chunk-recover.c

   1 /*
   2  * Copyright (C) 2013 FUJITSU LIMITED.  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 "kerncompat.h"
  20 #include "androidcompat.h"
  21
  22 #include <stdio.h>
  23 #include <stdio_ext.h>
  24 #include <stdlib.h>
  25 #include <sys/types.h>
  26 #include <sys/stat.h>
  27 #include <fcntl.h>
  28 #include <unistd.h>
  29 #include <uuid/uuid.h>
  30 #include <pthread.h>
  31
  32 #include "list.h"
  33 #include "radix-tree.h"
  34 #include "ctree.h"
  35 #include "extent-cache.h"
  36 #include "disk-io.h"
  37 #include "volumes.h"
  38 #include "transaction.h"
  39 #include "crc32c.h"
  40 #include "utils.h"
  41 #include "btrfsck.h"
  42 #include "commands.h"
  43
  44 struct recover_control {
  45         int verbose;
  46         int yes;
  47
  48         u16 csum_size;
  49         u32 sectorsize;
  50         u32 leafsize;
  51         u64 generation;
  52         u64 chunk_root_generation;
  53
  54         struct btrfs_fs_devices *fs_devices;
  55
  56         struct cache_tree chunk;
  57         struct block_group_tree bg;
  58         struct device_extent_tree devext;
  59         struct cache_tree eb_cache;
  60
  61         struct list_head good_chunks;
  62         struct list_head bad_chunks;
  63         struct list_head rebuild_chunks;
  64         struct list_head unrepaired_chunks;
  65         pthread_mutex_t rc_lock;
  66 };
  67
  68 struct extent_record {
  69         struct cache_extent cache;
  70         u64 generation;
  71         u8 csum[BTRFS_CSUM_SIZE];
  72         struct btrfs_device *devices[BTRFS_MAX_MIRRORS];
  73         u64 offsets[BTRFS_MAX_MIRRORS];
  74         int nmirrors;
  75 };
  76
  77 struct device_scan {
  78         struct recover_control *rc;
  79         struct btrfs_device *dev;
  80         int fd;
  81 };
  82
  83 static struct extent_record *btrfs_new_extent_record(struct extent_buffer *eb)
  84 {
  85         struct extent_record *rec;
  86
  87         rec = malloc(sizeof(*rec));
  88         if (!rec) {
  89                 fprintf(stderr, "Fail to allocate memory for extent record.\n");
  90                 exit(1);
  91         }
  92
  93         memset(rec, 0, sizeof(*rec));
  94         rec->cache.start = btrfs_header_bytenr(eb);
  95         rec->cache.size = eb->len;
  96         rec->generation = btrfs_header_generation(eb);
  97         read_extent_buffer(eb, rec->csum, (unsigned long)btrfs_header_csum(eb),
  98                            BTRFS_CSUM_SIZE);
  99         return rec;
 100 }
 101
 102 static int process_extent_buffer(struct cache_tree *eb_cache,
 103                                  struct extent_buffer *eb,
 104                                  struct btrfs_device *device, u64 offset)
 105 {
 106         struct extent_record *rec;
 107         struct extent_record *exist;
 108         struct cache_extent *cache;
 109         int ret = 0;
 110
 111         rec = btrfs_new_extent_record(eb);
 112         if (!rec->cache.size)
 113                 goto free_out;
 114 again:
 115         cache = lookup_cache_extent(eb_cache,
 116                                     rec->cache.start,
 117                                     rec->cache.size);
 118         if (cache) {
 119                 exist = container_of(cache, struct extent_record, cache);
 120
 121                 if (exist->generation > rec->generation)
 122                         goto free_out;
 123                 if (exist->generation == rec->generation) {
 124                         if (exist->cache.start != rec->cache.start ||
 125                             exist->cache.size != rec->cache.size ||
 126                             memcmp(exist->csum, rec->csum, BTRFS_CSUM_SIZE)) {
 127                                 ret = -EEXIST;
 128                         } else {
 129                                 BUG_ON(exist->nmirrors >= BTRFS_MAX_MIRRORS);
 130                                 exist->devices[exist->nmirrors] = device;
 131                                 exist->offsets[exist->nmirrors] = offset;
 132                                 exist->nmirrors++;
 133                         }
 134                         goto free_out;
 135                 }
 136                 remove_cache_extent(eb_cache, cache);
 137                 free(exist);
 138                 goto again;
 139         }
 140
 141         rec->devices[0] = device;
 142         rec->offsets[0] = offset;
 143         rec->nmirrors++;
 144         ret = insert_cache_extent(eb_cache, &rec->cache);
 145         BUG_ON(ret);
 146 out:
 147         return ret;
 148 free_out:
 149         free(rec);
 150         goto out;
 151 }
 152
 153 static void free_extent_record(struct cache_extent *cache)
 154 {
 155         struct extent_record *er;
 156
 157         er = container_of(cache, struct extent_record, cache);
 158         free(er);
 159 }
 160
 161 FREE_EXTENT_CACHE_BASED_TREE(extent_record, free_extent_record);
 162
 163 static struct btrfs_chunk *create_chunk_item(struct chunk_record *record)
 164 {
 165         struct btrfs_chunk *ret;
 166         struct btrfs_stripe *chunk_stripe;
 167         int i;
 168
 169         if (!record || record->num_stripes == 0)
 170                 return NULL;
 171         ret = malloc(btrfs_chunk_item_size(record->num_stripes));
 172         if (!ret)
 173                 return NULL;
 174         btrfs_set_stack_chunk_length(ret, record->length);
 175         btrfs_set_stack_chunk_owner(ret, record->owner);
 176         btrfs_set_stack_chunk_stripe_len(ret, record->stripe_len);
 177         btrfs_set_stack_chunk_type(ret, record->type_flags);
 178         btrfs_set_stack_chunk_io_align(ret, record->io_align);
 179         btrfs_set_stack_chunk_io_width(ret, record->io_width);
 180         btrfs_set_stack_chunk_sector_size(ret, record->sector_size);
 181         btrfs_set_stack_chunk_num_stripes(ret, record->num_stripes);
 182         btrfs_set_stack_chunk_sub_stripes(ret, record->sub_stripes);
 183         for (i = 0, chunk_stripe = &ret->stripe; i < record->num_stripes;
 184              i++, chunk_stripe++) {
 185                 btrfs_set_stack_stripe_devid(chunk_stripe,
 186                                 record->stripes[i].devid);
 187                 btrfs_set_stack_stripe_offset(chunk_stripe,
 188                                 record->stripes[i].offset);
 189                 memcpy(chunk_stripe->dev_uuid, record->stripes[i].dev_uuid,
 190                        BTRFS_UUID_SIZE);
 191         }
 192         return ret;
 193 }
 194
 195 static void init_recover_control(struct recover_control *rc, int verbose,
 196                 int yes)
 197 {
 198         memset(rc, 0, sizeof(struct recover_control));
 199         cache_tree_init(&rc->chunk);
 200         cache_tree_init(&rc->eb_cache);
 201         block_group_tree_init(&rc->bg);
 202         device_extent_tree_init(&rc->devext);
 203
 204         INIT_LIST_HEAD(&rc->good_chunks);
 205         INIT_LIST_HEAD(&rc->bad_chunks);
 206         INIT_LIST_HEAD(&rc->rebuild_chunks);
 207         INIT_LIST_HEAD(&rc->unrepaired_chunks);
 208
 209         rc->verbose = verbose;
 210         rc->yes = yes;
 211         pthread_mutex_init(&rc->rc_lock, NULL);
 212 }
 213
 214 static void free_recover_control(struct recover_control *rc)
 215 {
 216         free_block_group_tree(&rc->bg);
 217         free_chunk_cache_tree(&rc->chunk);
 218         free_device_extent_tree(&rc->devext);
 219         free_extent_record_tree(&rc->eb_cache);
 220         pthread_mutex_destroy(&rc->rc_lock);
 221 }
 222
 223 static int process_block_group_item(struct block_group_tree *bg_cache,
 224                                     struct extent_buffer *leaf,
 225                                     struct btrfs_key *key, int slot)
 226 {
 227         struct block_group_record *rec;
 228         struct block_group_record *exist;
 229         struct cache_extent *cache;
 230         int ret = 0;
 231
 232         rec = btrfs_new_block_group_record(leaf, key, slot);
 233         if (!rec->cache.size)
 234                 goto free_out;
 235 again:
 236         cache = lookup_cache_extent(&bg_cache->tree,
 237                                     rec->cache.start,
 238                                     rec->cache.size);
 239         if (cache) {
 240                 exist = container_of(cache, struct block_group_record, cache);
 241
 242                 /*check the generation and replace if needed*/
 243                 if (exist->generation > rec->generation)
 244                         goto free_out;
 245                 if (exist->generation == rec->generation) {
 246                         int offset = offsetof(struct block_group_record,
 247                                               generation);
 248                         /*
 249                          * According to the current kernel code, the following
 250                          * case is impossble, or there is something wrong in
 251                          * the kernel code.
 252                          */
 253                         if (memcmp(((void *)exist) + offset,
 254                                    ((void *)rec) + offset,
 255                                    sizeof(*rec) - offset))
 256                                 ret = -EEXIST;
 257                         goto free_out;
 258                 }
 259                 remove_cache_extent(&bg_cache->tree, cache);
 260                 list_del_init(&exist->list);
 261                 free(exist);
 262                 /*
 263                  * We must do seach again to avoid the following cache.
 264                  * /--old bg 1--//--old bg 2--/
 265                  *        /--new bg--/
 266                  */
 267                 goto again;
 268         }
 269
 270         ret = insert_block_group_record(bg_cache, rec);
 271         BUG_ON(ret);
 272 out:
 273         return ret;
 274 free_out:
 275         free(rec);
 276         goto out;
 277 }
 278
 279 static int process_chunk_item(struct cache_tree *chunk_cache,
 280                               struct extent_buffer *leaf, struct btrfs_key *key,
 281                               int slot)
 282 {
 283         struct chunk_record *rec;
 284         struct chunk_record *exist;
 285         struct cache_extent *cache;
 286         int ret = 0;
 287
 288         rec = btrfs_new_chunk_record(leaf, key, slot);
 289         if (!rec->cache.size)
 290                 goto free_out;
 291 again:
 292         cache = lookup_cache_extent(chunk_cache, rec->offset, rec->length);
 293         if (cache) {
 294                 exist = container_of(cache, struct chunk_record, cache);
 295
 296                 if (exist->generation > rec->generation)
 297                         goto free_out;
 298                 if (exist->generation == rec->generation) {
 299                         int num_stripes = rec->num_stripes;
 300                         int rec_size = btrfs_chunk_record_size(num_stripes);
 301                         int offset = offsetof(struct chunk_record, generation);
 302
 303                         if (exist->num_stripes != rec->num_stripes ||
 304                             memcmp(((void *)exist) + offset,
 305                                    ((void *)rec) + offset,
 306                                    rec_size - offset))
 307                                 ret = -EEXIST;
 308                         goto free_out;
 309                 }
 310                 remove_cache_extent(chunk_cache, cache);
 311                 free(exist);
 312                 goto again;
 313         }
 314         ret = insert_cache_extent(chunk_cache, &rec->cache);
 315         BUG_ON(ret);
 316 out:
 317         return ret;
 318 free_out:
 319         free(rec);
 320         goto out;
 321 }
 322
 323 static int process_device_extent_item(struct device_extent_tree *devext_cache,
 324                                       struct extent_buffer *leaf,
 325                                       struct btrfs_key *key, int slot)
 326 {
 327         struct device_extent_record *rec;
 328         struct device_extent_record *exist;
 329         struct cache_extent *cache;
 330         int ret = 0;
 331
 332         rec = btrfs_new_device_extent_record(leaf, key, slot);
 333         if (!rec->cache.size)
 334                 goto free_out;
 335 again:
 336         cache = lookup_cache_extent2(&devext_cache->tree,
 337                                      rec->cache.objectid,
 338                                      rec->cache.start,
 339                                      rec->cache.size);
 340         if (cache) {
 341                 exist = container_of(cache, struct device_extent_record, cache);
 342                 if (exist->generation > rec->generation)
 343                         goto free_out;
 344                 if (exist->generation == rec->generation) {
 345                         int offset = offsetof(struct device_extent_record,
 346                                               generation);
 347                         if (memcmp(((void *)exist) + offset,
 348                                    ((void *)rec) + offset,
 349                                    sizeof(*rec) - offset))
 350                                 ret = -EEXIST;
 351                         goto free_out;
 352                 }
 353                 remove_cache_extent(&devext_cache->tree, cache);
 354                 list_del_init(&exist->chunk_list);
 355                 list_del_init(&exist->device_list);
 356                 free(exist);
 357                 goto again;
 358         }
 359
 360         ret = insert_device_extent_record(devext_cache, rec);
 361         BUG_ON(ret);
 362 out:
 363         return ret;
 364 free_out:
 365         free(rec);
 366         goto out;
 367 }
 368
 369 static void print_block_group_info(struct block_group_record *rec, char *prefix)
 370 {
 371         if (prefix)
 372                 printf("%s", prefix);
 373         printf("Block Group: start = %llu, len = %llu, flag = %llx\n",
 374                rec->objectid, rec->offset, rec->flags);
 375 }
 376
 377 static void print_block_group_tree(struct block_group_tree *tree)
 378 {
 379         struct cache_extent *cache;
 380         struct block_group_record *rec;
 381
 382         printf("All Block Groups:\n");
 383         for (cache = first_cache_extent(&tree->tree); cache;
 384              cache = next_cache_extent(cache)) {
 385                 rec = container_of(cache, struct block_group_record, cache);
 386                 print_block_group_info(rec, "\t");
 387         }
 388         printf("\n");
 389 }
 390
 391 static void print_stripe_info(struct stripe *data, char *prefix1, char *prefix2,
 392                               int index)
 393 {
 394         if (prefix1)
 395                 printf("%s", prefix1);
 396         if (prefix2)
 397                 printf("%s", prefix2);
 398         printf("[%2d] Stripe: devid = %llu, offset = %llu\n",
 399                index, data->devid, data->offset);
 400 }
 401
 402 static void print_chunk_self_info(struct chunk_record *rec, char *prefix)
 403 {
 404         int i;
 405
 406         if (prefix)
 407                 printf("%s", prefix);
 408         printf("Chunk: start = %llu, len = %llu, type = %llx, num_stripes = %u\n",
 409                rec->offset, rec->length, rec->type_flags, rec->num_stripes);
 410         if (prefix)
 411                 printf("%s", prefix);
 412         printf("    Stripes list:\n");
 413         for (i = 0; i < rec->num_stripes; i++)
 414                 print_stripe_info(&rec->stripes[i], prefix, "    ", i);
 415 }
 416
 417 static void print_chunk_tree(struct cache_tree *tree)
 418 {
 419         struct cache_extent *n;
 420         struct chunk_record *entry;
 421
 422         printf("All Chunks:\n");
 423         for (n = first_cache_extent(tree); n;
 424              n = next_cache_extent(n)) {
 425                 entry = container_of(n, struct chunk_record, cache);
 426                 print_chunk_self_info(entry, "\t");
 427         }
 428         printf("\n");
 429 }
 430
 431 static void print_device_extent_info(struct device_extent_record *rec,
 432                                      char *prefix)
 433 {
 434         if (prefix)
 435                 printf("%s", prefix);
 436         printf("Device extent: devid = %llu, start = %llu, len = %llu, chunk offset = %llu\n",
 437                rec->objectid, rec->offset, rec->length, rec->chunk_offset);
 438 }
 439
 440 static void print_device_extent_tree(struct device_extent_tree *tree)
 441 {
 442         struct cache_extent *n;
 443         struct device_extent_record *entry;
 444
 445         printf("All Device Extents:\n");
 446         for (n = first_cache_extent(&tree->tree); n;
 447              n = next_cache_extent(n)) {
 448                 entry = container_of(n, struct device_extent_record, cache);
 449                 print_device_extent_info(entry, "\t");
 450         }
 451         printf("\n");
 452 }
 453
 454 static void print_device_info(struct btrfs_device *device, char *prefix)
 455 {
 456         if (prefix)
 457                 printf("%s", prefix);
 458         printf("Device: id = %llu, name = %s\n",
 459                device->devid, device->name);
 460 }
 461
 462 static void print_all_devices(struct list_head *devices)
 463 {
 464         struct btrfs_device *dev;
 465
 466         printf("All Devices:\n");
 467         list_for_each_entry(dev, devices, dev_list)
 468                 print_device_info(dev, "\t");
 469         printf("\n");
 470 }
 471
 472 static void print_scan_result(struct recover_control *rc)
 473 {
 474         if (!rc->verbose)
 475                 return;
 476
 477         printf("DEVICE SCAN RESULT:\n");
 478         printf("Filesystem Information:\n");
 479         printf("\tsectorsize: %d\n", rc->sectorsize);
 480         printf("\tleafsize: %d\n", rc->leafsize);
 481         printf("\ttree root generation: %llu\n", rc->generation);
 482         printf("\tchunk root generation: %llu\n", rc->chunk_root_generation);
 483         printf("\n");
 484
 485         print_all_devices(&rc->fs_devices->devices);
 486         print_block_group_tree(&rc->bg);
 487         print_chunk_tree(&rc->chunk);
 488         print_device_extent_tree(&rc->devext);
 489 }
 490
 491 static void print_chunk_info(struct chunk_record *chunk, char *prefix)
 492 {
 493         struct device_extent_record *devext;
 494         int i;
 495
 496         print_chunk_self_info(chunk, prefix);
 497         if (prefix)
 498                 printf("%s", prefix);
 499         if (chunk->bg_rec)
 500                 print_block_group_info(chunk->bg_rec, "    ");
 501         else
 502                 printf("    No block group.\n");
 503         if (prefix)
 504                 printf("%s", prefix);
 505         if (list_empty(&chunk->dextents)) {
 506                 printf("    No device extent.\n");
 507         } else {
 508                 printf("    Device extent list:\n");
 509                 i = 0;
 510                 list_for_each_entry(devext, &chunk->dextents, chunk_list) {
 511                         if (prefix)
 512                                 printf("%s", prefix);
 513                         printf("%s[%2d]", "        ", i);
 514                         print_device_extent_info(devext, NULL);
 515                         i++;
 516                 }
 517         }
 518 }
 519
 520 static void print_check_result(struct recover_control *rc)
 521 {
 522         struct chunk_record *chunk;
 523         struct block_group_record *bg;
 524         struct device_extent_record *devext;
 525         int total = 0;
 526         int good = 0;
 527         int bad = 0;
 528
 529         if (!rc->verbose)
 530                 return;
 531
 532         printf("CHECK RESULT:\n");
 533         printf("Recoverable Chunks:\n");
 534         list_for_each_entry(chunk, &rc->good_chunks, list) {
 535                 print_chunk_info(chunk, "  ");
 536                 good++;
 537                 total++;
 538         }
 539         list_for_each_entry(chunk, &rc->rebuild_chunks, list) {
 540                 print_chunk_info(chunk, "  ");
 541                 good++;
 542                 total++;
 543         }
 544         list_for_each_entry(chunk, &rc->unrepaired_chunks, list) {
 545                 print_chunk_info(chunk, "  ");
 546                 good++;
 547                 total++;
 548         }
 549         printf("Unrecoverable Chunks:\n");
 550         list_for_each_entry(chunk, &rc->bad_chunks, list) {
 551                 print_chunk_info(chunk, "  ");
 552                 bad++;
 553                 total++;
 554         }
 555         printf("\n");
 556         printf("Total Chunks:\t\t%d\n", total);
 557         printf("  Recoverable:\t\t%d\n", good);
 558         printf("  Unrecoverable:\t%d\n", bad);
 559
 560         printf("\n");
 561         printf("Orphan Block Groups:\n");
 562         list_for_each_entry(bg, &rc->bg.block_groups, list)
 563                 print_block_group_info(bg, "  ");
 564
 565         printf("\n");
 566         printf("Orphan Device Extents:\n");
 567         list_for_each_entry(devext, &rc->devext.no_chunk_orphans, chunk_list)
 568                 print_device_extent_info(devext, "  ");
 569         printf("\n");
 570 }
 571
 572 static int check_chunk_by_metadata(struct recover_control *rc,
 573                                    struct btrfs_root *root,
 574                                    struct chunk_record *chunk, int bg_only)
 575 {
 576         int ret;
 577         int i;
 578         int slot;
 579         struct btrfs_path path;
 580         struct btrfs_key key;
 581         struct btrfs_root *dev_root;
 582         struct stripe *stripe;
 583         struct btrfs_dev_extent *dev_extent;
 584         struct btrfs_block_group_item *bg_ptr;
 585         struct extent_buffer *l;
 586
 587         btrfs_init_path(&path);
 588
 589         if (bg_only)
 590                 goto bg_check;
 591
 592         dev_root = root->fs_info->dev_root;
 593         for (i = 0; i < chunk->num_stripes; i++) {
 594                 stripe = &chunk->stripes[i];
 595
 596                 key.objectid = stripe->devid;
 597                 key.offset = stripe->offset;
 598                 key.type = BTRFS_DEV_EXTENT_KEY;
 599
 600                 ret = btrfs_search_slot(NULL, dev_root, &key, &path, 0, 0);
 601                 if (ret < 0) {
 602                         fprintf(stderr, "Search device extent failed(%d)\n",
 603                                 ret);
 604                         btrfs_release_path(&path);
 605                         return ret;
 606                 } else if (ret > 0) {
 607                         if (rc->verbose)
 608                                 fprintf(stderr,
 609                                         "No device extent[%llu, %llu]\n",
 610                                         stripe->devid, stripe->offset);
 611                         btrfs_release_path(&path);
 612                         return -ENOENT;
 613                 }
 614                 l = path.nodes[0];
 615                 slot = path.slots[0];
 616                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 617                 if (chunk->offset !=
 618                     btrfs_dev_extent_chunk_offset(l, dev_extent)) {
 619                         if (rc->verbose)
 620                                 fprintf(stderr,
 621                                         "Device tree unmatch with chunks dev_extent[%llu, %llu], chunk[%llu, %llu]\n",
 622                                         btrfs_dev_extent_chunk_offset(l,
 623                                                                 dev_extent),
 624                                         btrfs_dev_extent_length(l, dev_extent),
 625                                         chunk->offset, chunk->length);
 626                         btrfs_release_path(&path);
 627                         return -ENOENT;
 628                 }
 629                 btrfs_release_path(&path);
 630         }
 631
 632 bg_check:
 633         key.objectid = chunk->offset;
 634         key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
 635         key.offset = chunk->length;
 636
 637         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, &path,
 638                                 0, 0);
 639         if (ret < 0) {
 640                 fprintf(stderr, "Search block group failed(%d)\n", ret);
 641                 btrfs_release_path(&path);
 642                 return ret;
 643         } else if (ret > 0) {
 644                 if (rc->verbose)
 645                         fprintf(stderr, "No block group[%llu, %llu]\n",
 646                                 key.objectid, key.offset);
 647                 btrfs_release_path(&path);
 648                 return -ENOENT;
 649         }
 650
 651         l = path.nodes[0];
 652         slot = path.slots[0];
 653         bg_ptr = btrfs_item_ptr(l, slot, struct btrfs_block_group_item);
 654         if (chunk->type_flags != btrfs_disk_block_group_flags(l, bg_ptr)) {
 655                 if (rc->verbose)
 656                         fprintf(stderr,
 657                                 "Chunk[%llu, %llu]'s type(%llu) is differemt with Block Group's type(%llu)\n",
 658                                 chunk->offset, chunk->length, chunk->type_flags,
 659                                 btrfs_disk_block_group_flags(l, bg_ptr));
 660                 btrfs_release_path(&path);
 661                 return -ENOENT;
 662         }
 663         btrfs_release_path(&path);
 664         return 0;
 665 }
 666
 667 static int check_all_chunks_by_metadata(struct recover_control *rc,
 668                                         struct btrfs_root *root)
 669 {
 670         struct chunk_record *chunk;
 671         struct chunk_record *next;
 672         LIST_HEAD(orphan_chunks);
 673         int ret = 0;
 674         int err;
 675
 676         list_for_each_entry_safe(chunk, next, &rc->good_chunks, list) {
 677                 err = check_chunk_by_metadata(rc, root, chunk, 0);
 678                 if (err) {
 679                         if (err == -ENOENT)
 680                                 list_move_tail(&chunk->list, &orphan_chunks);
 681                         else if (err && !ret)
 682                                 ret = err;
 683                 }
 684         }
 685
 686         list_for_each_entry_safe(chunk, next, &rc->unrepaired_chunks, list) {
 687                 err = check_chunk_by_metadata(rc, root, chunk, 1);
 688                 if (err == -ENOENT)
 689                         list_move_tail(&chunk->list, &orphan_chunks);
 690                 else if (err && !ret)
 691                         ret = err;
 692         }
 693
 694         list_for_each_entry(chunk, &rc->bad_chunks, list) {
 695                 err = check_chunk_by_metadata(rc, root, chunk, 1);
 696                 if (err != -ENOENT && !ret)
 697                         ret = err ? err : -EINVAL;
 698         }
 699         list_splice(&orphan_chunks, &rc->bad_chunks);
 700         return ret;
 701 }
 702
 703 static int extract_metadata_record(struct recover_control *rc,
 704                                    struct extent_buffer *leaf)
 705 {
 706         struct btrfs_key key;
 707         int ret = 0;
 708         int i;
 709         u32 nritems;
 710
 711         nritems = btrfs_header_nritems(leaf);
 712         for (i = 0; i < nritems; i++) {
 713                 btrfs_item_key_to_cpu(leaf, &key, i);
 714                 switch (key.type) {
 715                 case BTRFS_BLOCK_GROUP_ITEM_KEY:
 716                         pthread_mutex_lock(&rc->rc_lock);
 717                         ret = process_block_group_item(&rc->bg, leaf, &key, i);
 718                         pthread_mutex_unlock(&rc->rc_lock);
 719                         break;
 720                 case BTRFS_CHUNK_ITEM_KEY:
 721                         pthread_mutex_lock(&rc->rc_lock);
 722                         ret = process_chunk_item(&rc->chunk, leaf, &key, i);
 723                         pthread_mutex_unlock(&rc->rc_lock);
 724                         break;
 725                 case BTRFS_DEV_EXTENT_KEY:
 726                         pthread_mutex_lock(&rc->rc_lock);
 727                         ret = process_device_extent_item(&rc->devext, leaf,
 728                                                          &key, i);
 729                         pthread_mutex_unlock(&rc->rc_lock);
 730                         break;
 731                 }
 732                 if (ret)
 733                         break;
 734         }
 735         return ret;
 736 }
 737
 738 static inline int is_super_block_address(u64 offset)
 739 {
 740         int i;
 741
 742         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 743                 if (offset == btrfs_sb_offset(i))
 744                         return 1;
 745         }
 746         return 0;
 747 }
 748
 749 static int scan_one_device(void *dev_scan_struct)
 750 {
 751         struct extent_buffer *buf;
 752         u64 bytenr;
 753         int ret = 0;
 754         struct device_scan *dev_scan = (struct device_scan *)dev_scan_struct;
 755         struct recover_control *rc = dev_scan->rc;
 756         struct btrfs_device *device = dev_scan->dev;
 757         int fd = dev_scan->fd;
 758         int oldtype;
 759
 760         ret = pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS, &oldtype);
 761         if (ret)
 762                 return 1;
 763
 764         buf = malloc(sizeof(*buf) + rc->leafsize);
 765         if (!buf)
 766                 return -ENOMEM;
 767         buf->len = rc->leafsize;
 768
 769         bytenr = 0;
 770         while (1) {
 771                 if (is_super_block_address(bytenr))
 772                         bytenr += rc->sectorsize;
 773
 774                 if (pread64(fd, buf->data, rc->leafsize, bytenr) <
 775                     rc->leafsize)
 776                         break;
 777
 778                 if (memcmp_extent_buffer(buf, rc->fs_devices->fsid,
 779                                          btrfs_header_fsid(),
 780                                          BTRFS_FSID_SIZE)) {
 781                         bytenr += rc->sectorsize;
 782                         continue;
 783                 }
 784
 785                 if (verify_tree_block_csum_silent(buf, rc->csum_size)) {
 786                         bytenr += rc->sectorsize;
 787                         continue;
 788                 }
 789
 790                 pthread_mutex_lock(&rc->rc_lock);
 791                 ret = process_extent_buffer(&rc->eb_cache, buf, device, bytenr);
 792                 pthread_mutex_unlock(&rc->rc_lock);
 793                 if (ret)
 794                         goto out;
 795
 796                 if (btrfs_header_level(buf) != 0)
 797                         goto next_node;
 798
 799                 switch (btrfs_header_owner(buf)) {
 800                 case BTRFS_EXTENT_TREE_OBJECTID:
 801                 case BTRFS_DEV_TREE_OBJECTID:
 802                         /* different tree use different generation */
 803                         if (btrfs_header_generation(buf) > rc->generation)
 804                                 break;
 805                         ret = extract_metadata_record(rc, buf);
 806                         if (ret)
 807                                 goto out;
 808                         break;
 809                 case BTRFS_CHUNK_TREE_OBJECTID:
 810                         if (btrfs_header_generation(buf) >
 811                             rc->chunk_root_generation)
 812                                 break;
 813                         ret = extract_metadata_record(rc, buf);
 814                         if (ret)
 815                                 goto out;
 816                         break;
 817                 }
 818 next_node:
 819                 bytenr += rc->leafsize;
 820         }
 821 out:
 822         close(fd);
 823         free(buf);
 824         return ret;
 825 }
 826
 827 static int scan_devices(struct recover_control *rc)
 828 {
 829         int ret = 0;
 830         int fd;
 831         struct btrfs_device *dev;
 832         struct device_scan *dev_scans;
 833         pthread_t *t_scans;
 834         int *t_rets;
 835         int devnr = 0;
 836         int devidx = 0;
 837         int cancel_from = 0;
 838         int cancel_to = 0;
 839         int i;
 840
 841         list_for_each_entry(dev, &rc->fs_devices->devices, dev_list)
 842                 devnr++;
 843         dev_scans = (struct device_scan *)malloc(sizeof(struct device_scan)
 844                                                  * devnr);
 845         if (!dev_scans)
 846                 return -ENOMEM;
 847         t_scans = (pthread_t *)malloc(sizeof(pthread_t) * devnr);
 848         if (!t_scans)
 849                 return -ENOMEM;
 850         t_rets = (int *)malloc(sizeof(int) * devnr);
 851         if (!t_rets)
 852                 return -ENOMEM;
 853
 854         list_for_each_entry(dev, &rc->fs_devices->devices, dev_list) {
 855                 fd = open(dev->name, O_RDONLY);
 856                 if (fd < 0) {
 857                         fprintf(stderr, "Failed to open device %s\n",
 858                                 dev->name);
 859                         ret = 1;
 860                         goto out2;
 861                 }
 862                 dev_scans[devidx].rc = rc;
 863                 dev_scans[devidx].dev = dev;
 864                 dev_scans[devidx].fd = fd;
 865                 ret = pthread_create(&t_scans[devidx], NULL,
 866                                      (void *)scan_one_device,
 867                                      (void *)&dev_scans[devidx]);
 868                 if (ret) {
 869                         cancel_from = 0;
 870                         cancel_to = devidx - 1;
 871                         goto out1;
 872                 }
 873                 devidx++;
 874         }
 875
 876         i = 0;
 877         while (i < devidx) {
 878                 ret = pthread_join(t_scans[i], (void **)&t_rets[i]);
 879                 if (ret || t_rets[i]) {
 880                         ret = 1;
 881                         cancel_from = i + 1;
 882                         cancel_to = devnr - 1;
 883                         goto out1;
 884                 }
 885                 i++;
 886         }
 887 out1:
 888         while (ret && (cancel_from <= cancel_to)) {
 889                 pthread_cancel(t_scans[cancel_from]);
 890                 cancel_from++;
 891         }
 892 out2:
 893         free(dev_scans);
 894         free(t_scans);
 895         free(t_rets);
 896         return !!ret;
 897 }
 898
 899 static int build_device_map_by_chunk_record(struct btrfs_root *root,
 900                                             struct chunk_record *chunk)
 901 {
 902         int ret = 0;
 903         int i;
 904         u64 devid;
 905         u8 uuid[BTRFS_UUID_SIZE];
 906         u16 num_stripes;
 907         struct btrfs_mapping_tree *map_tree;
 908         struct map_lookup *map;
 909         struct stripe *stripe;
 910
 911         map_tree = &root->fs_info->mapping_tree;
 912         num_stripes = chunk->num_stripes;
 913         map = malloc(btrfs_map_lookup_size(num_stripes));
 914         if (!map)
 915                 return -ENOMEM;
 916         map->ce.start = chunk->offset;
 917         map->ce.size = chunk->length;
 918         map->num_stripes = num_stripes;
 919         map->io_width = chunk->io_width;
 920         map->io_align = chunk->io_align;
 921         map->sector_size = chunk->sector_size;
 922         map->stripe_len = chunk->stripe_len;
 923         map->type = chunk->type_flags;
 924         map->sub_stripes = chunk->sub_stripes;
 925
 926         for (i = 0, stripe = chunk->stripes; i < num_stripes; i++, stripe++) {
 927                 devid = stripe->devid;
 928                 memcpy(uuid, stripe->dev_uuid, BTRFS_UUID_SIZE);
 929                 map->stripes[i].physical = stripe->offset;
 930                 map->stripes[i].dev = btrfs_find_device(root, devid,
 931                                                         uuid, NULL);
 932                 if (!map->stripes[i].dev) {
 933                         kfree(map);
 934                         return -EIO;
 935                 }
 936         }
 937
 938         ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
 939         return ret;
 940 }
 941
 942 static int build_device_maps_by_chunk_records(struct recover_control *rc,
 943                                               struct btrfs_root *root)
 944 {
 945         int ret = 0;
 946         struct chunk_record *chunk;
 947
 948         list_for_each_entry(chunk, &rc->good_chunks, list) {
 949                 ret = build_device_map_by_chunk_record(root, chunk);
 950                 if (ret)
 951                         return ret;
 952         }
 953         list_for_each_entry(chunk, &rc->rebuild_chunks, list) {
 954                 ret = build_device_map_by_chunk_record(root, chunk);
 955                 if (ret)
 956                         return ret;
 957         }
 958         return ret;
 959 }
 960
 961 static int block_group_remove_all_extent_items(struct btrfs_trans_handle *trans,
 962                                                struct btrfs_root *root,
 963                                                struct block_group_record *bg)
 964 {
 965         struct btrfs_fs_info *fs_info = root->fs_info;
 966         struct btrfs_key key;
 967         struct btrfs_path path;
 968         struct extent_buffer *leaf;
 969         u64 start = bg->objectid;
 970         u64 end = bg->objectid + bg->offset;
 971         u64 old_val;
 972         int nitems;
 973         int ret;
 974         int i;
 975         int del_s, del_nr;
 976
 977         btrfs_init_path(&path);
 978         root = root->fs_info->extent_root;
 979
 980         key.objectid = start;
 981         key.offset = 0;
 982         key.type = BTRFS_EXTENT_ITEM_KEY;
 983 again:
 984         ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
 985         if (ret < 0)
 986                 goto err;
 987         else if (ret > 0)
 988                 ret = 0;
 989
 990         leaf = path.nodes[0];
 991         nitems = btrfs_header_nritems(leaf);
 992         if (!nitems) {
 993                 /* The tree is empty. */
 994                 ret = 0;
 995                 goto err;
 996         }
 997
 998         if (path.slots[0] >= nitems) {
 999                 ret = btrfs_next_leaf(root, &path);
1000                 if (ret < 0)
1001                         goto err;
1002                 if (ret > 0) {
1003                         ret = 0;
1004                         goto err;
1005                 }
1006                 leaf = path.nodes[0];
1007                 btrfs_item_key_to_cpu(leaf, &key, 0);
1008                 if (key.objectid >= end)
1009                         goto err;
1010                 btrfs_release_path(&path);
1011                 goto again;
1012         }
1013
1014         del_nr = 0;
1015         del_s = -1;
1016         for (i = path.slots[0]; i < nitems; i++) {
1017                 btrfs_item_key_to_cpu(leaf, &key, i);
1018                 if (key.objectid >= end)
1019                         break;
1020
1021                 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
1022                         if (del_nr == 0)
1023                                 continue;
1024                         else
1025                                 break;
1026                 }
1027
1028                 if (del_s == -1)
1029                         del_s = i;
1030                 del_nr++;
1031                 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
1032                     key.type == BTRFS_METADATA_ITEM_KEY) {
1033                         old_val = btrfs_super_bytes_used(fs_info->super_copy);
1034                         if (key.type == BTRFS_METADATA_ITEM_KEY)
1035                                 old_val += root->leafsize;
1036                         else
1037                                 old_val += key.offset;
1038                         btrfs_set_super_bytes_used(fs_info->super_copy,
1039                                                    old_val);
1040                 }
1041         }
1042
1043         if (del_nr) {
1044                 ret = btrfs_del_items(trans, root, &path, del_s, del_nr);
1045                 if (ret)
1046                         goto err;
1047         }
1048
1049         if (key.objectid < end) {
1050                 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
1051                         key.objectid += root->sectorsize;
1052                         key.type = BTRFS_EXTENT_ITEM_KEY;
1053                         key.offset = 0;
1054                 }
1055                 btrfs_release_path(&path);
1056                 goto again;
1057         }
1058 err:
1059         btrfs_release_path(&path);
1060         return ret;
1061 }
1062
1063 static int block_group_free_all_extent(struct btrfs_trans_handle *trans,
1064                                        struct btrfs_root *root,
1065                                        struct block_group_record *bg)
1066 {
1067         struct btrfs_block_group_cache *cache;
1068         struct btrfs_fs_info *info;
1069         u64 start;
1070         u64 end;
1071
1072         info = root->fs_info;
1073         cache = btrfs_lookup_block_group(info, bg->objectid);
1074         if (!cache)
1075                 return -ENOENT;
1076
1077         start = cache->key.objectid;
1078         end = start + cache->key.offset - 1;
1079
1080         set_extent_bits(&info->block_group_cache, start, end,
1081                         BLOCK_GROUP_DIRTY, GFP_NOFS);
1082         set_extent_dirty(&info->free_space_cache, start, end, GFP_NOFS);
1083
1084         btrfs_set_block_group_used(&cache->item, 0);
1085
1086         return 0;
1087 }
1088
1089 static int remove_chunk_extent_item(struct btrfs_trans_handle *trans,
1090                                     struct recover_control *rc,
1091                                     struct btrfs_root *root)
1092 {
1093         struct chunk_record *chunk;
1094         int ret = 0;
1095
1096         list_for_each_entry(chunk, &rc->good_chunks, list) {
1097                 if (!(chunk->type_flags & BTRFS_BLOCK_GROUP_SYSTEM))
1098                         continue;
1099                 ret = block_group_remove_all_extent_items(trans, root,
1100                                                           chunk->bg_rec);
1101                 if (ret)
1102                         return ret;
1103
1104                 ret = block_group_free_all_extent(trans, root, chunk->bg_rec);
1105                 if (ret)
1106                         return ret;
1107         }
1108         return ret;
1109 }
1110
1111 static int __rebuild_chunk_root(struct btrfs_trans_handle *trans,
1112                                 struct recover_control *rc,
1113                                 struct btrfs_root *root)
1114 {
1115         u64 min_devid = -1;
1116         struct btrfs_device *dev;
1117         struct extent_buffer *cow;
1118         struct btrfs_disk_key disk_key;
1119         int ret = 0;
1120
1121         list_for_each_entry(dev, &rc->fs_devices->devices, dev_list) {
1122                 if (min_devid > dev->devid)
1123                         min_devid = dev->devid;
1124         }
1125         disk_key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1126         disk_key.type = BTRFS_DEV_ITEM_KEY;
1127         disk_key.offset = min_devid;
1128
1129         cow = btrfs_alloc_free_block(trans, root, root->nodesize,
1130                                      BTRFS_CHUNK_TREE_OBJECTID,
1131                                      &disk_key, 0, 0, 0);
1132         btrfs_set_header_bytenr(cow, cow->start);
1133         btrfs_set_header_generation(cow, trans->transid);
1134         btrfs_set_header_nritems(cow, 0);
1135         btrfs_set_header_level(cow, 0);
1136         btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
1137         btrfs_set_header_owner(cow, BTRFS_CHUNK_TREE_OBJECTID);
1138         write_extent_buffer(cow, root->fs_info->fsid,
1139                         btrfs_header_fsid(), BTRFS_FSID_SIZE);
1140
1141         write_extent_buffer(cow, root->fs_info->chunk_tree_uuid,
1142                         btrfs_header_chunk_tree_uuid(cow),
1143                         BTRFS_UUID_SIZE);
1144
1145         root->node = cow;
1146         btrfs_mark_buffer_dirty(cow);
1147
1148         return ret;
1149 }
1150
1151 static int __rebuild_device_items(struct btrfs_trans_handle *trans,
1152                                   struct recover_control *rc,
1153                                   struct btrfs_root *root)
1154 {
1155         struct btrfs_device *dev;
1156         struct btrfs_key key;
1157         struct btrfs_dev_item *dev_item;
1158         int ret = 0;
1159
1160         dev_item = malloc(sizeof(struct btrfs_dev_item));
1161         if (!dev_item)
1162                 return -ENOMEM;
1163
1164         list_for_each_entry(dev, &rc->fs_devices->devices, dev_list) {
1165                 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1166                 key.type = BTRFS_DEV_ITEM_KEY;
1167                 key.offset = dev->devid;
1168
1169                 btrfs_set_stack_device_generation(dev_item, 0);
1170                 btrfs_set_stack_device_type(dev_item, dev->type);
1171                 btrfs_set_stack_device_id(dev_item, dev->devid);
1172                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1173                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1174                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1175                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1176                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1177                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1178                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
1179
1180                 ret = btrfs_insert_item(trans, root, &key,
1181                                         dev_item, sizeof(*dev_item));
1182         }
1183
1184         free(dev_item);
1185         return ret;
1186 }
1187
1188 static int __insert_chunk_item(struct btrfs_trans_handle *trans,
1189                                 struct chunk_record *chunk_rec,
1190                                 struct btrfs_root *chunk_root)
1191 {
1192         struct btrfs_key key;
1193         struct btrfs_chunk *chunk = NULL;
1194         int ret = 0;
1195
1196         chunk = create_chunk_item(chunk_rec);
1197         if (!chunk)
1198                 return -ENOMEM;
1199         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1200         key.type = BTRFS_CHUNK_ITEM_KEY;
1201         key.offset = chunk_rec->offset;
1202
1203         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1204                                 btrfs_chunk_item_size(chunk->num_stripes));
1205         free(chunk);
1206         return ret;
1207 }
1208
1209 static int __rebuild_chunk_items(struct btrfs_trans_handle *trans,
1210                                  struct recover_control *rc,
1211                                  struct btrfs_root *root)
1212 {
1213         struct btrfs_root *chunk_root;
1214         struct chunk_record *chunk_rec;
1215         int ret;
1216
1217         chunk_root = root->fs_info->chunk_root;
1218
1219         list_for_each_entry(chunk_rec, &rc->good_chunks, list) {
1220                 ret = __insert_chunk_item(trans, chunk_rec, chunk_root);
1221                 if (ret)
1222                         return ret;
1223         }
1224         list_for_each_entry(chunk_rec, &rc->rebuild_chunks, list) {
1225                 ret = __insert_chunk_item(trans, chunk_rec, chunk_root);
1226                 if (ret)
1227                         return ret;
1228         }
1229         return 0;
1230 }
1231
1232 static int rebuild_chunk_tree(struct btrfs_trans_handle *trans,
1233                               struct recover_control *rc,
1234                               struct btrfs_root *root)
1235 {
1236         int ret = 0;
1237
1238         root = root->fs_info->chunk_root;
1239
1240         ret = __rebuild_chunk_root(trans, rc, root);
1241         if (ret)
1242                 return ret;
1243
1244         ret = __rebuild_device_items(trans, rc, root);
1245         if (ret)
1246                 return ret;
1247
1248         ret = __rebuild_chunk_items(trans, rc, root);
1249
1250         return ret;
1251 }
1252
1253 static int rebuild_sys_array(struct recover_control *rc,
1254                              struct btrfs_root *root)
1255 {
1256         struct btrfs_chunk *chunk;
1257         struct btrfs_key key;
1258         struct chunk_record *chunk_rec;
1259         int ret = 0;
1260         u16 num_stripes;
1261
1262         btrfs_set_super_sys_array_size(root->fs_info->super_copy, 0);
1263
1264         list_for_each_entry(chunk_rec, &rc->good_chunks, list) {
1265                 if (!(chunk_rec->type_flags & BTRFS_BLOCK_GROUP_SYSTEM))
1266                         continue;
1267
1268                 num_stripes = chunk_rec->num_stripes;
1269                 chunk = create_chunk_item(chunk_rec);
1270                 if (!chunk) {
1271                         ret = -ENOMEM;
1272                         break;
1273                 }
1274
1275                 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1276                 key.type = BTRFS_CHUNK_ITEM_KEY;
1277                 key.offset = chunk_rec->offset;
1278
1279                 ret = btrfs_add_system_chunk(NULL, root, &key, chunk,
1280                                 btrfs_chunk_item_size(num_stripes));
1281                 free(chunk);
1282                 if (ret)
1283                         break;
1284         }
1285         return ret;
1286
1287 }
1288
1289 static int calculate_bg_used(struct btrfs_root *extent_root,
1290                              struct chunk_record *chunk_rec,
1291                              struct btrfs_path *path,
1292                              u64 *used)
1293 {
1294         struct extent_buffer *node;
1295         struct btrfs_key found_key;
1296         int slot;
1297         int ret = 0;
1298         u64 used_ret = 0;
1299
1300         while (1) {
1301                 node = path->nodes[0];
1302                 slot = path->slots[0];
1303                 btrfs_item_key_to_cpu(node, &found_key, slot);
1304                 if (found_key.objectid >= chunk_rec->offset + chunk_rec->length)
1305                         break;
1306                 if (found_key.type != BTRFS_METADATA_ITEM_KEY &&
1307                     found_key.type != BTRFS_EXTENT_DATA_KEY)
1308                         goto next;
1309                 if (found_key.type == BTRFS_METADATA_ITEM_KEY)
1310                         used_ret += extent_root->nodesize;
1311                 else
1312                         used_ret += found_key.offset;
1313 next:
1314                 if (slot + 1 < btrfs_header_nritems(node)) {
1315                         slot++;
1316                 } else {
1317                         ret = btrfs_next_leaf(extent_root, path);
1318                         if (ret > 0) {
1319                                 ret = 0;
1320                                 break;
1321                         }
1322                         if (ret < 0)
1323                                 break;
1324                 }
1325         }
1326         if (!ret)
1327                 *used = used_ret;
1328         return ret;
1329 }
1330
1331 static int __insert_block_group(struct btrfs_trans_handle *trans,
1332                                 struct chunk_record *chunk_rec,
1333                                 struct btrfs_root *extent_root,
1334                                 u64 used)
1335 {
1336         struct btrfs_block_group_item bg_item;
1337         struct btrfs_key key;
1338         int ret = 0;
1339
1340         btrfs_set_block_group_used(&bg_item, used);
1341         btrfs_set_block_group_chunk_objectid(&bg_item, used);
1342         btrfs_set_block_group_flags(&bg_item, chunk_rec->type_flags);
1343         key.objectid = chunk_rec->offset;
1344         key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
1345         key.offset = chunk_rec->length;
1346
1347         ret = btrfs_insert_item(trans, extent_root, &key, &bg_item,
1348                                 sizeof(bg_item));
1349         return ret;
1350 }
1351
1352 /*
1353  * Search through the extent tree to rebuild the 'used' member of the block
1354  * group.
1355  * However, since block group and extent item shares the extent tree,
1356  * the extent item may also missing.
1357  * In that case, we fill the 'used' with the length of the block group to
1358  * ensure no write into the block group.
1359  * Btrfsck will hate it but we will inform user to call '--init-extent-tree'
1360  * if possible, or just salvage as much data as possible from the fs.
1361  */
1362 static int rebuild_block_group(struct btrfs_trans_handle *trans,
1363                                struct recover_control *rc,
1364                                struct btrfs_root *root)
1365 {
1366         struct chunk_record *chunk_rec;
1367         struct btrfs_key search_key;
1368         struct btrfs_path *path;
1369         u64 used = 0;
1370         int ret = 0;
1371
1372         if (list_empty(&rc->rebuild_chunks))
1373                 return 0;
1374
1375         path = btrfs_alloc_path();
1376         if (!path)
1377                 return -ENOMEM;
1378         list_for_each_entry(chunk_rec, &rc->rebuild_chunks, list) {
1379                 search_key.objectid = chunk_rec->offset;
1380                 search_key.type = BTRFS_EXTENT_ITEM_KEY;
1381                 search_key.offset = 0;
1382                 ret = btrfs_search_slot(NULL, root->fs_info->extent_root,
1383                                         &search_key, path, 0, 0);
1384                 if (ret < 0)
1385                         goto out;
1386                 ret = calculate_bg_used(root->fs_info->extent_root,
1387                                         chunk_rec, path, &used);
1388                 /*
1389                  * Extent tree is damaged, better to rebuild the whole extent
1390                  * tree. Currently, change the used to chunk's len to prevent
1391                  * write/block reserve happening in that block group.
1392                  */
1393                 if (ret < 0) {
1394                         fprintf(stderr,
1395                                 "Fail to search extent tree for block group: [%llu,%llu]\n",
1396                                 chunk_rec->offset,
1397                                 chunk_rec->offset + chunk_rec->length);
1398                         fprintf(stderr,
1399                                 "Mark the block group full to prevent block rsv problems\n");
1400                         used = chunk_rec->length;
1401                 }
1402                 btrfs_release_path(path);
1403                 ret = __insert_block_group(trans, chunk_rec,
1404                                            root->fs_info->extent_root,
1405                                            used);
1406                 if (ret < 0)
1407                         goto out;
1408         }
1409 out:
1410         btrfs_free_path(path);
1411         return ret;
1412 }
1413
1414 static struct btrfs_root *
1415 open_ctree_with_broken_chunk(struct recover_control *rc)
1416 {
1417         struct btrfs_fs_info *fs_info;
1418         struct btrfs_super_block *disk_super;
1419         struct extent_buffer *eb;
1420         u32 sectorsize;
1421         u32 nodesize;
1422         u32 leafsize;
1423         u32 stripesize;
1424         int ret;
1425
1426         fs_info = btrfs_new_fs_info(1, BTRFS_SUPER_INFO_OFFSET);
1427         if (!fs_info) {
1428                 fprintf(stderr, "Failed to allocate memory for fs_info\n");
1429                 return ERR_PTR(-ENOMEM);
1430         }
1431         fs_info->is_chunk_recover = 1;
1432
1433         fs_info->fs_devices = rc->fs_devices;
1434         ret = btrfs_open_devices(fs_info->fs_devices, O_RDWR);
1435         if (ret)
1436                 goto out;
1437
1438         disk_super = fs_info->super_copy;
1439         ret = btrfs_read_dev_super(fs_info->fs_devices->latest_bdev,
1440                                    disk_super, fs_info->super_bytenr, 1);
1441         if (ret) {
1442                 fprintf(stderr, "No valid btrfs found\n");
1443                 goto out_devices;
1444         }
1445
1446         memcpy(fs_info->fsid, &disk_super->fsid, BTRFS_FSID_SIZE);
1447
1448         ret = btrfs_check_fs_compatibility(disk_super, 1);
1449         if (ret)
1450                 goto out_devices;
1451
1452         nodesize = btrfs_super_nodesize(disk_super);
1453         leafsize = btrfs_super_leafsize(disk_super);
1454         sectorsize = btrfs_super_sectorsize(disk_super);
1455         stripesize = btrfs_super_stripesize(disk_super);
1456
1457         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1458                      fs_info->chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1459
1460         ret = build_device_maps_by_chunk_records(rc, fs_info->chunk_root);
1461         if (ret)
1462                 goto out_cleanup;
1463
1464         ret = btrfs_setup_all_roots(fs_info, 0, 0);
1465         if (ret)
1466                 goto out_failed;
1467
1468         eb = fs_info->tree_root->node;
1469         read_extent_buffer(eb, fs_info->chunk_tree_uuid,
1470                            btrfs_header_chunk_tree_uuid(eb),
1471                            BTRFS_UUID_SIZE);
1472
1473         return fs_info->fs_root;
1474 out_failed:
1475         btrfs_release_all_roots(fs_info);
1476 out_cleanup:
1477         btrfs_cleanup_all_caches(fs_info);
1478 out_devices:
1479         btrfs_close_devices(fs_info->fs_devices);
1480 out:
1481         btrfs_free_fs_info(fs_info);
1482         return ERR_PTR(ret);
1483 }
1484
1485 static int recover_prepare(struct recover_control *rc, char *path)
1486 {
1487         int ret;
1488         int fd;
1489         struct btrfs_super_block *sb;
1490         struct btrfs_fs_devices *fs_devices;
1491
1492         ret = 0;
1493         fd = open(path, O_RDONLY);
1494         if (fd < 0) {
1495                 fprintf(stderr, "open %s\n error.\n", path);
1496                 return -1;
1497         }
1498
1499         sb = malloc(BTRFS_SUPER_INFO_SIZE);
1500         if (!sb) {
1501                 fprintf(stderr, "allocating memory for sb failed.\n");
1502                 ret = -ENOMEM;
1503                 goto fail_close_fd;
1504         }
1505
1506         ret = btrfs_read_dev_super(fd, sb, BTRFS_SUPER_INFO_OFFSET, 1);
1507         if (ret) {
1508                 fprintf(stderr, "read super block error\n");
1509                 goto fail_free_sb;
1510         }
1511
1512         rc->sectorsize = btrfs_super_sectorsize(sb);
1513         rc->leafsize = btrfs_super_leafsize(sb);
1514         rc->generation = btrfs_super_generation(sb);
1515         rc->chunk_root_generation = btrfs_super_chunk_root_generation(sb);
1516         rc->csum_size = btrfs_super_csum_size(sb);
1517
1518         /* if seed, the result of scanning below will be partial */
1519         if (btrfs_super_flags(sb) & BTRFS_SUPER_FLAG_SEEDING) {
1520                 fprintf(stderr, "this device is seed device\n");
1521                 ret = -1;
1522                 goto fail_free_sb;
1523         }
1524
1525         ret = btrfs_scan_fs_devices(fd, path, &fs_devices, 0, 1, 0);
1526         if (ret)
1527                 goto fail_free_sb;
1528
1529         rc->fs_devices = fs_devices;
1530
1531         if (rc->verbose)
1532                 print_all_devices(&rc->fs_devices->devices);
1533
1534 fail_free_sb:
1535         free(sb);
1536 fail_close_fd:
1537         close(fd);
1538         return ret;
1539 }
1540
1541 static int btrfs_get_device_extents(u64 chunk_object,
1542                                     struct list_head *orphan_devexts,
1543                                     struct list_head *ret_list)
1544 {
1545         struct device_extent_record *devext;
1546         struct device_extent_record *next;
1547         int count = 0;
1548
1549         list_for_each_entry_safe(devext, next, orphan_devexts, chunk_list) {
1550                 if (devext->chunk_offset == chunk_object) {
1551                         list_move_tail(&devext->chunk_list, ret_list);
1552                         count++;
1553                 }
1554         }
1555         return count;
1556 }
1557
1558 static int calc_num_stripes(u64 type)
1559 {
1560         if (type & (BTRFS_BLOCK_GROUP_RAID0 |
1561                     BTRFS_BLOCK_GROUP_RAID10 |
1562                     BTRFS_BLOCK_GROUP_RAID5 |
1563                     BTRFS_BLOCK_GROUP_RAID6))
1564                 return 0;
1565         else if (type & (BTRFS_BLOCK_GROUP_RAID1 |
1566                          BTRFS_BLOCK_GROUP_DUP))
1567                 return 2;
1568         else
1569                 return 1;
1570 }
1571
1572 static inline int calc_sub_nstripes(u64 type)
1573 {
1574         if (type & BTRFS_BLOCK_GROUP_RAID10)
1575                 return 2;
1576         else
1577                 return 1;
1578 }
1579
1580 static int btrfs_verify_device_extents(struct block_group_record *bg,
1581                                        struct list_head *devexts, int ndevexts)
1582 {
1583         struct device_extent_record *devext;
1584         u64 strpie_length;
1585         int expected_num_stripes;
1586
1587         expected_num_stripes = calc_num_stripes(bg->flags);
1588         if (expected_num_stripes && expected_num_stripes != ndevexts)
1589                 return 1;
1590
1591         strpie_length = calc_stripe_length(bg->flags, bg->offset, ndevexts);
1592         list_for_each_entry(devext, devexts, chunk_list) {
1593                 if (devext->length != strpie_length)
1594                         return 1;
1595         }
1596         return 0;
1597 }
1598
1599 static int btrfs_rebuild_unordered_chunk_stripes(struct recover_control *rc,
1600                                                  struct chunk_record *chunk)
1601 {
1602         struct device_extent_record *devext;
1603         struct btrfs_device *device;
1604         int i;
1605
1606         devext = list_first_entry(&chunk->dextents, struct device_extent_record,
1607                                   chunk_list);
1608         for (i = 0; i < chunk->num_stripes; i++) {
1609                 chunk->stripes[i].devid = devext->objectid;
1610                 chunk->stripes[i].offset = devext->offset;
1611                 device = btrfs_find_device_by_devid(rc->fs_devices,
1612                                                     devext->objectid,
1613                                                     0);
1614                 if (!device)
1615                         return -ENOENT;
1616                 BUG_ON(btrfs_find_device_by_devid(rc->fs_devices,
1617                                                   devext->objectid,
1618                                                   1));
1619                 memcpy(chunk->stripes[i].dev_uuid, device->uuid,
1620                        BTRFS_UUID_SIZE);
1621                 devext = list_next_entry(devext, chunk_list);
1622         }
1623         return 0;
1624 }
1625
1626 static int btrfs_calc_stripe_index(struct chunk_record *chunk, u64 logical)
1627 {
1628         u64 offset = logical - chunk->offset;
1629         int stripe_nr;
1630         int nr_data_stripes;
1631         int index;
1632
1633         stripe_nr = offset / chunk->stripe_len;
1634         if (chunk->type_flags & BTRFS_BLOCK_GROUP_RAID0) {
1635                 index = stripe_nr % chunk->num_stripes;
1636         } else if (chunk->type_flags & BTRFS_BLOCK_GROUP_RAID10) {
1637                 index = stripe_nr % (chunk->num_stripes / chunk->sub_stripes);
1638                 index *= chunk->sub_stripes;
1639         } else if (chunk->type_flags & BTRFS_BLOCK_GROUP_RAID5) {
1640                 nr_data_stripes = chunk->num_stripes - 1;
1641                 index = stripe_nr % nr_data_stripes;
1642                 stripe_nr /= nr_data_stripes;
1643                 index = (index + stripe_nr) % chunk->num_stripes;
1644         } else if (chunk->type_flags & BTRFS_BLOCK_GROUP_RAID6) {
1645                 nr_data_stripes = chunk->num_stripes - 2;
1646                 index = stripe_nr % nr_data_stripes;
1647                 stripe_nr /= nr_data_stripes;
1648                 index = (index + stripe_nr) % chunk->num_stripes;
1649         } else {
1650                 return -1;
1651         }
1652         return index;
1653 }
1654
1655 /* calc the logical offset which is the start of the next stripe. */
1656 static inline u64 btrfs_next_stripe_logical_offset(struct chunk_record *chunk,
1657                                                    u64 logical)
1658 {
1659         u64 offset = logical - chunk->offset;
1660
1661         offset /= chunk->stripe_len;
1662         offset *= chunk->stripe_len;
1663         offset += chunk->stripe_len;
1664
1665         return offset + chunk->offset;
1666 }
1667
1668 static int is_extent_record_in_device_extent(struct extent_record *er,
1669                                              struct device_extent_record *dext,
1670                                              int *mirror)
1671 {
1672         int i;
1673
1674         for (i = 0; i < er->nmirrors; i++) {
1675                 if (er->devices[i]->devid == dext->objectid &&
1676                     er->offsets[i] >= dext->offset &&
1677                     er->offsets[i] < dext->offset + dext->length) {
1678                         *mirror = i;
1679                         return 1;
1680                 }
1681         }
1682         return 0;
1683 }
1684
1685 static int
1686 btrfs_rebuild_ordered_meta_chunk_stripes(struct recover_control *rc,
1687                                          struct chunk_record *chunk)
1688 {
1689         u64 start = chunk->offset;
1690         u64 end = chunk->offset + chunk->length;
1691         struct cache_extent *cache;
1692         struct extent_record *er;
1693         struct device_extent_record *devext;
1694         struct device_extent_record *next;
1695         struct btrfs_device *device;
1696         LIST_HEAD(devexts);
1697         int index;
1698         int mirror;
1699         int ret;
1700
1701         cache = lookup_cache_extent(&rc->eb_cache,
1702                                     start, chunk->length);
1703         if (!cache) {
1704                 /* No used space, we can reorder the stripes freely. */
1705                 ret = btrfs_rebuild_unordered_chunk_stripes(rc, chunk);
1706                 return ret;
1707         }
1708
1709         list_splice_init(&chunk->dextents, &devexts);
1710 again:
1711         er = container_of(cache, struct extent_record, cache);
1712         index = btrfs_calc_stripe_index(chunk, er->cache.start);
1713         BUG_ON(index == -1);
1714         if (chunk->stripes[index].devid)
1715                 goto next;
1716         list_for_each_entry_safe(devext, next, &devexts, chunk_list) {
1717                 if (is_extent_record_in_device_extent(er, devext, &mirror)) {
1718                         chunk->stripes[index].devid = devext->objectid;
1719                         chunk->stripes[index].offset = devext->offset;
1720                         memcpy(chunk->stripes[index].dev_uuid,
1721                                er->devices[mirror]->uuid,
1722                                BTRFS_UUID_SIZE);
1723                         index++;
1724                         list_move(&devext->chunk_list, &chunk->dextents);
1725                 }
1726         }
1727 next:
1728         start = btrfs_next_stripe_logical_offset(chunk, er->cache.start);
1729         if (start >= end)
1730                 goto no_extent_record;
1731
1732         cache = lookup_cache_extent(&rc->eb_cache, start, end - start);
1733         if (cache)
1734                 goto again;
1735 no_extent_record:
1736         if (list_empty(&devexts))
1737                 return 0;
1738
1739         if (chunk->type_flags & (BTRFS_BLOCK_GROUP_RAID5 |
1740                                  BTRFS_BLOCK_GROUP_RAID6)) {
1741                 /* Fixme: try to recover the order by the parity block. */
1742                 list_splice_tail(&devexts, &chunk->dextents);
1743                 return -EINVAL;
1744         }
1745
1746         /* There is no data on the lost stripes, we can reorder them freely. */
1747         for (index = 0; index < chunk->num_stripes; index++) {
1748                 if (chunk->stripes[index].devid)
1749                         continue;
1750
1751                 devext = list_first_entry(&devexts,
1752                                           struct device_extent_record,
1753                                            chunk_list);
1754                 list_move(&devext->chunk_list, &chunk->dextents);
1755
1756                 chunk->stripes[index].devid = devext->objectid;
1757                 chunk->stripes[index].offset = devext->offset;
1758                 device = btrfs_find_device_by_devid(rc->fs_devices,
1759                                                     devext->objectid,
1760                                                     0);
1761                 if (!device) {
1762                         list_splice_tail(&devexts, &chunk->dextents);
1763                         return -EINVAL;
1764                 }
1765                 BUG_ON(btrfs_find_device_by_devid(rc->fs_devices,
1766                                                   devext->objectid,
1767                                                   1));
1768                 memcpy(chunk->stripes[index].dev_uuid, device->uuid,
1769                        BTRFS_UUID_SIZE);
1770         }
1771         return 0;
1772 }
1773
1774 #define BTRFS_ORDERED_RAID      (BTRFS_BLOCK_GROUP_RAID0 |      \
1775                                  BTRFS_BLOCK_GROUP_RAID10 |     \
1776                                  BTRFS_BLOCK_GROUP_RAID5 |      \
1777                                  BTRFS_BLOCK_GROUP_RAID6)
1778
1779 static int btrfs_rebuild_chunk_stripes(struct recover_control *rc,
1780                                        struct chunk_record *chunk)
1781 {
1782         int ret;
1783
1784         /*
1785          * All the data in the system metadata chunk will be dropped,
1786          * so we need not guarantee that the data is right or not, that
1787          * is we can reorder the stripes in the system metadata chunk.
1788          */
1789         if ((chunk->type_flags & BTRFS_BLOCK_GROUP_METADATA) &&
1790             (chunk->type_flags & BTRFS_ORDERED_RAID))
1791                 ret =btrfs_rebuild_ordered_meta_chunk_stripes(rc, chunk);
1792         else if ((chunk->type_flags & BTRFS_BLOCK_GROUP_DATA) &&
1793                  (chunk->type_flags & BTRFS_ORDERED_RAID))
1794                 ret = 1;        /* Be handled after the fs is opened. */
1795         else
1796                 ret = btrfs_rebuild_unordered_chunk_stripes(rc, chunk);
1797
1798         return ret;
1799 }
1800
1801 static int next_csum(struct btrfs_root *root,
1802                      struct extent_buffer **leaf,
1803                      struct btrfs_path *path,
1804                      int *slot,
1805                      u64 *csum_offset,
1806                      u32 *tree_csum,
1807                      u64 end,
1808                      struct btrfs_key *key)
1809 {
1810         int ret = 0;
1811         struct btrfs_root *csum_root = root->fs_info->csum_root;
1812         struct btrfs_csum_item *csum_item;
1813         u32 blocksize = root->sectorsize;
1814         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
1815         int csums_in_item = btrfs_item_size_nr(*leaf, *slot) / csum_size;
1816
1817         if (*csum_offset >= csums_in_item) {
1818                 ++(*slot);
1819                 *csum_offset = 0;
1820                 if (*slot >= btrfs_header_nritems(*leaf)) {
1821                         ret = btrfs_next_leaf(csum_root, path);
1822                         if (ret < 0)
1823                                 return -1;
1824                         else if (ret > 0)
1825                                 return 1;
1826                         *leaf = path->nodes[0];
1827                         *slot = path->slots[0];
1828                 }
1829                 btrfs_item_key_to_cpu(*leaf, key, *slot);
1830         }
1831
1832         if (key->offset + (*csum_offset) * blocksize >= end)
1833                 return 2;
1834         csum_item = btrfs_item_ptr(*leaf, *slot, struct btrfs_csum_item);
1835         csum_item = (struct btrfs_csum_item *)((unsigned char *)csum_item
1836                                              + (*csum_offset) * csum_size);
1837         read_extent_buffer(*leaf, tree_csum,
1838                           (unsigned long)csum_item, csum_size);
1839         return ret;
1840 }
1841
1842 static u64 calc_data_offset(struct btrfs_key *key,
1843                             struct chunk_record *chunk,
1844                             u64 dev_offset,
1845                             u64 csum_offset,
1846                             u32 blocksize)
1847 {
1848         u64 data_offset;
1849         int logical_stripe_nr;
1850         int dev_stripe_nr;
1851         int nr_data_stripes;
1852
1853         data_offset = key->offset + csum_offset * blocksize - chunk->offset;
1854         nr_data_stripes = chunk->num_stripes;
1855
1856         if (chunk->type_flags & BTRFS_BLOCK_GROUP_RAID5)
1857                 nr_data_stripes -= 1;
1858         else if (chunk->type_flags & BTRFS_BLOCK_GROUP_RAID6)
1859                 nr_data_stripes -= 2;
1860
1861         logical_stripe_nr = data_offset / chunk->stripe_len;
1862         dev_stripe_nr = logical_stripe_nr / nr_data_stripes;
1863
1864         data_offset -= logical_stripe_nr * chunk->stripe_len;
1865         data_offset += dev_stripe_nr * chunk->stripe_len;
1866
1867         return dev_offset + data_offset;
1868 }
1869
1870 static int check_one_csum(int fd, u64 start, u32 len, u32 tree_csum)
1871 {
1872         char *data;
1873         int ret = 0;
1874         u32 csum_result = ~(u32)0;
1875
1876         data = malloc(len);
1877         if (!data)
1878                 return -1;
1879         ret = pread64(fd, data, len, start);
1880         if (ret < 0 || ret != len) {
1881                 ret = -1;
1882                 goto out;
1883         }
1884         ret = 0;
1885         csum_result = btrfs_csum_data(NULL, data, csum_result, len);
1886         btrfs_csum_final(csum_result, (char *)&csum_result);
1887         if (csum_result != tree_csum)
1888                 ret = 1;
1889 out:
1890         free(data);
1891         return ret;
1892 }
1893
1894 static u64 item_end_offset(struct btrfs_root *root, struct btrfs_key *key,
1895                            struct extent_buffer *leaf, int slot) {
1896         u32 blocksize = root->sectorsize;
1897         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
1898
1899         u64 offset = btrfs_item_size_nr(leaf, slot);
1900         offset /= csum_size;
1901         offset *= blocksize;
1902         offset += key->offset;
1903
1904         return offset;
1905 }
1906
1907 static int insert_stripe(struct list_head *devexts,
1908                          struct recover_control *rc,
1909                          struct chunk_record *chunk,
1910                          int index) {
1911         struct device_extent_record *devext;
1912         struct btrfs_device *dev;
1913
1914         devext = list_entry(devexts->next, struct device_extent_record,
1915                             chunk_list);
1916         dev = btrfs_find_device_by_devid(rc->fs_devices, devext->objectid,
1917                                         0);
1918         if (!dev)
1919                 return 1;
1920         BUG_ON(btrfs_find_device_by_devid(rc->fs_devices, devext->objectid,
1921                                         1));
1922
1923         chunk->stripes[index].devid = devext->objectid;
1924         chunk->stripes[index].offset = devext->offset;
1925         memcpy(chunk->stripes[index].dev_uuid, dev->uuid, BTRFS_UUID_SIZE);
1926
1927         list_move(&devext->chunk_list, &chunk->dextents);
1928
1929         return 0;
1930 }
1931
1932 static inline int count_devext_records(struct list_head *record_list)
1933 {
1934         int num_of_records = 0;
1935         struct device_extent_record *devext;
1936
1937         list_for_each_entry(devext, record_list, chunk_list)
1938                 num_of_records++;
1939
1940         return num_of_records;
1941 }
1942
1943 static int fill_chunk_up(struct chunk_record *chunk, struct list_head *devexts,
1944                          struct recover_control *rc)
1945 {
1946         int ret = 0;
1947         int i;
1948
1949         for (i = 0; i < chunk->num_stripes; i++) {
1950                 if (!chunk->stripes[i].devid) {
1951                         ret = insert_stripe(devexts, rc, chunk, i);
1952                         if (ret)
1953                                 break;
1954                 }
1955         }
1956
1957         return ret;
1958 }
1959
1960 #define EQUAL_STRIPE (1 << 0)
1961
1962 static int rebuild_raid_data_chunk_stripes(struct recover_control *rc,
1963                                            struct btrfs_root *root,
1964                                            struct chunk_record *chunk,
1965                                            u8 *flags)
1966 {
1967         int i;
1968         int ret = 0;
1969         int slot;
1970         struct btrfs_path path;
1971         struct btrfs_key prev_key;
1972         struct btrfs_key key;
1973         struct btrfs_root *csum_root;
1974         struct extent_buffer *leaf;
1975         struct device_extent_record *devext;
1976         struct device_extent_record *next;
1977         struct btrfs_device *dev;
1978         u64 start = chunk->offset;
1979         u64 end = start + chunk->stripe_len;
1980         u64 chunk_end = chunk->offset + chunk->length;
1981         u64 csum_offset = 0;
1982         u64 data_offset;
1983         u32 blocksize = root->sectorsize;
1984         u32 tree_csum;
1985         int index = 0;
1986         int num_unordered = 0;
1987         LIST_HEAD(unordered);
1988         LIST_HEAD(candidates);
1989
1990         csum_root = root->fs_info->csum_root;
1991         btrfs_init_path(&path);
1992         list_splice_init(&chunk->dextents, &candidates);
1993 again:
1994         if (list_is_last(candidates.next, &candidates))
1995                 goto out;
1996
1997         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1998         key.type = BTRFS_EXTENT_CSUM_KEY;
1999         key.offset = start;
2000
2001         ret = btrfs_search_slot(NULL, csum_root, &key, &path, 0, 0);
2002         if (ret < 0) {
2003                 fprintf(stderr, "Search csum failed(%d)\n", ret);
2004                 goto fail_out;
2005         }
2006         leaf = path.nodes[0];
2007         slot = path.slots[0];
2008         if (ret > 0) {
2009                 if (slot >= btrfs_header_nritems(leaf)) {
2010                         ret = btrfs_next_leaf(csum_root, &path);
2011                         if (ret < 0) {
2012                                 fprintf(stderr,
2013                                         "Walk tree failed(%d)\n", ret);
2014                                 goto fail_out;
2015                         } else if (ret > 0) {
2016                                 slot = btrfs_header_nritems(leaf) - 1;
2017                                 btrfs_item_key_to_cpu(leaf, &key, slot);
2018                                 if (item_end_offset(root, &key, leaf, slot)
2019                                                                 > start) {
2020                                         csum_offset = start - key.offset;
2021                                         csum_offset /= blocksize;
2022                                         goto next_csum;
2023                                 }
2024                                 goto next_stripe;
2025                         }
2026                         leaf = path.nodes[0];
2027                         slot = path.slots[0];
2028                 }
2029                 btrfs_item_key_to_cpu(leaf, &key, slot);
2030                 ret = btrfs_previous_item(csum_root, &path, 0,
2031                                           BTRFS_EXTENT_CSUM_KEY);
2032                 if (ret < 0)
2033                         goto fail_out;
2034                 else if (ret > 0) {
2035                         if (key.offset >= end)
2036                                 goto next_stripe;
2037                         else
2038                                 goto next_csum;
2039                 }
2040                 leaf = path.nodes[0];
2041                 slot = path.slots[0];
2042
2043                 btrfs_item_key_to_cpu(leaf, &prev_key, slot);
2044                 if (item_end_offset(root, &prev_key, leaf, slot) > start) {
2045                         csum_offset = start - prev_key.offset;
2046                         csum_offset /= blocksize;
2047                         btrfs_item_key_to_cpu(leaf, &key, slot);
2048                 } else {
2049                         if (key.offset >= end)
2050                                 goto next_stripe;
2051                 }
2052
2053                 if (key.offset + csum_offset * blocksize > chunk_end)
2054                         goto out;
2055         }
2056 next_csum:
2057         ret = next_csum(root, &leaf, &path, &slot, &csum_offset, &tree_csum,
2058                         end, &key);
2059         if (ret < 0) {
2060                 fprintf(stderr, "Fetch csum failed\n");
2061                 goto fail_out;
2062         } else if (ret == 1) {
2063                 if (!(*flags & EQUAL_STRIPE))
2064                         *flags |= EQUAL_STRIPE;
2065                 goto out;
2066         } else if (ret == 2)
2067                 goto next_stripe;
2068
2069         list_for_each_entry_safe(devext, next, &candidates, chunk_list) {
2070                 data_offset = calc_data_offset(&key, chunk, devext->offset,
2071                                                csum_offset, blocksize);
2072                 dev = btrfs_find_device_by_devid(rc->fs_devices,
2073                                                  devext->objectid, 0);
2074                 if (!dev) {
2075                         ret = 1;
2076                         goto fail_out;
2077                 }
2078                 BUG_ON(btrfs_find_device_by_devid(rc->fs_devices,
2079                                                   devext->objectid, 1));
2080
2081                 ret = check_one_csum(dev->fd, data_offset, blocksize,
2082                                      tree_csum);
2083                 if (ret < 0)
2084                         goto fail_out;
2085                 else if (ret > 0)
2086                         list_move(&devext->chunk_list, &unordered);
2087         }
2088
2089         if (list_empty(&candidates)) {
2090                 num_unordered = count_devext_records(&unordered);
2091                 if (chunk->type_flags & BTRFS_BLOCK_GROUP_RAID6
2092                                         && num_unordered == 2) {
2093                         btrfs_release_path(&path);
2094                         ret = fill_chunk_up(chunk, &unordered, rc);
2095                         return ret;
2096                 }
2097
2098                 goto next_stripe;
2099         }
2100
2101         if (list_is_last(candidates.next, &candidates)) {
2102                 index = btrfs_calc_stripe_index(chunk,
2103                         key.offset + csum_offset * blocksize);
2104                 BUG_ON(index == -1);
2105                 if (chunk->stripes[index].devid)
2106                         goto next_stripe;
2107                 ret = insert_stripe(&candidates, rc, chunk, index);
2108                 if (ret)
2109                         goto fail_out;
2110         } else {
2111                 csum_offset++;
2112                 goto next_csum;
2113         }
2114 next_stripe:
2115         start = btrfs_next_stripe_logical_offset(chunk, start);
2116         end = min(start + chunk->stripe_len, chunk_end);
2117         list_splice_init(&unordered, &candidates);
2118         btrfs_release_path(&path);
2119         csum_offset = 0;
2120         if (end < chunk_end)
2121                 goto again;
2122 out:
2123         ret = 0;
2124         list_splice_init(&candidates, &unordered);
2125         num_unordered = count_devext_records(&unordered);
2126         if (num_unordered == 1) {
2127                 for (i = 0; i < chunk->num_stripes; i++) {
2128                         if (!chunk->stripes[i].devid) {
2129                                 index = i;
2130                                 break;
2131                         }
2132                 }
2133                 ret = insert_stripe(&unordered, rc, chunk, index);
2134                 if (ret)
2135                         goto fail_out;
2136         } else {
2137                 if ((num_unordered == 2 && chunk->type_flags
2138                         & BTRFS_BLOCK_GROUP_RAID5)
2139                  || (num_unordered == 3 && chunk->type_flags
2140                         & BTRFS_BLOCK_GROUP_RAID6)) {
2141                         ret = fill_chunk_up(chunk, &unordered, rc);
2142                 }
2143         }
2144 fail_out:
2145         ret = !!ret || (list_empty(&unordered) ? 0 : 1);
2146         list_splice_init(&candidates, &chunk->dextents);
2147         list_splice_init(&unordered, &chunk->dextents);
2148         btrfs_release_path(&path);
2149
2150         return ret;
2151 }
2152
2153 static int btrfs_rebuild_ordered_data_chunk_stripes(struct recover_control *rc,
2154                                            struct btrfs_root *root)
2155 {
2156         struct chunk_record *chunk;
2157         struct chunk_record *next;
2158         int ret = 0;
2159         int err;
2160         u8 flags;
2161
2162         list_for_each_entry_safe(chunk, next, &rc->unrepaired_chunks, list) {
2163                 if ((chunk->type_flags & BTRFS_BLOCK_GROUP_DATA)
2164                  && (chunk->type_flags & BTRFS_ORDERED_RAID)) {
2165                         flags = 0;
2166                         err = rebuild_raid_data_chunk_stripes(rc, root, chunk,
2167                                                               &flags);
2168                         if (err) {
2169                                 list_move(&chunk->list, &rc->bad_chunks);
2170                                 if (flags & EQUAL_STRIPE)
2171                                         fprintf(stderr,
2172                         "Failure: too many equal stripes in chunk[%llu %llu]\n",
2173                                                 chunk->offset, chunk->length);
2174                                 if (!ret)
2175                                         ret = err;
2176                         } else
2177                                 list_move(&chunk->list, &rc->good_chunks);
2178                 }
2179         }
2180         return ret;
2181 }
2182
2183 static int btrfs_recover_chunks(struct recover_control *rc)
2184 {
2185         struct chunk_record *chunk;
2186         struct block_group_record *bg;
2187         struct block_group_record *next;
2188         LIST_HEAD(new_chunks);
2189         LIST_HEAD(devexts);
2190         int nstripes;
2191         int ret;
2192
2193         /* create the chunk by block group */
2194         list_for_each_entry_safe(bg, next, &rc->bg.block_groups, list) {
2195                 nstripes = btrfs_get_device_extents(bg->objectid,
2196                                                     &rc->devext.no_chunk_orphans,
2197                                                     &devexts);
2198                 chunk = malloc(btrfs_chunk_record_size(nstripes));
2199                 if (!chunk)
2200                         return -ENOMEM;
2201                 memset(chunk, 0, btrfs_chunk_record_size(nstripes));
2202                 INIT_LIST_HEAD(&chunk->dextents);
2203                 chunk->bg_rec = bg;
2204                 chunk->cache.start = bg->objectid;
2205                 chunk->cache.size = bg->offset;
2206                 chunk->objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2207                 chunk->type = BTRFS_CHUNK_ITEM_KEY;
2208                 chunk->offset = bg->objectid;
2209                 chunk->generation = bg->generation;
2210                 chunk->length = bg->offset;
2211                 chunk->owner = BTRFS_CHUNK_TREE_OBJECTID;
2212                 chunk->stripe_len = BTRFS_STRIPE_LEN;
2213                 chunk->type_flags = bg->flags;
2214                 chunk->io_width = BTRFS_STRIPE_LEN;
2215                 chunk->io_align = BTRFS_STRIPE_LEN;
2216                 chunk->sector_size = rc->sectorsize;
2217                 chunk->sub_stripes = calc_sub_nstripes(bg->flags);
2218
2219                 ret = insert_cache_extent(&rc->chunk, &chunk->cache);
2220                 BUG_ON(ret);
2221
2222                 list_del_init(&bg->list);
2223                 if (!nstripes) {
2224                         list_add_tail(&chunk->list, &rc->bad_chunks);
2225                         continue;
2226                 }
2227
2228                 list_splice_init(&devexts, &chunk->dextents);
2229
2230                 ret = btrfs_verify_device_extents(bg, &devexts, nstripes);
2231                 if (ret) {
2232                         list_add_tail(&chunk->list, &rc->bad_chunks);
2233                         continue;
2234                 }
2235
2236                 chunk->num_stripes = nstripes;
2237                 ret = btrfs_rebuild_chunk_stripes(rc, chunk);
2238                 if (ret > 0)
2239                         list_add_tail(&chunk->list, &rc->unrepaired_chunks);
2240                 else if (ret < 0)
2241                         list_add_tail(&chunk->list, &rc->bad_chunks);
2242                 else
2243                         list_add_tail(&chunk->list, &rc->good_chunks);
2244         }
2245         /*
2246          * Don't worry about the lost orphan device extents, they don't
2247          * have its chunk and block group, they must be the old ones that
2248          * we have dropped.
2249          */
2250         return 0;
2251 }
2252
2253 static inline int is_chunk_overlap(struct chunk_record *chunk1,
2254                                    struct chunk_record *chunk2)
2255 {
2256         if (chunk1->offset >= chunk2->offset + chunk2->length ||
2257             chunk1->offset + chunk1->length <= chunk2->offset)
2258                 return 0;
2259         return 1;
2260 }
2261
2262 /* Move invalid(overlap with good chunks) rebuild chunks to bad chunk list */
2263 static void validate_rebuild_chunks(struct recover_control *rc)
2264 {
2265         struct chunk_record *good;
2266         struct chunk_record *rebuild;
2267         struct chunk_record *tmp;
2268
2269         list_for_each_entry_safe(rebuild, tmp, &rc->rebuild_chunks, list) {
2270                 list_for_each_entry(good, &rc->good_chunks, list) {
2271                         if (is_chunk_overlap(rebuild, good)) {
2272                                 list_move_tail(&rebuild->list,
2273                                                &rc->bad_chunks);
2274                                 break;
2275                         }
2276                 }
2277         }
2278 }
2279
2280 /*
2281  * Return 0 when succesful, < 0 on error and > 0 if aborted by user
2282  */
2283 int btrfs_recover_chunk_tree(char *path, int verbose, int yes)
2284 {
2285         int ret = 0;
2286         struct btrfs_root *root = NULL;
2287         struct btrfs_trans_handle *trans;
2288         struct recover_control rc;
2289
2290         init_recover_control(&rc, verbose, yes);
2291
2292         ret = recover_prepare(&rc, path);
2293         if (ret) {
2294                 fprintf(stderr, "recover prepare error\n");
2295                 return ret;
2296         }
2297
2298         ret = scan_devices(&rc);
2299         if (ret) {
2300                 fprintf(stderr, "scan chunk headers error\n");
2301                 goto fail_rc;
2302         }
2303
2304         if (cache_tree_empty(&rc.chunk) &&
2305             cache_tree_empty(&rc.bg.tree) &&
2306             cache_tree_empty(&rc.devext.tree)) {
2307                 fprintf(stderr, "no recoverable chunk\n");
2308                 goto fail_rc;
2309         }
2310
2311         print_scan_result(&rc);
2312
2313         ret = check_chunks(&rc.chunk, &rc.bg, &rc.devext, &rc.good_chunks,
2314                            &rc.bad_chunks, &rc.rebuild_chunks, 1);
2315         if (ret) {
2316                 if (!list_empty(&rc.bg.block_groups) ||
2317                     !list_empty(&rc.devext.no_chunk_orphans)) {
2318                         ret = btrfs_recover_chunks(&rc);
2319                         if (ret)
2320                                 goto fail_rc;
2321                 }
2322         } else {
2323                 print_check_result(&rc);
2324                 printf("Check chunks successfully with no orphans\n");
2325                 goto fail_rc;
2326         }
2327         validate_rebuild_chunks(&rc);
2328         print_check_result(&rc);
2329
2330         root = open_ctree_with_broken_chunk(&rc);
2331         if (IS_ERR(root)) {
2332                 fprintf(stderr, "open with broken chunk error\n");
2333                 ret = PTR_ERR(root);
2334                 goto fail_rc;
2335         }
2336
2337         ret = check_all_chunks_by_metadata(&rc, root);
2338         if (ret) {
2339                 fprintf(stderr, "The chunks in memory can not match the metadata of the fs. Repair failed.\n");
2340                 goto fail_close_ctree;
2341         }
2342
2343         ret = btrfs_rebuild_ordered_data_chunk_stripes(&rc, root);
2344         if (ret) {
2345                 fprintf(stderr, "Failed to rebuild ordered chunk stripes.\n");
2346                 goto fail_close_ctree;
2347         }
2348
2349         if (!rc.yes) {
2350                 ret = ask_user("We are going to rebuild the chunk tree on disk, it might destroy the old metadata on the disk, Are you sure?");
2351                 if (!ret) {
2352                         ret = 1;
2353                         goto fail_close_ctree;
2354                 }
2355         }
2356
2357         trans = btrfs_start_transaction(root, 1);
2358         ret = remove_chunk_extent_item(trans, &rc, root);
2359         BUG_ON(ret);
2360
2361         ret = rebuild_chunk_tree(trans, &rc, root);
2362         BUG_ON(ret);
2363
2364         ret = rebuild_sys_array(&rc, root);
2365         BUG_ON(ret);
2366
2367         ret = rebuild_block_group(trans, &rc, root);
2368         if (ret) {
2369                 printf("Fail to rebuild block groups.\n");
2370                 printf("Recommend to run 'btrfs check --init-extent-tree <dev>' after recovery\n");
2371         }
2372
2373         btrfs_commit_transaction(trans, root);
2374 fail_close_ctree:
2375         close_ctree(root);
2376 fail_rc:
2377         free_recover_control(&rc);
2378         return ret;
2379 }