERENCER_MISSING;
9471         }
9472
9473         /* Iterate dev_extents to calculate the used space of a device */
9474         while (1) {
9475                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
9476
9477                 if (key.objectid > dev_id)
9478                         break;
9479                 if (key.type != BTRFS_DEV_EXTENT_KEY || key.objectid != dev_id)
9480                         goto next;
9481
9482                 ptr = btrfs_item_ptr(path.nodes[0], path.slots[0],
9483                                      struct btrfs_dev_extent);
9484                 total += btrfs_dev_extent_length(path.nodes[0], ptr);
9485 next:
9486                 ret = btrfs_next_item(dev_root, &path);
9487                 if (ret)
9488                         break;
9489         }
9490         btrfs_release_path(&path);
9491
9492         if (used != total) {
9493                 btrfs_item_key_to_cpu(eb, &key, slot);
9494                 error(
9495 "Dev extent's total-byte %llu is not equal to bytes-used %llu in dev[%llu, %u, %llu]",
9496                         total, used, BTRFS_ROOT_TREE_OBJECTID,
9497                         BTRFS_DEV_EXTENT_KEY, dev_id);
9498                 return ACCOUNTING_MISMATCH;
9499         }
9500         return 0;
9501 }
9502
9503 /*
9504  * Check a block group item with its referener (chunk) and its used space
9505  * with extent/metadata item
9506  */
9507 static int check_block_group_item(struct btrfs_fs_info *fs_info,
9508                                   struct extent_buffer *eb, int slot)
9509 {
9510         struct btrfs_root *extent_root = fs_info->extent_root;
9511         struct btrfs_root *chunk_root = fs_info->chunk_root;
9512         struct btrfs_block_group_item *bi;
9513         struct btrfs_block_group_item bg_item;
9514         struct btrfs_path path;
9515         struct btrfs_key bg_key;
9516         struct btrfs_key chunk_key;
9517         struct btrfs_key extent_key;
9518         struct btrfs_chunk *chunk;
9519         struct extent_buffer *leaf;
9520         struct btrfs_extent_item *ei;
9521         u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
9522         u64 flags;
9523         u64 bg_flags;
9524         u64 used;
9525         u64 total = 0;
9526         int ret;
9527         int err = 0;
9528
9529         btrfs_item_key_to_cpu(eb, &bg_key, slot);
9530         bi = btrfs_item_ptr(eb, slot, struct btrfs_block_group_item);
9531         read_extent_buffer(eb, &bg_item, (unsigned long)bi, sizeof(bg_item));
9532         used = btrfs_block_group_used(&bg_item);
9533         bg_flags = btrfs_block_group_flags(&bg_item);
9534
9535         chunk_key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
9536         chunk_key.type = BTRFS_CHUNK_ITEM_KEY;
9537         chunk_key.offset = bg_key.objectid;
9538
9539         btrfs_init_path(&path);
9540         /* Search for the referencer chunk */
9541         ret = btrfs_search_slot(NULL, chunk_root, &chunk_key, &path, 0, 0);
9542         if (ret) {
9543                 error(
9544                 "block group[%llu %llu] did not find the related chunk item",
9545                         bg_key.objectid, bg_key.offset);
9546                 err |= REFERENCER_MISSING;
9547         } else {
9548                 chunk = btrfs_item_ptr(path.nodes[0], path.slots[0],
9549                                         struct btrfs_chunk);
9550                 if (btrfs_chunk_length(path.nodes[0], chunk) !=
9551                                                 bg_key.offset) {
9552                         error(
9553         "block group[%llu %llu] related chunk item length does not match",
9554                                 bg_key.objectid, bg_key.offset);
9555                         err |= REFERENCER_MISMATCH;
9556                 }
9557         }
9558         btrfs_release_path(&path);
9559
9560         /* Search from the block group bytenr */
9561         extent_key.objectid = bg_key.objectid;
9562         extent_key.type = 0;
9563         extent_key.offset = 0;
9564
9565         btrfs_init_path(&path);
9566         ret = btrfs_search_slot(NULL, extent_root, &extent_key, &path, 0, 0);
9567         if (ret < 0)
9568                 goto out;
9569
9570         /* Iterate extent tree to account used space */
9571         while (1) {
9572                 leaf = path.nodes[0];
9573                 btrfs_item_key_to_cpu(leaf, &extent_key, path.slots[0]);
9574                 if (extent_key.objectid >= bg_key.objectid + bg_key.offset)
9575                         break;
9576
9577                 if (extent_key.type != BTRFS_METADATA_ITEM_KEY &&
9578                     extent_key.type != BTRFS_EXTENT_ITEM_KEY)
9579                         goto next;
9580                 if (extent_key.objectid < bg_key.objectid)
9581                         goto next;
9582
9583                 if (extent_key.type == BTRFS_METADATA_ITEM_KEY)
9584                         total += nodesize;
9585                 else
9586                         total += extent_key.offset;
9587
9588                 ei = btrfs_item_ptr(leaf, path.slots[0],
9589                                     struct btrfs_extent_item);
9590                 flags = btrfs_extent_flags(leaf, ei);
9591                 if (flags & BTRFS_EXTENT_FLAG_DATA) {
9592                         if (!(bg_flags & BTRFS_BLOCK_GROUP_DATA)) {
9593                                 error(
9594                         "bad extent[%llu, %llu) type mismatch with chunk",
9595                                         extent_key.objectid,
9596                                         extent_key.objectid + extent_key.offset);
9597                                 err |= CHUNK_TYPE_MISMATCH;
9598                         }
9599                 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
9600                         if (!(bg_flags & (BTRFS_BLOCK_GROUP_SYSTEM |
9601                                     BTRFS_BLOCK_GROUP_METADATA))) {
9602                                 error(
9603                         "bad extent[%llu, %llu) type mismatch with chunk",
9604                                         extent_key.objectid,
9605                                         extent_key.objectid + nodesize);
9606                                 err |= CHUNK_TYPE_MISMATCH;
9607                         }
9608                 }
9609 next:
9610                 ret = btrfs_next_item(extent_root, &path);
9611                 if (ret)
9612                         break;
9613         }
9614
9615 out:
9616         btrfs_release_path(&path);
9617
9618         if (total != used) {
9619                 error(
9620                 "block group[%llu %llu] used %llu but extent items used %llu",
9621                         bg_key.objectid, bg_key.offset, used, total);
9622                 err |= ACCOUNTING_MISMATCH;
9623         }
9624         return err;
9625 }
9626
9627 /*
9628  * Check a chunk item.
9629  * Including checking all referred dev_extents and block group
9630  */
9631 static int check_chunk_item(struct btrfs_fs_info *fs_info,
9632                             struct extent_buffer *eb, int slot)
9633 {
9634         struct btrfs_root *extent_root = fs_info->extent_root;
9635         struct btrfs_root *dev_root = fs_info->dev_root;
9636         struct btrfs_path path;
9637         struct btrfs_key chunk_key;
9638         struct btrfs_key bg_key;
9639         struct btrfs_key devext_key;
9640         struct btrfs_chunk *chunk;
9641         struct extent_buffer *leaf;
9642         struct btrfs_block_group_item *bi;
9643         struct btrfs_block_group_item bg_item;
9644         struct btrfs_dev_extent *ptr;
9645         u32 sectorsize = btrfs_super_sectorsize(fs_info->super_copy);
9646         u64 length;
9647         u64 chunk_end;
9648         u64 type;
9649         u64 profile;
9650         int num_stripes;
9651         u64 offset;
9652         u64 objectid;
9653         int i;
9654         int ret;
9655         int err = 0;
9656
9657         btrfs_item_key_to_cpu(eb, &chunk_key, slot);
9658         chunk = btrfs_item_ptr(eb, slot, struct btrfs_chunk);
9659         length = btrfs_chunk_length(eb, chunk);
9660         chunk_end = chunk_key.offset + length;
9661         if (!IS_ALIGNED(length, sectorsize)) {
9662                 error("chunk[%llu %llu) not aligned to %u",
9663                         chunk_key.offset, chunk_end, sectorsize);
9664                 err |= BYTES_UNALIGNED;
9665                 goto out;
9666         }
9667
9668         type = btrfs_chunk_type(eb, chunk);
9669         profile = type & BTRFS_BLOCK_GROUP_PROFILE_MASK;
9670         if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
9671                 error("chunk[%llu %llu) has no chunk type",
9672                         chunk_key.offset, chunk_end);
9673                 err |= UNKNOWN_TYPE;
9674         }
9675         if (profile && (profile & (profile - 1))) {
9676                 error("chunk[%llu %llu) multiple profiles detected: %llx",
9677                         chunk_key.offset, chunk_end, profile);
9678                 err |= UNKNOWN_TYPE;
9679         }
9680
9681         bg_key.objectid = chunk_key.offset;
9682         bg_key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9683         bg_key.offset = length;
9684
9685         btrfs_init_path(&path);
9686         ret = btrfs_search_slot(NULL, extent_root, &bg_key, &path, 0, 0);
9687         if (ret) {
9688                 error(
9689                 "chunk[%llu %llu) did not find the related block group item",
9690                         chunk_key.offset, chunk_end);
9691                 err |= REFERENCER_MISSING;
9692         } else{
9693                 leaf = path.nodes[0];
9694                 bi = btrfs_item_ptr(leaf, path.slots[0],
9695                                     struct btrfs_block_group_item);
9696                 read_extent_buffer(leaf, &bg_item, (unsigned long)bi,
9697                                    sizeof(bg_item));
9698                 if (btrfs_block_group_flags(&bg_item) != type) {
9699                         error(
9700 "chunk[%llu %llu) related block group item flags mismatch, wanted: %llu, have: %llu",
9701                                 chunk_key.offset, chunk_end, type,
9702                                 btrfs_block_group_flags(&bg_item));
9703                         err |= REFERENCER_MISSING;
9704                 }
9705         }
9706
9707         num_stripes = btrfs_chunk_num_stripes(eb, chunk);
9708         for (i = 0; i < num_stripes; i++) {
9709                 btrfs_release_path(&path);
9710                 btrfs_init_path(&path);
9711                 devext_key.objectid = btrfs_stripe_devid_nr(eb, chunk, i);
9712                 devext_key.type = BTRFS_DEV_EXTENT_KEY;
9713                 devext_key.offset = btrfs_stripe_offset_nr(eb, chunk, i);
9714
9715                 ret = btrfs_search_slot(NULL, dev_root, &devext_key, &path,
9716                                         0, 0);
9717                 if (ret)
9718                         goto not_match_dev;
9719
9720                 leaf = path.nodes[0];
9721                 ptr = btrfs_item_ptr(leaf, path.slots[0],
9722                                      struct btrfs_dev_extent);
9723                 objectid = btrfs_dev_extent_chunk_objectid(leaf, ptr);
9724                 offset = btrfs_dev_extent_chunk_offset(leaf, ptr);
9725                 if (objectid != chunk_key.objectid ||
9726                     offset != chunk_key.offset ||
9727                     btrfs_dev_extent_length(leaf, ptr) != length)
9728                         goto not_match_dev;
9729                 continue;
9730 not_match_dev:
9731                 err |= BACKREF_MISSING;
9732                 error(
9733                 "chunk[%llu %llu) stripe %d did not find the related dev extent",
9734                         chunk_key.objectid, chunk_end, i);
9735                 continue;
9736         }
9737         btrfs_release_path(&path);
9738 out:
9739         return err;
9740 }
9741
9742 /*
9743  * Main entry function to check known items and update related accounting info
9744  */
9745 static int check_leaf_items(struct btrfs_root *root, struct extent_buffer *eb)
9746 {
9747         struct btrfs_fs_info *fs_info = root->fs_info;
9748         struct btrfs_key key;
9749         int slot = 0;
9750         int type;
9751         struct btrfs_extent_data_ref *dref;
9752         int ret;
9753         int err = 0;
9754
9755 next:
9756         btrfs_item_key_to_cpu(eb, &key, slot);
9757         type = key.type;
9758
9759         switch (type) {
9760         case BTRFS_EXTENT_DATA_KEY:
9761                 ret = check_extent_data_item(root, eb, slot);
9762                 err |= ret;
9763                 break;
9764         case BTRFS_BLOCK_GROUP_ITEM_KEY:
9765                 ret = check_block_group_item(fs_info, eb, slot);
9766                 err |= ret;
9767                 break;
9768         case BTRFS_DEV_ITEM_KEY:
9769                 ret = check_dev_item(fs_info, eb, slot);
9770                 err |= ret;
9771                 break;
9772         case BTRFS_CHUNK_ITEM_KEY:
9773                 ret = check_chunk_item(fs_info, eb, slot);
9774                 err |= ret;
9775                 break;
9776         case BTRFS_DEV_EXTENT_KEY:
9777                 ret = check_dev_extent_item(fs_info, eb, slot);
9778                 err |= ret;
9779                 break;
9780         case BTRFS_EXTENT_ITEM_KEY:
9781         case BTRFS_METADATA_ITEM_KEY:
9782                 ret = check_extent_item(fs_info, eb, slot);
9783                 err |= ret;
9784                 break;
9785         case BTRFS_EXTENT_CSUM_KEY:
9786                 total_csum_bytes += btrfs_item_size_nr(eb, slot);
9787                 break;
9788         case BTRFS_TREE_BLOCK_REF_KEY:
9789                 ret = check_tree_block_backref(fs_info, key.offset,
9790                                                key.objectid, -1);
9791                 err |= ret;
9792                 break;
9793         case BTRFS_EXTENT_DATA_REF_KEY:
9794                 dref = btrfs_item_ptr(eb, slot, struct btrfs_extent_data_ref);
9795                 ret = check_extent_data_backref(fs_info,
9796                                 btrfs_extent_data_ref_root(eb, dref),
9797                                 btrfs_extent_data_ref_objectid(eb, dref),
9798                                 btrfs_extent_data_ref_offset(eb, dref),
9799                                 key.objectid, 0,
9800                                 btrfs_extent_data_ref_count(eb, dref));
9801                 err |= ret;
9802                 break;
9803         case BTRFS_SHARED_BLOCK_REF_KEY:
9804                 ret = check_shared_block_backref(fs_info, key.offset,
9805                                                  key.objectid, -1);
9806                 err |= ret;
9807                 break;
9808         case BTRFS_SHARED_DATA_REF_KEY:
9809                 ret = check_shared_data_backref(fs_info, key.offset,
9810                                                 key.objectid);
9811                 err |= ret;
9812                 break;
9813         default:
9814                 break;
9815         }
9816
9817         if (++slot < btrfs_header_nritems(eb))
9818                 goto next;
9819
9820         return err;
9821 }
9822
9823 /*
9824  * Helper function for later fs/subvol tree check.  To determine if a tree
9825  * block should be checked.
9826  * This function will ensure only the direct referencer with lowest rootid to
9827  * check a fs/subvolume tree block.
9828  *
9829  * Backref check at extent tree would detect errors like missing subvolume
9830  * tree, so we can do aggressive check to reduce duplicated checks.
9831  */
9832 static int should_check(struct btrfs_root *root, struct extent_buffer *eb)
9833 {
9834         struct btrfs_root *extent_root = root->fs_info->extent_root;
9835         struct btrfs_key key;
9836         struct btrfs_path path;
9837         struct extent_buffer *leaf;
9838         int slot;
9839         struct btrfs_extent_item *ei;
9840         unsigned long ptr;
9841         unsigned long end;
9842         int type;
9843         u32 item_size;
9844         u64 offset;
9845         struct btrfs_extent_inline_ref *iref;
9846         int ret;
9847
9848         btrfs_init_path(&path);
9849         key.objectid = btrfs_header_bytenr(eb);
9850         key.type = BTRFS_METADATA_ITEM_KEY;
9851         key.offset = (u64)-1;
9852
9853         /*
9854          * Any failure in backref resolving means we can't determine
9855          * whom the tree block belongs to.
9856          * So in that case, we need to check that tree block
9857          */
9858         ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
9859         if (ret < 0)
9860                 goto need_check;
9861
9862         ret = btrfs_previous_extent_item(extent_root, &path,
9863                                          btrfs_header_bytenr(eb));
9864         if (ret)
9865                 goto need_check;
9866
9867         leaf = path.nodes[0];
9868         slot = path.slots[0];
9869         btrfs_item_key_to_cpu(leaf, &key, slot);
9870         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
9871
9872         if (key.type == BTRFS_METADATA_ITEM_KEY) {
9873                 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
9874         } else {
9875                 struct btrfs_tree_block_info *info;
9876
9877                 info = (struct btrfs_tree_block_info *)(ei + 1);
9878                 iref = (struct btrfs_extent_inline_ref *)(info + 1);
9879         }
9880
9881         item_size = btrfs_item_size_nr(leaf, slot);
9882         ptr = (unsigned long)iref;
9883         end = (unsigned long)ei + item_size;
9884         while (ptr < end) {
9885                 iref = (struct btrfs_extent_inline_ref *)ptr;
9886                 type = btrfs_extent_inline_ref_type(leaf, iref);
9887                 offset = btrfs_extent_inline_ref_offset(leaf, iref);
9888
9889                 /*
9890                  * We only check the tree block if current root is
9891                  * the lowest referencer of it.
9892                  */
9893                 if (type == BTRFS_TREE_BLOCK_REF_KEY &&
9894                     offset < root->objectid) {
9895                         btrfs_release_path(&path);
9896                         return 0;
9897                 }
9898
9899                 ptr += btrfs_extent_inline_ref_size(type);
9900         }
9901         /*
9902          * Normally we should also check keyed tree block ref, but that may be
9903          * very time consuming.  Inlined ref should already make us skip a lot
9904          * of refs now.  So skip search keyed tree block ref.
9905          */
9906
9907 need_check:
9908         btrfs_release_path(&path);
9909         return 1;
9910 }
9911
9912 /*
9913  * Traversal function for tree block. We will do:
9914  * 1) Skip shared fs/subvolume tree blocks
9915  * 2) Update related bytes accounting
9916  * 3) Pre-order traversal
9917  */
9918 static int traverse_tree_block(struct btrfs_root *root,
9919                                 struct extent_buffer *node)
9920 {
9921         struct extent_buffer *eb;
9922         struct btrfs_key key;
9923         struct btrfs_key drop_key;
9924         int level;
9925         u64 nr;
9926         int i;
9927         int err = 0;
9928         int ret;
9929
9930         /*
9931          * Skip shared fs/subvolume tree block, in that case they will
9932          * be checked by referencer with lowest rootid
9933          */
9934         if (is_fstree(root->objectid) && !should_check(root, node))
9935                 return 0;
9936
9937         /* Update bytes accounting */
9938         total_btree_bytes += node->len;
9939         if (fs_root_objectid(btrfs_header_owner(node)))
9940                 total_fs_tree_bytes += node->len;
9941         if (btrfs_header_owner(node) == BTRFS_EXTENT_TREE_OBJECTID)
9942                 total_extent_tree_bytes += node->len;
9943         if (!found_old_backref &&
9944             btrfs_header_owner(node) == BTRFS_TREE_RELOC_OBJECTID &&
9945             btrfs_header_backref_rev(node) == BTRFS_MIXED_BACKREF_REV &&
9946             !btrfs_header_flag(node, BTRFS_HEADER_FLAG_RELOC))
9947                 found_old_backref = 1;
9948
9949         /* pre-order tranversal, check itself first */
9950         level = btrfs_header_level(node);
9951         ret = check_tree_block_ref(root, node, btrfs_header_bytenr(node),
9952                                    btrfs_header_level(node),
9953                                    btrfs_header_owner(node));
9954         err |= ret;
9955         if (err)
9956                 error(
9957         "check %s failed root %llu bytenr %llu level %d, force continue check",
9958                         level ? "node":"leaf", root->objectid,
9959                         btrfs_header_bytenr(node), btrfs_header_level(node));
9960
9961         if (!level) {
9962                 btree_space_waste += btrfs_leaf_free_space(root, node);
9963                 ret = check_leaf_items(root, node);
9964                 err |= ret;
9965                 return err;
9966         }
9967
9968         nr = btrfs_header_nritems(node);
9969         btrfs_disk_key_to_cpu(&drop_key, &root->root_item.drop_progress);
9970         btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) - nr) *
9971                 sizeof(struct btrfs_key_ptr);
9972
9973         /* Then check all its children */
9974         for (i = 0; i < nr; i++) {
9975                 u64 blocknr = btrfs_node_blockptr(node, i);
9976
9977                 btrfs_node_key_to_cpu(node, &key, i);
9978                 if (level == root->root_item.drop_level &&
9979                     is_dropped_key(&key, &drop_key))
9980                         continue;
9981
9982                 /*
9983                  * As a btrfs tree has most 8 levels (0..7), so it's quite safe
9984                  * to call the function itself.
9985                  */
9986                 eb = read_tree_block(root, blocknr, root->nodesize, 0);
9987                 if (extent_buffer_uptodate(eb)) {
9988                         ret = traverse_tree_block(root, eb);
9989                         err |= ret;
9990                 }
9991                 free_extent_buffer(eb);
9992         }
9993
9994         return err;
9995 }
9996
9997 /*
9998  * Low memory usage version check_chunks_and_extents.
9999  */
10000 static int check_chunks_and_extents_v2(struct btrfs_root *root)
10001 {
10002         struct btrfs_path path;
10003         struct btrfs_key key;
10004         struct btrfs_root *root1;
10005         struct btrfs_root *cur_root;
10006         int err = 0;
10007         int ret;
10008
10009         root1 = root->fs_info->chunk_root;
10010         ret = traverse_tree_block(root1, root1->node);
10011         err |= ret;
10012
10013         root1 = root->fs_info->tree_root;
10014         ret = traverse_tree_block(root1, root1->node);
10015         err |= ret;
10016
10017         btrfs_init_path(&path);
10018         key.objectid = BTRFS_EXTENT_TREE_OBJECTID;
10019         key.offset = 0;
10020         key.type = BTRFS_ROOT_ITEM_KEY;
10021
10022         ret = btrfs_search_slot(NULL, root1, &key, &path, 0, 0);
10023         if (ret) {
10024                 error("cannot find extent treet in tree_root");
10025                 goto out;
10026         }
10027
10028         while (1) {
10029                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
10030                 if (key.type != BTRFS_ROOT_ITEM_KEY)
10031                         goto next;
10032                 key.offset = (u64)-1;
10033
10034                 cur_root = btrfs_read_fs_root(root->fs_info, &key);
10035                 if (IS_ERR(cur_root) || !cur_root) {
10036                         error("failed to read tree: %lld", key.objectid);
10037                         goto next;
10038                 }
10039
10040                 ret = traverse_tree_block(cur_root, cur_root->node);
10041                 err |= ret;
10042
10043 next:
10044                 ret = btrfs_next_item(root1, &path);
10045                 if (ret)
10046                         goto out;
10047         }
10048
10049 out:
10050         btrfs_release_path(&path);
10051         return err;
10052 }
10053
10054 static int btrfs_fsck_reinit_root(struct btrfs_trans_handle *trans,
10055                            struct btrfs_root *root, int overwrite)
10056 {
10057         struct extent_buffer *c;
10058         struct extent_buffer *old = root->node;
10059         int level;
10060         int ret;
10061         struct btrfs_disk_key disk_key = {0,0,0};
10062
10063         level = 0;
10064
10065         if (overwrite) {
10066                 c = old;
10067                 extent_buffer_get(c);
10068                 goto init;
10069         }
10070         c = btrfs_alloc_free_block(trans, root,
10071                                    root->nodesize,
10072                                    root->root_key.objectid,
10073                                    &disk_key, level, 0, 0);
10074         if (IS_ERR(c)) {
10075                 c = old;
10076                 extent_buffer_get(c);
10077                 overwrite = 1;
10078         }
10079 init:
10080         memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
10081         btrfs_set_header_level(c, level);
10082         btrfs_set_header_bytenr(c, c->start);
10083         btrfs_set_header_generation(c, trans->transid);
10084         btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
10085         btrfs_set_header_owner(c, root->root_key.objectid);
10086
10087         write_extent_buffer(c, root->fs_info->fsid,
10088                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
10089
10090         write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
10091                             btrfs_header_chunk_tree_uuid(c),
10092                             BTRFS_UUID_SIZE);
10093
10094         btrfs_mark_buffer_dirty(c);
10095         /*
10096          * this case can happen in the following case:
10097          *
10098          * 1.overwrite previous root.
10099          *
10100          * 2.reinit reloc data root, this is because we skip pin
10101          * down reloc data tree before which means we can allocate
10102          * same block bytenr here.
10103          */
10104         if (old->start == c->start) {
10105                 btrfs_set_root_generation(&root->root_item,
10106                                           trans->transid);
10107                 root->root_item.level = btrfs_header_level(root->node);
10108                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
10109                                         &root->root_key, &root->root_item);
10110                 if (ret) {
10111                         free_extent_buffer(c);
10112                         return ret;
10113                 }
10114         }
10115         free_extent_buffer(old);
10116         root->node = c;
10117         add_root_to_dirty_list(root);
10118         return 0;
10119 }
10120
10121 static int pin_down_tree_blocks(struct btrfs_fs_info *fs_info,
10122                                 struct extent_buffer *eb, int tree_root)
10123 {
10124         struct extent_buffer *tmp;
10125         struct btrfs_root_item *ri;
10126         struct btrfs_key key;
10127         u64 bytenr;
10128         u32 nodesize;
10129         int level = btrfs_header_level(eb);
10130         int nritems;
10131         int ret;
10132         int i;
10133
10134         /*
10135          * If we have pinned this block before, don't pin it again.
10136          * This can not only avoid forever loop with broken filesystem
10137          * but also give us some speedups.
10138          */
10139         if (test_range_bit(&fs_info->pinned_extents, eb->start,
10140                            eb->start + eb->len - 1, EXTENT_DIRTY, 0))
10141                 return 0;
10142
10143         btrfs_pin_extent(fs_info, eb->start, eb->len);
10144
10145         nodesize = btrfs_super_nodesize(fs_info->super_copy);
10146         nritems = btrfs_header_nritems(eb);
10147         for (i = 0; i < nritems; i++) {
10148                 if (level == 0) {
10149                         btrfs_item_key_to_cpu(eb, &key, i);
10150                         if (key.type != BTRFS_ROOT_ITEM_KEY)
10151                                 continue;
10152                         /* Skip the extent root and reloc roots */
10153                         if (key.objectid == BTRFS_EXTENT_TREE_OBJECTID ||
10154                             key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
10155                             key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
10156                                 continue;
10157                         ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
10158                         bytenr = btrfs_disk_root_bytenr(eb, ri);
10159
10160                         /*
10161                          * If at any point we start needing the real root we
10162                          * will have to build a stump root for the root we are
10163                          * in, but for now this doesn't actually use the root so
10164                          * just pass in extent_root.
10165                          */
10166                         tmp = read_tree_block(fs_info->extent_root, bytenr,
10167                                               nodesize, 0);
10168                         if (!extent_buffer_uptodate(tmp)) {
10169                                 fprintf(stderr, "Error reading root block\n");
10170                                 return -EIO;
10171                         }
10172                         ret = pin_down_tree_blocks(fs_info, tmp, 0);
10173                         free_extent_buffer(tmp);
10174                         if (ret)
10175                                 return ret;
10176                 } else {
10177                         bytenr = btrfs_node_blockptr(eb, i);
10178
10179                         /* If we aren't the tree root don't read the block */
10180                         if (level == 1 && !tree_root) {
10181                                 btrfs_pin_extent(fs_info, bytenr, nodesize);
10182                                 continue;
10183                         }
10184
10185                         tmp = read_tree_block(fs_info->extent_root, bytenr,
10186                                               nodesize, 0);
10187                         if (!extent_buffer_uptodate(tmp)) {
10188                                 fprintf(stderr, "Error reading tree block\n");
10189                                 return -EIO;
10190                         }
10191                         ret = pin_down_tree_blocks(fs_info, tmp, tree_root);
10192                         free_extent_buffer(tmp);
10193                         if (ret)
10194                                 return ret;
10195                 }
10196         }
10197
10198         return 0;
10199 }
10200
10201 static int pin_metadata_blocks(struct btrfs_fs_info *fs_info)
10202 {
10203         int ret;
10204
10205         ret = pin_down_tree_blocks(fs_info, fs_info->chunk_root->node, 0);
10206         if (ret)
10207                 return ret;
10208
10209         return pin_down_tree_blocks(fs_info, fs_info->tree_root->node, 1);
10210 }
10211
10212 static int reset_block_groups(struct btrfs_fs_info *fs_info)
10213 {
10214         struct btrfs_block_group_cache *cache;
10215         struct btrfs_path path;
10216         struct extent_buffer *leaf;
10217         struct btrfs_chunk *chunk;
10218         struct btrfs_key key;
10219         int ret;
10220         u64 start;
10221
10222         btrfs_init_path(&path);
10223         key.objectid = 0;
10224         key.type = BTRFS_CHUNK_ITEM_KEY;
10225         key.offset = 0;
10226         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, &path, 0, 0);
10227         if (ret < 0) {
10228                 btrfs_release_path(&path);
10229                 return ret;
10230         }
10231
10232         /*
10233          * We do this in case the block groups were screwed up and had alloc
10234          * bits that aren't actually set on the chunks.  This happens with
10235          * restored images every time and could happen in real life I guess.
10236          */
10237         fs_info->avail_data_alloc_bits = 0;
10238         fs_info->avail_metadata_alloc_bits = 0;
10239         fs_info->avail_system_alloc_bits = 0;
10240
10241         /* First we need to create the in-memory block groups */
10242         while (1) {
10243                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
10244                         ret = btrfs_next_leaf(fs_info->chunk_root, &path);
10245                         if (ret < 0) {
10246                                 btrfs_release_path(&path);
10247                                 return ret;
10248                         }
10249                         if (ret) {
10250                                 ret = 0;
10251                                 break;
10252                         }
10253                 }
10254                 leaf = path.nodes[0];
10255                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
10256                 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
10257                         path.slots[0]++;
10258                         continue;
10259                 }
10260
10261                 chunk = btrfs_item_ptr(leaf, path.slots[0], struct btrfs_chunk);
10262                 btrfs_add_block_group(fs_info, 0,
10263                                       btrfs_chunk_type(leaf, chunk),
10264                                       key.objectid, key.offset,
10265                                       btrfs_chunk_length(leaf, chunk));
10266                 set_extent_dirty(&fs_info->free_space_cache, key.offset,
10267                                  key.offset + btrfs_chunk_length(leaf, chunk),
10268                                  GFP_NOFS);
10269                 path.slots[0]++;
10270         }
10271         start = 0;
10272         while (1) {
10273                 cache = btrfs_lookup_first_block_group(fs_info, start);
10274                 if (!cache)
10275                         break;
10276                 cache->cached = 1;
10277                 start = cache->key.objectid + cache->key.offset;
10278         }
10279
10280         btrfs_release_path(&path);
10281         return 0;
10282 }
10283
10284 static int reset_balance(struct btrfs_trans_handle *trans,
10285                          struct btrfs_fs_info *fs_info)
10286 {
10287         struct btrfs_root *root = fs_info->tree_root;
10288         struct btrfs_path path;
10289         struct extent_buffer *leaf;
10290         struct btrfs_key key;
10291         int del_slot, del_nr = 0;
10292         int ret;
10293         int found = 0;
10294
10295         btrfs_init_path(&path);
10296         key.objectid = BTRFS_BALANCE_OBJECTID;
10297         key.type = BTRFS_BALANCE_ITEM_KEY;
10298         key.offset = 0;
10299         ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
10300         if (ret) {
10301                 if (ret > 0)
10302                         ret = 0;
10303                 if (!ret)
10304                         goto reinit_data_reloc;
10305                 else
10306                         goto out;
10307         }
10308
10309         ret = btrfs_del_item(trans, root, &path);
10310         if (ret)
10311                 goto out;
10312         btrfs_release_path(&path);
10313
10314         key.objectid = BTRFS_TREE_RELOC_OBJECTID;
10315         key.type = BTRFS_ROOT_ITEM_KEY;
10316         key.offset = 0;
10317         ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
10318         if (ret < 0)
10319                 goto out;
10320         while (1) {
10321                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
10322                         if (!found)
10323                                 break;
10324
10325                         if (del_nr) {
10326                                 ret = btrfs_del_items(trans, root, &path,
10327                                                       del_slot, del_nr);
10328                                 del_nr = 0;
10329                                 if (ret)
10330                                         goto out;
10331                         }
10332                         key.offset++;
10333                         btrfs_release_path(&path);
10334
10335                         found = 0;
10336                         ret = btrfs_search_slot(trans, root, &key, &path,
10337                                                 -1, 1);
10338                         if (ret < 0)
10339                                 goto out;
10340                         continue;
10341                 }
10342                 found = 1;
10343                 leaf = path.nodes[0];
10344                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
10345                 if (key.objectid > BTRFS_TREE_RELOC_OBJECTID)
10346                         break;
10347                 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
10348                         path.slots[0]++;
10349                         continue;
10350                 }
10351                 if (!del_nr) {
10352                         del_slot = path.slots[0];
10353                         del_nr = 1;
10354                 } else {
10355                         del_nr++;
10356                 }
10357                 path.slots[0]++;
10358         }
10359
10360         if (del_nr) {
10361                 ret = btrfs_del_items(trans, root, &path, del_slot, del_nr);
10362                 if (ret)
10363                         goto out;
10364         }
10365         btrfs_release_path(&path);
10366
10367 reinit_data_reloc:
10368         key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
10369         key.type = BTRFS_ROOT_ITEM_KEY;
10370         key.offset = (u64)-1;
10371         root = btrfs_read_fs_root(fs_info, &key);
10372         if (IS_ERR(root)) {
10373                 fprintf(stderr, "Error reading data reloc tree\n");
10374                 ret = PTR_ERR(root);
10375                 goto out;
10376         }
10377         record_root_in_trans(trans, root);
10378         ret = btrfs_fsck_reinit_root(trans, root, 0);
10379         if (ret)
10380                 goto out;
10381         ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
10382 out:
10383         btrfs_release_path(&path);
10384         return ret;
10385 }
10386
10387 static int reinit_extent_tree(struct btrfs_trans_handle *trans,
10388                               struct btrfs_fs_info *fs_info)
10389 {
10390         u64 start = 0;
10391         int ret;
10392
10393         /*
10394          * The only reason we don't do this is because right now we're just
10395          * walking the trees we find and pinning down their bytes, we don't look
10396          * at any of the leaves.  In order to do mixed groups we'd have to check
10397          * the leaves of any fs roots and pin down the bytes for any file
10398          * extents we find.  Not hard but why do it if we don't have to?
10399          */
10400         if (btrfs_fs_incompat(fs_info, BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)) {
10401                 fprintf(stderr, "We don't support re-initing the extent tree "
10402                         "for mixed block groups yet, please notify a btrfs "
10403                         "developer you want to do this so they can add this "
10404                         "functionality.\n");
10405                 return -EINVAL;
10406         }
10407
10408         /*
10409          * first we need to walk all of the trees except the extent tree and pin
10410          * down the bytes that are in use so we don't overwrite any existing
10411          * metadata.
10412          */
10413         ret = pin_metadata_blocks(fs_info);
10414         if (ret) {
10415                 fprintf(stderr, "error pinning down used bytes\n");
10416                 return ret;
10417         }
10418
10419         /*
10420          * Need to drop all the block groups since we're going to recreate all
10421          * of them again.
10422          */
10423         btrfs_free_block_groups(fs_info);
10424         ret = reset_block_groups(fs_info);
10425         if (ret) {
10426                 fprintf(stderr, "error resetting the block groups\n");
10427                 return ret;
10428         }
10429
10430         /* Ok we can allocate now, reinit the extent root */
10431         ret = btrfs_fsck_reinit_root(trans, fs_info->extent_root, 0);
10432         if (ret) {
10433                 fprintf(stderr, "extent root initialization failed\n");
10434                 /*
10435                  * When the transaction code is updated we should end the
10436                  * transaction, but for now progs only knows about commit so
10437                  * just return an error.
10438                  */
10439                 return ret;
10440         }
10441
10442         /*
10443          * Now we have all the in-memory block groups setup so we can make
10444          * allocations properly, and the metadata we care about is safe since we
10445          * pinned all of it above.
10446          */
10447         while (1) {
10448                 struct btrfs_block_group_cache *cache;
10449
10450                 cache = btrfs_lookup_first_block_group(fs_info, start);
10451                 if (!cache)
10452                         break;
10453                 start = cache->key.objectid + cache->key.offset;
10454                 ret = btrfs_insert_item(trans, fs_info->extent_root,
10455                                         &cache->key, &cache->item,
10456                                         sizeof(cache->item));
10457                 if (ret) {
10458                         fprintf(stderr, "Error adding block group\n");
10459                         return ret;
10460                 }
10461                 btrfs_extent_post_op(trans, fs_info->extent_root);
10462         }
10463
10464         ret = reset_balance(trans, fs_info);
10465         if (ret)
10466                 fprintf(stderr, "error resetting the pending balance\n");
10467
10468         return ret;
10469 }
10470
10471 static int recow_extent_buffer(struct btrfs_root *root, struct extent_buffer *eb)
10472 {
10473         struct btrfs_path path;
10474         struct btrfs_trans_handle *trans;
10475         struct btrfs_key key;
10476         int ret;
10477
10478         printf("Recowing metadata block %llu\n", eb->start);
10479         key.objectid = btrfs_header_owner(eb);
10480         key.type = BTRFS_ROOT_ITEM_KEY;
10481         key.offset = (u64)-1;
10482
10483         root = btrfs_read_fs_root(root->fs_info, &key);
10484         if (IS_ERR(root)) {
10485                 fprintf(stderr, "Couldn't find owner root %llu\n",
10486                         key.objectid);
10487                 return PTR_ERR(root);
10488         }
10489
10490         trans = btrfs_start_transaction(root, 1);
10491         if (IS_ERR(trans))
10492                 return PTR_ERR(trans);
10493
10494         btrfs_init_path(&path);
10495         path.lowest_level = btrfs_header_level(eb);
10496         if (path.lowest_level)
10497                 btrfs_node_key_to_cpu(eb, &key, 0);
10498         else
10499                 btrfs_item_key_to_cpu(eb, &key, 0);
10500
10501         ret = btrfs_search_slot(trans, root, &key, &path, 0, 1);
10502         btrfs_commit_transaction(trans, root);
10503         btrfs_release_path(&path);
10504         return ret;
10505 }
10506
10507 static int delete_bad_item(struct btrfs_root *root, struct bad_item *bad)
10508 {
10509         struct btrfs_path path;
10510         struct btrfs_trans_handle *trans;
10511         struct btrfs_key key;
10512         int ret;
10513
10514         printf("Deleting bad item [%llu,%u,%llu]\n", bad->key.objectid,
10515                bad->key.type, bad->key.offset);
10516         key.objectid = bad->root_id;
10517         key.type = BTRFS_ROOT_ITEM_KEY;
10518         key.offset = (u64)-1;
10519
10520         root = btrfs_read_fs_root(root->fs_info, &key);
10521         if (IS_ERR(root)) {
10522                 fprintf(stderr, "Couldn't find owner root %llu\n",
10523                         key.objectid);
10524                 return PTR_ERR(root);
10525         }
10526
10527         trans = btrfs_start_transaction(root, 1);
10528         if (IS_ERR(trans))
10529                 return PTR_ERR(trans);
10530
10531         btrfs_init_path(&path);
10532         ret = btrfs_search_slot(trans, root, &bad->key, &path, -1, 1);
10533         if (ret) {
10534                 if (ret > 0)
10535                         ret = 0;
10536                 goto out;
10537         }
10538         ret = btrfs_del_item(trans, root, &path);
10539 out:
10540         btrfs_commit_transaction(trans, root);
10541         btrfs_release_path(&path);
10542         return ret;
10543 }
10544
10545 static int zero_log_tree(struct btrfs_root *root)
10546 {
10547         struct btrfs_trans_handle *trans;
10548         int ret;
10549
10550         trans = btrfs_start_transaction(root, 1);
10551         if (IS_ERR(trans)) {
10552                 ret = PTR_ERR(trans);
10553                 return ret;
10554         }
10555         btrfs_set_super_log_root(root->fs_info->super_copy, 0);
10556         btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
10557         ret = btrfs_commit_transaction(trans, root);
10558         return ret;
10559 }
10560
10561 static int populate_csum(struct btrfs_trans_handle *trans,
10562                          struct btrfs_root *csum_root, char *buf, u64 start,
10563                          u64 len)
10564 {
10565         u64 offset = 0;
10566         u64 sectorsize;
10567         int ret = 0;
10568
10569         while (offset < len) {
10570                 sectorsize = csum_root->sectorsize;
10571                 ret = read_extent_data(csum_root, buf, start + offset,
10572                                        &sectorsize, 0);
10573                 if (ret)
10574                         break;
10575                 ret = btrfs_csum_file_block(trans, csum_root, start + len,
10576                                             start + offset, buf, sectorsize);
10577                 if (ret)
10578                         break;
10579                 offset += sectorsize;
10580         }
10581         return ret;
10582 }
10583
10584 static int fill_csum_tree_from_one_fs_root(struct btrfs_trans_handle *trans,
10585                                       struct btrfs_root *csum_root,
10586                                       struct btrfs_root *cur_root)
10587 {
10588         struct btrfs_path path;
10589         struct btrfs_key key;
10590         struct extent_buffer *node;
10591         struct btrfs_file_extent_item *fi;
10592         char *buf = NULL;
10593         u64 start = 0;
10594         u64 len = 0;
10595         int slot = 0;
10596         int ret = 0;
10597
10598         buf = malloc(cur_root->fs_info->csum_root->sectorsize);
10599         if (!buf)
10600                 return -ENOMEM;
10601
10602         btrfs_init_path(&path);
10603         key.objectid = 0;
10604         key.offset = 0;
10605         key.type = 0;
10606         ret = btrfs_search_slot(NULL, cur_root, &key, &path, 0, 0);
10607         if (ret < 0)
10608                 goto out;
10609         /* Iterate all regular file extents and fill its csum */
10610         while (1) {
10611                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
10612
10613                 if (key.type != BTRFS_EXTENT_DATA_KEY)
10614                         goto next;
10615                 node = path.nodes[0];
10616                 slot = path.slots[0];
10617                 fi = btrfs_item_ptr(node, slot, struct btrfs_file_extent_item);
10618                 if (btrfs_file_extent_type(node, fi) != BTRFS_FILE_EXTENT_REG)
10619                         goto next;
10620                 start = btrfs_file_extent_disk_bytenr(node, fi);
10621                 len = btrfs_file_extent_disk_num_bytes(node, fi);
10622
10623                 ret = populate_csum(trans, csum_root, buf, start, len);
10624                 if (ret == -EEXIST)
10625                         ret = 0;
10626                 if (ret < 0)
10627                         goto out;
10628 next:
10629                 /*
10630                  * TODO: if next leaf is corrupted, jump to nearest next valid
10631                  * leaf.
10632                  */
10633                 ret = btrfs_next_item(cur_root, &path);
10634                 if (ret < 0)
10635                         goto out;
10636                 if (ret > 0) {
10637                         ret = 0;
10638                         goto out;
10639                 }
10640         }
10641
10642 out:
10643         btrfs_release_path(&path);
10644         free(buf);
10645         return ret;
10646 }
10647
10648 static int fill_csum_tree_from_fs(struct btrfs_trans_handle *trans,
10649                                   struct btrfs_root *csum_root)
10650 {
10651         struct btrfs_fs_info *fs_info = csum_root->fs_info;
10652         struct btrfs_path path;
10653         struct btrfs_root *tree_root = fs_info->tree_root;
10654         struct btrfs_root *cur_root;
10655         struct extent_buffer *node;
10656         struct btrfs_key key;
10657         int slot = 0;
10658         int ret = 0;
10659
10660         btrfs_init_path(&path);
10661         key.objectid = BTRFS_FS_TREE_OBJECTID;
10662         key.offset = 0;
10663         key.type = BTRFS_ROOT_ITEM_KEY;
10664         ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
10665         if (ret < 0)
10666                 goto out;
10667         if (ret > 0) {
10668                 ret = -ENOENT;
10669                 goto out;
10670         }
10671
10672         while (1) {
10673                 node = path.nodes[0];
10674                 slot = path.slots[0];
10675                 btrfs_item_key_to_cpu(node, &key, slot);
10676                 if (key.objectid > BTRFS_LAST_FREE_OBJECTID)
10677                         goto out;
10678                 if (key.type != BTRFS_ROOT_ITEM_KEY)
10679                         goto next;
10680                 if (!is_fstree(key.objectid))
10681                         goto next;
10682                 key.offset = (u64)-1;
10683
10684                 cur_root = btrfs_read_fs_root(fs_info, &key);
10685                 if (IS_ERR(cur_root) || !cur_root) {
10686                         fprintf(stderr, "Fail to read fs/subvol tree: %lld\n",
10687                                 key.objectid);
10688                         goto out;
10689                 }
10690                 ret = fill_csum_tree_from_one_fs_root(trans, csum_root,
10691                                 cur_root);
10692                 if (ret < 0)
10693                         goto out;
10694 next:
10695                 ret = btrfs_next_item(tree_root, &path);
10696                 if (ret > 0) {
10697                         ret = 0;
10698                         goto out;
10699                 }
10700                 if (ret < 0)
10701                         goto out;
10702         }
10703
10704 out:
10705         btrfs_release_path(&path);
10706         return ret;
10707 }
10708
10709 static int fill_csum_tree_from_extent(struct btrfs_trans_handle *trans,
10710                                       struct btrfs_root *csum_root)
10711 {
10712         struct btrfs_root *extent_root = csum_root->fs_info->extent_root;
10713         struct btrfs_path path;
10714         struct btrfs_extent_item *ei;
10715         struct extent_buffer *leaf;
10716         char *buf;
10717         struct btrfs_key key;
10718         int ret;
10719
10720         btrfs_init_path(&path);
10721         key.objectid = 0;
10722         key.type = BTRFS_EXTENT_ITEM_KEY;
10723         key.offset = 0;
10724         ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
10725         if (ret < 0) {
10726                 btrfs_release_path(&path);
10727                 return ret;
10728         }
10729
10730         buf = malloc(csum_root->sectorsize);
10731         if (!buf) {
10732                 btrfs_release_path(&path);
10733                 return -ENOMEM;
10734         }
10735
10736         while (1) {
10737                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
10738                         ret = btrfs_next_leaf(extent_root, &path);
10739                         if (ret < 0)
10740                                 break;
10741                         if (ret) {
10742                                 ret = 0;
10743                                 break;
10744                         }
10745                 }
10746                 leaf = path.nodes[0];
10747
10748                 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
10749                 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
10750                         path.slots[0]++;
10751                         continue;
10752                 }
10753
10754                 ei = btrfs_item_ptr(leaf, path.slots[0],
10755                                     struct btrfs_extent_item);
10756                 if (!(btrfs_extent_flags(leaf, ei) &
10757                       BTRFS_EXTENT_FLAG_DATA)) {
10758                         path.slots[0]++;
10759                         continue;
10760                 }
10761
10762                 ret = populate_csum(trans, csum_root, buf, key.objectid,
10763                                     key.offset);
10764                 if (ret)
10765                         break;
10766                 path.slots[0]++;
10767         }
10768
10769         btrfs_release_path(&path);
10770         free(buf);
10771         return ret;
10772 }
10773
10774 /*
10775  * Recalculate the csum and put it into the csum tree.
10776  *
10777  * Extent tree init will wipe out all the extent info, so in that case, we
10778  * can't depend on extent tree, but use fs tree.  If search_fs_tree is set, we
10779  * will use fs/subvol trees to init the csum tree.
10780  */
10781 static int fill_csum_tree(struct btrfs_trans_handle *trans,
10782                           struct btrfs_root *csum_root,
10783                           int search_fs_tree)
10784 {
10785         if (search_fs_tree)
10786                 return fill_csum_tree_from_fs(trans, csum_root);
10787         else
10788                 return fill_csum_tree_from_extent(trans, csum_root);
10789 }
10790
10791 static void free_roots_info_cache(void)
10792 {
10793         if (!roots_info_cache)
10794                 return;
10795
10796         while (!cache_tree_empty(roots_info_cache)) {
10797                 struct cache_extent *entry;
10798                 struct root_item_info *rii;
10799
10800                 entry = first_cache_extent(roots_info_cache);
10801                 if (!entry)
10802                         break;
10803                 remove_cache_extent(roots_info_cache, entry);
10804                 rii = container_of(entry, struct root_item_info, cache_extent);
10805                 free(rii);
10806         }
10807
10808         free(roots_info_cache);
10809         roots_info_cache = NULL;
10810 }
10811
10812 static int build_roots_info_cache(struct btrfs_fs_info *info)
10813 {
10814         int ret = 0;
10815         struct btrfs_key key;
10816         struct extent_buffer *leaf;
10817         struct btrfs_path path;
10818
10819         if (!roots_info_cache) {
10820                 roots_info_cache = malloc(sizeof(*roots_info_cache));
10821                 if (!roots_info_cache)
10822                         return -ENOMEM;
10823                 cache_tree_init(roots_info_cache);
10824         }
10825
10826         btrfs_init_path(&path);
10827         key.objectid = 0;
10828         key.type = BTRFS_EXTENT_ITEM_KEY;
10829         key.offset = 0;
10830         ret = btrfs_search_slot(NULL, info->extent_root, &key, &path, 0, 0);
10831         if (ret < 0)
10832                 goto out;
10833         leaf = path.nodes[0];
10834
10835         while (1) {
10836                 struct btrfs_key found_key;
10837                 struct btrfs_extent_item *ei;
10838                 struct btrfs_extent_inline_ref *iref;
10839                 int slot = path.slots[0];
10840                 int type;
10841                 u64 flags;
10842                 u64 root_id;
10843                 u8 level;
10844                 struct cache_extent *entry;
10845                 struct root_item_info *rii;
10846
10847                 if (slot >= btrfs_header_nritems(leaf)) {
10848                         ret = btrfs_next_leaf(info->extent_root, &path);
10849                         if (ret < 0) {
10850                                 break;
10851                         } else if (ret) {
10852                                 ret = 0;
10853                                 break;
10854                         }
10855                         leaf = path.nodes[0];
10856                         slot = path.slots[0];
10857                 }
10858
10859                 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
10860
10861                 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
10862                     found_key.type != BTRFS_METADATA_ITEM_KEY)
10863                         goto next;
10864
10865                 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
10866                 flags = btrfs_extent_flags(leaf, ei);
10867
10868                 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
10869                     !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
10870                         goto next;
10871
10872                 if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
10873                         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
10874                         level = found_key.offset;
10875                 } else {
10876                         struct btrfs_tree_block_info *binfo;
10877
10878                         binfo = (struct btrfs_tree_block_info *)(ei + 1);
10879                         iref = (struct btrfs_extent_inline_ref *)(binfo + 1);
10880                         level = btrfs_tree_block_level(leaf, binfo);
10881                 }
10882
10883                 /*
10884                  * For a root extent, it must be of the following type and the
10885                  * first (and only one) iref in the item.
10886                  */
10887                 type = btrfs_extent_inline_ref_type(leaf, iref);
10888                 if (type != BTRFS_TREE_BLOCK_REF_KEY)
10889                         goto next;
10890
10891                 root_id = btrfs_extent_inline_ref_offset(leaf, iref);
10892                 entry = lookup_cache_extent(roots_info_cache, root_id, 1);
10893                 if (!entry) {
10894                         rii = malloc(sizeof(struct root_item_info));
10895                         if (!rii) {
10896                                 ret = -ENOMEM;
10897                                 goto out;
10898                         }
10899                         rii->cache_extent.start = root_id;
10900                         rii->cache_extent.size = 1;
10901                         rii->level = (u8)-1;
10902                         entry = &rii->cache_extent;
10903                         ret = insert_cache_extent(roots_info_cache, entry);
10904                         ASSERT(ret == 0);
10905                 } else {
10906                         rii = container_of(entry, struct root_item_info,
10907                                            cache_extent);
10908                 }
10909
10910                 ASSERT(rii->cache_extent.start == root_id);
10911                 ASSERT(rii->cache_extent.size == 1);
10912
10913                 if (level > rii->level || rii->level == (u8)-1) {
10914                         rii->level = level;
10915                         rii->bytenr = found_key.objectid;
10916                         rii->gen = btrfs_extent_generation(leaf, ei);
10917                         rii->node_count = 1;
10918                 } else if (level == rii->level) {
10919                         rii->node_count++;
10920                 }
10921 next:
10922                 path.slots[0]++;
10923         }
10924
10925 out:
10926         btrfs_release_path(&path);
10927
10928         return ret;
10929 }
10930
10931 static int maybe_repair_root_item(struct btrfs_fs_info *info,
10932                                   struct btrfs_path *path,
10933                                   const struct btrfs_key *root_key,
10934                                   const int read_only_mode)
10935 {
10936         const u64 root_id = root_key->objectid;
10937         struct cache_extent *entry;
10938         struct root_item_info *rii;
10939         struct btrfs_root_item ri;
10940         unsigned long offset;
10941
10942         entry = lookup_cache_extent(roots_info_cache, root_id, 1);
10943         if (!entry) {
10944                 fprintf(stderr,
10945                         "Error: could not find extent items for root %llu\n",
10946                         root_key->objectid);
10947                 return -ENOENT;
10948         }
10949
10950         rii = container_of(entry, struct root_item_info, cache_extent);
10951         ASSERT(rii->cache_extent.start == root_id);
10952         ASSERT(rii->cache_extent.size == 1);
10953
10954         if (rii->node_count != 1) {
10955                 fprintf(stderr,
10956                         "Error: could not find btree root extent for root %llu\n",
10957                         root_id);
10958                 return -ENOENT;
10959         }
10960
10961         offset = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
10962         read_extent_buffer(path->nodes[0], &ri, offset, sizeof(ri));
10963
10964         if (btrfs_root_bytenr(&ri) != rii->bytenr ||
10965             btrfs_root_level(&ri) != rii->level ||
10966             btrfs_root_generation(&ri) != rii->gen) {
10967
10968                 /*
10969                  * If we're in repair mode but our caller told us to not update
10970                  * the root item, i.e. just check if it needs to be updated, don't
10971                  * print this message, since the caller will call us again shortly
10972                  * for the same root item without read only mode (the caller will
10973                  * open a transaction first).
10974                  */
10975                 if (!(read_only_mode && repair))
10976                         fprintf(stderr,
10977                                 "%sroot item for root %llu,"
10978                                 " current bytenr %llu, current gen %llu, current level %u,"
10979                                 " new bytenr %llu, new gen %llu, new level %u\n",
10980                                 (read_only_mode ? "" : "fixing "),
10981                                 root_id,
10982                                 btrfs_root_bytenr(&ri), btrfs_root_generation(&ri),
10983                                 btrfs_root_level(&ri),
10984                                 rii->bytenr, rii->gen, rii->level);
10985
10986                 if (btrfs_root_generation(&ri) > rii->gen) {
10987                         fprintf(stderr,
10988                                 "root %llu has a root item with a more recent gen (%llu) compared to the found root node (%llu)\n",
10989                                 root_id, btrfs_root_generation(&ri), rii->gen);
10990                         return -EINVAL;
10991                 }
10992
10993                 if (!read_only_mode) {
10994                         btrfs_set_root_bytenr(&ri, rii->bytenr);
10995                         btrfs_set_root_level(&ri, rii->level);
10996                         btrfs_set_root_generation(&ri, rii->gen);
10997                         write_extent_buffer(path->nodes[0], &ri,
10998                                             offset, sizeof(ri));
10999                 }
11000
11001                 return 1;
11002         }
11003
11004         return 0;
11005 }
11006
11007 /*
11008  * A regression introduced in the 3.17 kernel (more specifically in 3.17-rc2),
11009  * caused read-only snapshots to be corrupted if they were created at a moment
11010  * when the source subvolume/snapshot had orphan items. The issue was that the
11011  * on-disk root items became incorrect, referring to the pre orphan cleanup root
11012  * node instead of the post orphan cleanup root node.
11013  * So this function, and its callees, just detects and fixes those cases. Even
11014  * though the regression was for read-only snapshots, this function applies to
11015  * any snapshot/subvolume root.
11016  * This must be run before any other repair code - not doing it so, makes other
11017  * repair code delete or modify backrefs in the extent tree for example, which
11018  * will result in an inconsistent fs after repairing the root items.
11019  */
11020 static int repair_root_items(struct btrfs_fs_info *info)
11021 {
11022         struct btrfs_path path;
11023         struct btrfs_key key;
11024         struct extent_buffer *leaf;
11025         struct btrfs_trans_handle *trans = NULL;
11026         int ret = 0;
11027         int bad_roots = 0;
11028         int need_trans = 0;
11029
11030         btrfs_init_path(&path);
11031
11032         ret = build_roots_info_cache(info);
11033         if (ret)
11034                 goto out;
11035
11036         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
11037         key.type = BTRFS_ROOT_ITEM_KEY;
11038         key.offset = 0;
11039
11040 again:
11041         /*
11042          * Avoid opening and committing transactions if a leaf doesn't have
11043          * any root items that need to be fixed, so that we avoid rotating
11044          * backup roots unnecessarily.
11045          */
11046         if (need_trans) {
11047                 trans = btrfs_start_transaction(info->tree_root, 1);
11048                 if (IS_ERR(trans)) {
11049                         ret = PTR_ERR(trans);
11050                         goto out;
11051                 }
11052         }
11053
11054         ret = btrfs_search_slot(trans, info->tree_root, &key, &path,
11055                                 0, trans ? 1 : 0);
11056         if (ret < 0)
11057                 goto out;
11058         leaf = path.nodes[0];
11059
11060         while (1) {
11061                 struct btrfs_key found_key;
11062
11063                 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
11064                         int no_more_keys = find_next_key(&path, &key);
11065
11066                         btrfs_release_path(&path);
11067                         if (trans) {
11068                                 ret = btrfs_commit_transaction(trans,
11069                                                                info->tree_root);
11070                                 trans = NULL;
11071                                 if (ret < 0)
11072                                         goto out;
11073                         }
11074                         need_trans = 0;
11075                         if (no_more_keys)
11076                                 break;
11077                         goto again;
11078                 }
11079
11080                 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
11081
11082                 if (found_key.type != BTRFS_ROOT_ITEM_KEY)
11083                         goto next;
11084                 if (found_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
11085                         goto next;
11086
11087                 ret = maybe_repair_root_item(info, &path, &found_key,
11088                                              trans ? 0 : 1);
11089                 if (ret < 0)
11090                         goto out;
11091                 if (ret) {
11092                         if (!trans && repair) {
11093                                 need_trans = 1;
11094                                 key = found_key;
11095                                 btrfs_release_path(&path);
11096                                 goto again;
11097                         }
11098                         bad_roots++;
11099                 }
11100 next:
11101                 path.slots[0]++;
11102         }
11103         ret = 0;
11104 out:
11105         free_roots_info_cache();
11106         btrfs_release_path(&path);
11107         if (trans)
11108                 btrfs_commit_transaction(trans, info->tree_root);
11109         if (ret < 0)
11110                 return ret;
11111
11112         return bad_roots;
11113 }
11114
11115 static int clear_free_space_cache(struct btrfs_fs_info *fs_info)
11116 {
11117         struct btrfs_trans_handle *trans;
11118         struct btrfs_block_group_cache *bg_cache;
11119         u64 current = 0;
11120         int ret = 0;
11121
11122         /* Clear all free space cache inodes and its extent data */
11123         while (1) {
11124                 bg_cache = btrfs_lookup_first_block_group(fs_info, current);
11125                 if (!bg_cache)
11126                         break;
11127                 ret = btrfs_clear_free_space_cache(fs_info, bg_cache);
11128                 if (ret < 0)
11129                         return ret;
11130                 current = bg_cache->key.objectid + bg_cache->key.offset;
11131         }
11132
11133         /* Don't forget to set cache_generation to -1 */
11134         trans = btrfs_start_transaction(fs_info->tree_root, 0);
11135         if (IS_ERR(trans)) {
11136                 error("failed to update super block cache generation");
11137                 return PTR_ERR(trans);
11138         }
11139         btrfs_set_super_cache_generation(fs_info->super_copy, (u64)-1);
11140         btrfs_commit_transaction(trans, fs_info->tree_root);
11141
11142         return ret;
11143 }
11144
11145 const char * const cmd_check_usage[] = {
11146         "btrfs check [options] <device>",
11147         "Check structural integrity of a filesystem (unmounted).",
11148         "Check structural integrity of an unmounted filesystem. Verify internal",
11149         "trees' consistency and item connectivity. In the repair mode try to",
11150         "fix the problems found. ",
11151         "WARNING: the repair mode is considered dangerous",
11152         "",
11153         "-s|--super <superblock>     use this superblock copy",
11154         "-b|--backup                 use the first valid backup root copy",
11155         "--repair                    try to repair the filesystem",
11156         "--readonly                  run in read-only mode (default)",
11157         "--init-csum-tree            create a new CRC tree",
11158         "--init-extent-tree          create a new extent tree",
11159         "--mode <MODE>               allows choice of memory/IO trade-offs",
11160         "                            where MODE is one of:",
11161         "                            original - read inodes and extents to memory (requires",
11162         "                                       more memory, does less IO)",
11163         "                            lowmem   - try to use less memory but read blocks again",
11164         "                                       when needed",
11165         "--check-data-csum           verify checksums of data blocks",
11166         "-Q|--qgroup-report          print a report on qgroup consistency",
11167         "-E|--subvol-extents <subvolid>",
11168         "                            print subvolume extents and sharing state",
11169         "-r|--tree-root <bytenr>     use the given bytenr for the tree root",
11170         "--chunk-root <bytenr>       use the given bytenr for the chunk tree root",
11171         "-p|--progress               indicate progress",
11172         "--clear-space-cache v1|v2   clear space cache for v1 or v2",
11173         "                            NOTE: v1 support implemented",
11174         NULL
11175 };
11176
11177 int cmd_check(int argc, char **argv)
11178 {
11179         struct cache_tree root_cache;
11180         struct btrfs_root *root;
11181         struct btrfs_fs_info *info;
11182         u64 bytenr = 0;
11183         u64 subvolid = 0;
11184         u64 tree_root_bytenr = 0;
11185         u64 chunk_root_bytenr = 0;
11186         char uuidbuf[BTRFS_UUID_UNPARSED_SIZE];
11187         int ret;
11188         u64 num;
11189         int init_csum_tree = 0;
11190         int readonly = 0;
11191         int clear_space_cache = 0;
11192         int qgroup_report = 0;
11193         int qgroups_repaired = 0;
11194         unsigned ctree_flags = OPEN_CTREE_EXCLUSIVE;
11195
11196         while(1) {
11197                 int c;
11198                 enum { GETOPT_VAL_REPAIR = 257, GETOPT_VAL_INIT_CSUM,
11199                         GETOPT_VAL_INIT_EXTENT, GETOPT_VAL_CHECK_CSUM,
11200                         GETOPT_VAL_READONLY, GETOPT_VAL_CHUNK_TREE,
11201                         GETOPT_VAL_MODE, GETOPT_VAL_CLEAR_SPACE_CACHE };
11202                 static const struct option long_options[] = {
11203                         { "super", required_argument, NULL, 's' },
11204                         { "repair", no_argument, NULL, GETOPT_VAL_REPAIR },
11205                         { "readonly", no_argument, NULL, GETOPT_VAL_READONLY },
11206                         { "init-csum-tree", no_argument, NULL,
11207                                 GETOPT_VAL_INIT_CSUM },
11208                         { "init-extent-tree", no_argument, NULL,
11209                                 GETOPT_VAL_INIT_EXTENT },
11210                         { "check-data-csum", no_argument, NULL,
11211                                 GETOPT_VAL_CHECK_CSUM },
11212                         { "backup", no_argument, NULL, 'b' },
11213                         { "subvol-extents", required_argument, NULL, 'E' },
11214                         { "qgroup-report", no_argument, NULL, 'Q' },
11215                         { "tree-root", required_argument, NULL, 'r' },
11216                         { "chunk-root", required_argument, NULL,
11217                                 GETOPT_VAL_CHUNK_TREE },
11218                         { "progress", no_argument, NULL, 'p' },
11219                         { "mode", required_argument, NULL,
11220                                 GETOPT_VAL_MODE },
11221                         { "clear-space-cache", required_argument, NULL,
11222                                 GETOPT_VAL_CLEAR_SPACE_CACHE},
11223                         { NULL, 0, NULL, 0}
11224                 };
11225
11226                 c = getopt_long(argc, argv, "as:br:p", long_options, NULL);
11227                 if (c < 0)
11228                         break;
11229                 switch(c) {
11230                         case 'a': /* ignored */ break;
11231                         case 'b':
11232                                 ctree_flags |= OPEN_CTREE_BACKUP_ROOT;
11233                                 break;
11234                         case 's':
11235                                 num = arg_strtou64(optarg);
11236                                 if (num >= BTRFS_SUPER_MIRROR_MAX) {
11237                                         error(
11238                                         "super mirror should be less than %d",
11239                                                 BTRFS_SUPER_MIRROR_MAX);
11240                                         exit(1);
11241                                 }
11242                                 bytenr = btrfs_sb_offset(((int)num));
11243                                 printf("using SB copy %llu, bytenr %llu\n", num,
11244                                        (unsigned long long)bytenr);
11245                                 break;
11246                         case 'Q':
11247                                 qgroup_report = 1;
11248                                 break;
11249                         case 'E':
11250                                 subvolid = arg_strtou64(optarg);
11251                                 break;
11252                         case 'r':
11253                                 tree_root_bytenr = arg_strtou64(optarg);
11254                                 break;
11255                         case GETOPT_VAL_CHUNK_TREE:
11256                                 chunk_root_bytenr = arg_strtou64(optarg);
11257                                 break;
11258                         case 'p':
11259                                 ctx.progress_enabled = true;
11260                                 break;
11261                         case '?':
11262                         case 'h':
11263                                 usage(cmd_check_usage);
11264                         case GETOPT_VAL_REPAIR:
11265                                 printf("enabling repair mode\n");
11266                                 repair = 1;
11267                                 ctree_flags |= OPEN_CTREE_WRITES;
11268                                 break;
11269                         case GETOPT_VAL_READONLY:
11270                                 readonly = 1;
11271                                 break;
11272                         case GETOPT_VAL_INIT_CSUM:
11273                                 printf("Creating a new CRC tree\n");
11274                                 init_csum_tree = 1;
11275                                 repair = 1;
11276                                 ctree_flags |= OPEN_CTREE_WRITES;
11277                                 break;
11278                         case GETOPT_VAL_INIT_EXTENT:
11279                                 init_extent_tree = 1;
11280                                 ctree_flags |= (OPEN_CTREE_WRITES |
11281                                                 OPEN_CTREE_NO_BLOCK_GROUPS);
11282                                 repair = 1;
11283                                 break;
11284                         case GETOPT_VAL_CHECK_CSUM:
11285                                 check_data_csum = 1;
11286                                 break;
11287                         case GETOPT_VAL_MODE:
11288                                 check_mode = parse_check_mode(optarg);
11289                                 if (check_mode == CHECK_MODE_UNKNOWN) {
11290                                         error("unknown mode: %s", optarg);
11291                                         exit(1);
11292                                 }
11293                                 break;
11294                         case GETOPT_VAL_CLEAR_SPACE_CACHE:
11295                                 if (strcmp(optarg, "v1") != 0) {
11296                                         error(
11297                         "only v1 support implmented, unrecognized value %s",
11298                         optarg);
11299                                         exit(1);
11300                                 }
11301                                 clear_space_cache = 1;
11302                                 ctree_flags |= OPEN_CTREE_WRITES;
11303                                 break;
11304                 }
11305         }
11306
11307         if (check_argc_exact(argc - optind, 1))
11308                 usage(cmd_check_usage);
11309
11310         if (ctx.progress_enabled) {
11311                 ctx.tp = TASK_NOTHING;
11312                 ctx.info = task_init(print_status_check, print_status_return, &ctx);
11313         }
11314
11315         /* This check is the only reason for --readonly to exist */
11316         if (readonly && repair) {
11317                 error("repair options are not compatible with --readonly");
11318                 exit(1);
11319         }
11320
11321         /*
11322          * Not supported yet
11323          */
11324         if (repair && check_mode == CHECK_MODE_LOWMEM) {
11325                 error("low memory mode doesn't support repair yet");
11326                 exit(1);
11327         }
11328
11329         radix_tree_init();
11330         cache_tree_init(&root_cache);
11331
11332         if((ret = check_mounted(argv[optind])) < 0) {
11333                 error("could not check mount status: %s", strerror(-ret));
11334                 goto err_out;
11335         } else if(ret) {
11336                 error("%s is currently mounted, aborting", argv[optind]);
11337                 ret = -EBUSY;
11338                 goto err_out;
11339         }
11340
11341         /* only allow partial opening under repair mode */
11342         if (repair)
11343                 ctree_flags |= OPEN_CTREE_PARTIAL;
11344
11345         info = open_ctree_fs_info(argv[optind], bytenr, tree_root_bytenr,
11346                                   chunk_root_bytenr, ctree_flags);
11347         if (!info) {
11348                 error("cannot open file system");
11349                 ret = -EIO;
11350                 goto err_out;
11351         }
11352
11353         global_info = info;
11354         root = info->fs_root;
11355         if (clear_space_cache) {
11356                 if (btrfs_fs_compat_ro(info,
11357                                 BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE)) {
11358                         error(
11359                         "free space cache v2 detected, clearing not implemented");
11360                         ret = 1;
11361                         goto close_out;
11362                 }
11363                 printf("Clearing free space cache\n");
11364                 ret = clear_free_space_cache(info);
11365                 if (ret) {
11366                         error("failed to clear free space cache");
11367                         ret = 1;
11368                 } else {
11369                         printf("Free space cache cleared\n");
11370                 }
11371                 goto close_out;
11372         }
11373
11374         /*
11375          * repair mode will force us to commit transaction which
11376          * will make us fail to load log tree when mounting.
11377          */
11378         if (repair && btrfs_super_log_root(info->super_copy)) {
11379                 ret = ask_user("repair mode will force to clear out log tree, are you sure?");
11380                 if (!ret) {
11381                         ret = 1;
11382                         goto close_out;
11383                 }
11384                 ret = zero_log_tree(root);
11385                 if (ret) {
11386                         error("failed to zero log tree: %d", ret);
11387                         goto close_out;
11388                 }
11389         }
11390
11391         uuid_unparse(info->super_copy->fsid, uuidbuf);
11392         if (qgroup_report) {
11393                 printf("Print quota groups for %s\nUUID: %s\n", argv[optind],
11394                        uuidbuf);
11395                 ret = qgroup_verify_all(info);
11396                 if (ret == 0)
11397                         report_qgroups(1);
11398                 goto close_out;
11399         }
11400         if (subvolid) {
11401                 printf("Print extent state for subvolume %llu on %s\nUUID: %s\n",
11402                        subvolid, argv[optind], uuidbuf);
11403                 ret = print_extent_state(info, subvolid);
11404                 goto close_out;
11405         }
11406         printf("Checking filesystem on %s\nUUID: %s\n", argv[optind], uuidbuf);
11407
11408         if (!extent_buffer_uptodate(info->tree_root->node) ||
11409             !extent_buffer_uptodate(info->dev_root->node) ||
11410             !extent_buffer_uptodate(info->chunk_root->node)) {
11411                 error("critical roots corrupted, unable to check the filesystem");
11412                 ret = -EIO;
11413                 goto close_out;
11414         }
11415
11416         if (init_extent_tree || init_csum_tree) {
11417                 struct btrfs_trans_handle *trans;
11418
11419                 trans = btrfs_start_transaction(info->extent_root, 0);
11420                 if (IS_ERR(trans)) {
11421                         error("error starting transaction");
11422                         ret = PTR_ERR(trans);
11423                         goto close_out;
11424                 }
11425
11426                 if (init_extent_tree) {
11427                         printf("Creating a new extent tree\n");
11428                         ret = reinit_extent_tree(trans, info);
11429                         if (ret)
11430                                 goto close_out;
11431                 }
11432
11433                 if (init_csum_tree) {
11434                         printf("Reinitialize checksum tree\n");
11435                         ret = btrfs_fsck_reinit_root(trans, info->csum_root, 0);
11436                         if (ret) {
11437                                 error("checksum tree initialization failed: %d",
11438                                                 ret);
11439                                 ret = -EIO;
11440                                 goto close_out;
11441                         }
11442
11443                         ret = fill_csum_tree(trans, info->csum_root,
11444                                              init_extent_tree);
11445                         if (ret) {
11446                                 error("checksum tree refilling failed: %d", ret);
11447                                 return -EIO;
11448                         }
11449                 }
11450                 /*
11451                  * Ok now we commit and run the normal fsck, which will add
11452                  * extent entries for all of the items it finds.
11453                  */
11454                 ret = btrfs_commit_transaction(trans, info->extent_root);
11455                 if (ret)
11456                         goto close_out;
11457         }
11458         if (!extent_buffer_uptodate(info->extent_root->node)) {
11459                 error("critical: extent_root, unable to check the filesystem");
11460                 ret = -EIO;
11461                 goto close_out;
11462         }
11463         if (!extent_buffer_uptodate(info->csum_root->node)) {
11464                 error("critical: csum_root, unable to check the filesystem");
11465                 ret = -EIO;
11466                 goto close_out;
11467         }
11468
11469         if (!ctx.progress_enabled)
11470                 printf("checking extents");
11471         if (check_mode == CHECK_MODE_LOWMEM)
11472                 ret = check_chunks_and_extents_v2(root);
11473         else
11474                 ret = check_chunks_and_extents(root);
11475         if (ret)
11476                 printf("Errors found in extent allocation tree or chunk allocation");
11477
11478         ret = repair_root_items(info);
11479         if (ret < 0)
11480                 goto close_out;
11481         if (repair) {
11482                 fprintf(stderr, "Fixed %d roots.\n", ret);
11483                 ret = 0;
11484         } else if (ret > 0) {
11485                 fprintf(stderr,
11486                        "Found %d roots with an outdated root item.\n",
11487                        ret);
11488                 fprintf(stderr,
11489                         "Please run a filesystem check with the option --repair to fix them.\n");
11490                 ret = 1;
11491                 goto close_out;
11492         }
11493
11494         if (!ctx.progress_enabled) {
11495                 if (btrfs_fs_compat_ro(info, BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE))
11496                         fprintf(stderr, "checking free space tree\n");
11497                 else
11498                         fprintf(stderr, "checking free space cache\n");
11499         }
11500         ret = check_space_cache(root);
11501         if (ret)
11502                 goto out;
11503
11504         /*
11505          * We used to have to have these hole extents in between our real
11506          * extents so if we don't have this flag set we need to make sure there
11507          * are no gaps in the file extents for inodes, otherwise we can just
11508          * ignore it when this happens.
11509          */
11510         no_holes = btrfs_fs_incompat(root->fs_info,
11511                                      BTRFS_FEATURE_INCOMPAT_NO_HOLES);
11512         if (!ctx.progress_enabled)
11513                 fprintf(stderr, "checking fs roots\n");
11514         ret = check_fs_roots(root, &root_cache);
11515         if (ret)
11516                 goto out;
11517
11518         fprintf(stderr, "checking csums\n");
11519         ret = check_csums(root);
11520         if (ret)
11521                 goto out;
11522
11523         fprintf(stderr, "checking root refs\n");
11524         ret = check_root_refs(root, &root_cache);
11525         if (ret)
11526                 goto out;
11527
11528         while (repair && !list_empty(&root->fs_info->recow_ebs)) {
11529                 struct extent_buffer *eb;
11530
11531                 eb = list_first_entry(&root->fs_info->recow_ebs,
11532                                       struct extent_buffer, recow);
11533                 list_del_init(&eb->recow);
11534                 ret = recow_extent_buffer(root, eb);
11535                 if (ret)
11536                         break;
11537         }
11538
11539         while (!list_empty(&delete_items)) {
11540                 struct bad_item *bad;
11541
11542                 bad = list_first_entry(&delete_items, struct bad_item, list);
11543                 list_del_init(&bad->list);
11544                 if (repair)
11545                         ret = delete_bad_item(root, bad);
11546                 free(bad);
11547         }
11548
11549         if (info->quota_enabled) {
11550                 int err;
11551                 fprintf(stderr, "checking quota groups\n");
11552                 err = qgroup_verify_all(info);
11553                 if (err)
11554                         goto out;
11555                 report_qgroups(0);
11556                 err = repair_qgroups(info, &qgroups_repaired);
11557                 if (err)
11558                         goto out;
11559         }
11560
11561         if (!list_empty(&root->fs_info->recow_ebs)) {
11562                 error("transid errors in file system");
11563                 ret = 1;
11564         }
11565 out:
11566         /* Don't override original ret */
11567         if (!ret && qgroups_repaired)
11568                 ret = qgroups_repaired;
11569
11570         if (found_old_backref) { /*
11571                  * there was a disk format change when mixed
11572                  * backref was in testing tree. The old format
11573                  * existed about one week.
11574                  */
11575                 printf("\n * Found old mixed backref format. "
11576                        "The old format is not supported! *"
11577                        "\n * Please mount the FS in readonly mode, "
11578                        "backup data and re-format the FS. *\n\n");
11579                 ret = 1;
11580         }
11581         printf("found %llu bytes used err is %d\n",
11582                (unsigned long long)bytes_used, ret);
11583         printf("total csum bytes: %llu\n",(unsigned long long)total_csum_bytes);
11584         printf("total tree bytes: %llu\n",
11585                (unsigned long long)total_btree_bytes);
11586         printf("total fs tree bytes: %llu\n",
11587                (unsigned long long)total_fs_tree_bytes);
11588         printf("total extent tree bytes: %llu\n",
11589                (unsigned long long)total_extent_tree_bytes);
11590         printf("btree space waste bytes: %llu\n",
11591                (unsigned long long)btree_space_waste);
11592         printf("file data blocks allocated: %llu\n referenced %llu\n",
11593                 (unsigned long long)data_bytes_allocated,
11594                 (unsigned long long)data_bytes_referenced);
11595
11596         free_qgroup_counts();
11597         free_root_recs_tree(&root_cache);
11598 close_out:
11599         close_ctree(root);
11600 err_out:
11601         if (ctx.progress_enabled)
11602                 task_deinit(ctx.info);
11603
11604         return ret;
11605 }