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