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 nodesize;
  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         u64 bytenr;
  82 };
  83
  84 static struct extent_record *btrfs_new_extent_record(struct extent_buffer *eb)
  85 {
  86         struct extent_record *rec;
  87
  88         rec = calloc(1, sizeof(*rec));
  89         if (!rec) {
  90                 fprintf(stderr, "Fail to allocate memory for extent record.\n");
  91                 exit(1);
  92         }
  93
  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 impossible, 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 search 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("\tnodesize: %d\n", rc->nodesize);
 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 mismatch 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 different 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->nodesize);
 765         if (!buf)
 766                 return -ENOMEM;
 767         buf->len = rc->nodesize;
 768
 769         bytenr = 0;
 770         while (1) {
 771                 dev_scan->bytenr = bytenr;
 772
 773                 if (is_super_block_address(bytenr))
 774                         bytenr += rc->sectorsize;
 775
 776                 if (pread64(fd, buf->data, rc->nodesize, bytenr) <
 777                     rc->nodesize)
 778                         break;
 779
 780                 if (memcmp_extent_buffer(buf, rc->fs_devices->fsid,
 781                                          btrfs_header_fsid(),
 782                                          BTRFS_FSID_SIZE)) {
 783                         bytenr += rc->sectorsize;
 784                         continue;
 785                 }
 786
 787                 if (verify_tree_block_csum_silent(buf, rc->csum_size)) {
 788                         bytenr += rc->sectorsize;
 789                         continue;
 790                 }
 791
 792                 pthread_mutex_lock(&rc->rc_lock);
 793                 ret = process_extent_buffer(&rc->eb_cache, buf, device, bytenr);
 794                 pthread_mutex_unlock(&rc->rc_lock);
 795                 if (ret)
 796                         goto out;
 797
 798                 if (btrfs_header_level(buf) != 0)
 799                         goto next_node;
 800
 801                 switch (btrfs_header_owner(buf)) {
 802                 case BTRFS_EXTENT_TREE_OBJECTID:
 803                 case BTRFS_DEV_TREE_OBJECTID:
 804                         /* different tree use different generation */
 805                         if (btrfs_header_generation(buf) > rc->generation)
 806                                 break;
 807                         ret = extract_metadata_record(rc, buf);
 808                         if (ret)
 809                                 goto out;
 810                         break;
 811                 case BTRFS_CHUNK_TREE_OBJECTID:
 812                         if (btrfs_header_generation(buf) >
 813                             rc->chunk_root_generation)
 814                                 break;
 815                         ret = extract_metadata_record(rc, buf);
 816                         if (ret)
 817                                 goto out;
 818                         break;
 819                 }
 820 next_node:
 821                 bytenr += rc->nodesize;
 822         }
 823 out:
 824         close(fd);
 825         free(buf);
 826         return ret;
 827 }
 828
 829 static int scan_devices(struct recover_control *rc)
 830 {
 831         int ret = 0;
 832         int fd;
 833         struct btrfs_device *dev;
 834         struct device_scan *dev_scans;
 835         pthread_t *t_scans;
 836         long *t_rets;
 837         int devnr = 0;
 838         int devidx = 0;
 839         int i;
 840         int all_done;
 841
 842         list_for_each_entry(dev, &rc->fs_devices->devices, dev_list)
 843                 devnr++;
 844         dev_scans = (struct device_scan *)malloc(sizeof(struct device_scan)
 845                                                  * devnr);
 846         if (!dev_scans)
 847                 return -ENOMEM;
 848         t_scans = (pthread_t *)malloc(sizeof(pthread_t) * devnr);
 849         if (!t_scans) {
 850                 free(dev_scans);
 851                 return -ENOMEM;
 852         }
 853         t_rets = (long *)malloc(sizeof(long) * devnr);
 854         if (!t_rets) {
 855                 free(dev_scans);
 856                 free(t_scans);
 857                 return -ENOMEM;
 858         }
 859
 860         list_for_each_entry(dev, &rc->fs_devices->devices, dev_list) {
 861                 fd = open(dev->name, O_RDONLY);
 862                 if (fd < 0) {
 863                         fprintf(stderr, "Failed to open device %s\n",
 864                                 dev->name);
 865                         ret = 1;
 866                         goto out2;
 867                 }
 868                 dev_scans[devidx].rc = rc;
 869                 dev_scans[devidx].dev = dev;
 870                 dev_scans[devidx].fd = fd;
 871                 dev_scans[devidx].bytenr = -1;
 872                 devidx++;
 873         }
 874
 875         for (i = 0; i < devidx; i++) {
 876                 ret = pthread_create(&t_scans[i], NULL,
 877                                      (void *)scan_one_device,
 878                                      (void *)&dev_scans[i]);
 879                 if (ret)
 880                         goto out1;
 881
 882                 dev_scans[i].bytenr = 0;
 883         }
 884
 885         while (1) {
 886                 all_done = 1;
 887                 for (i = 0; i < devidx; i++) {
 888                         if (dev_scans[i].bytenr == -1)
 889                                 continue;
 890                         ret = pthread_tryjoin_np(t_scans[i],
 891                                                  (void **)&t_rets[i]);
 892                         if (ret == EBUSY) {
 893                                 all_done = 0;
 894                                 continue;
 895                         }
 896                         if (ret || t_rets[i]) {
 897                                 ret = 1;
 898                                 goto out1;
 899                         }
 900                         dev_scans[i].bytenr = -1;
 901                 }
 902
 903                 printf("\rScanning: ");
 904                 for (i = 0; i < devidx; i++) {
 905                         if (dev_scans[i].bytenr == -1)
 906                                 printf("%sDONE in dev%d",
 907                                        i ? ", " : "", i);
 908                         else
 909                                 printf("%s%llu in dev%d",
 910                                        i ? ", " : "", dev_scans[i].bytenr, i);
 911                 }
 912                 /* clear chars if exist in tail */
 913                 printf("                ");
 914                 printf("\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b");
 915                 fflush(stdout);
 916
 917                 if (all_done) {
 918                         printf("\n");
 919                         break;
 920                 }
 921
 922                 sleep(1);
 923         }
 924 out1:
 925         for (i = 0; i < devidx; i++) {
 926                 if (dev_scans[i].bytenr == -1)
 927                         continue;
 928                 pthread_cancel(t_scans[i]);
 929         }
 930 out2:
 931         free(dev_scans);
 932         free(t_scans);
 933         free(t_rets);
 934         return !!ret;
 935 }
 936
 937 static int build_device_map_by_chunk_record(struct btrfs_root *root,
 938                                             struct chunk_record *chunk)
 939 {
 940         int ret = 0;
 941         int i;
 942         u64 devid;
 943         u8 uuid[BTRFS_UUID_SIZE];
 944         u16 num_stripes;
 945         struct btrfs_mapping_tree *map_tree;
 946         struct map_lookup *map;
 947         struct stripe *stripe;
 948
 949         map_tree = &root->fs_info->mapping_tree;
 950         num_stripes = chunk->num_stripes;
 951         map = malloc(btrfs_map_lookup_size(num_stripes));
 952         if (!map)
 953                 return -ENOMEM;
 954         map->ce.start = chunk->offset;
 955         map->ce.size = chunk->length;
 956         map->num_stripes = num_stripes;
 957         map->io_width = chunk->io_width;
 958         map->io_align = chunk->io_align;
 959         map->sector_size = chunk->sector_size;
 960         map->stripe_len = chunk->stripe_len;
 961         map->type = chunk->type_flags;
 962         map->sub_stripes = chunk->sub_stripes;
 963
 964         for (i = 0, stripe = chunk->stripes; i < num_stripes; i++, stripe++) {
 965                 devid = stripe->devid;
 966                 memcpy(uuid, stripe->dev_uuid, BTRFS_UUID_SIZE);
 967                 map->stripes[i].physical = stripe->offset;
 968                 map->stripes[i].dev = btrfs_find_device(root, devid,
 969                                                         uuid, NULL);
 970                 if (!map->stripes[i].dev) {
 971                         free(map);
 972                         return -EIO;
 973                 }
 974         }
 975
 976         ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
 977         return ret;
 978 }
 979
 980 static int build_device_maps_by_chunk_records(struct recover_control *rc,
 981                                               struct btrfs_root *root)
 982 {
 983         int ret = 0;
 984         struct chunk_record *chunk;
 985
 986         list_for_each_entry(chunk, &rc->good_chunks, list) {
 987                 ret = build_device_map_by_chunk_record(root, chunk);
 988                 if (ret)
 989                         return ret;
 990         }
 991         list_for_each_entry(chunk, &rc->rebuild_chunks, list) {
 992                 ret = build_device_map_by_chunk_record(root, chunk);
 993                 if (ret)
 994                         return ret;
 995         }
 996         return ret;
 997 }
 998
 999 static int block_group_remove_all_extent_items(struct btrfs_trans_handle *trans,
1000                                                struct btrfs_root *root,
1001                                                struct block_group_record *bg)
1002 {
1003         struct btrfs_fs_info *fs_info = root->fs_info;
1004         struct btrfs_key key;
1005         struct btrfs_path path;
1006         struct extent_buffer *leaf;
1007         u64 start = bg->objectid;
1008         u64 end = bg->objectid + bg->offset;
1009         u64 old_val;
1010         int nitems;
1011         int ret;
1012         int i;
1013         int del_s, del_nr;
1014
1015         btrfs_init_path(&path);
1016         root = root->fs_info->extent_root;
1017
1018         key.objectid = start;
1019         key.offset = 0;
1020         key.type = BTRFS_EXTENT_ITEM_KEY;
1021 again:
1022         ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
1023         if (ret < 0)
1024                 goto err;
1025         else if (ret > 0)
1026                 ret = 0;
1027
1028         leaf = path.nodes[0];
1029         nitems = btrfs_header_nritems(leaf);
1030         if (!nitems) {
1031                 /* The tree is empty. */
1032                 ret = 0;
1033                 goto err;
1034         }
1035
1036         if (path.slots[0] >= nitems) {
1037                 ret = btrfs_next_leaf(root, &path);
1038                 if (ret < 0)
1039                         goto err;
1040                 if (ret > 0) {
1041                         ret = 0;
1042                         goto err;
1043                 }
1044                 leaf = path.nodes[0];
1045                 btrfs_item_key_to_cpu(leaf, &key, 0);
1046                 if (key.objectid >= end)
1047                         goto err;
1048                 btrfs_release_path(&path);
1049                 goto again;
1050         }
1051
1052         del_nr = 0;
1053         del_s = -1;
1054         for (i = path.slots[0]; i < nitems; i++) {
1055                 btrfs_item_key_to_cpu(leaf, &key, i);
1056                 if (key.objectid >= end)
1057                         break;
1058
1059                 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
1060                         if (del_nr == 0)
1061                                 continue;
1062                         else
1063                                 break;
1064                 }
1065
1066                 if (del_s == -1)
1067                         del_s = i;
1068                 del_nr++;
1069                 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
1070                     key.type == BTRFS_METADATA_ITEM_KEY) {
1071                         old_val = btrfs_super_bytes_used(fs_info->super_copy);
1072                         if (key.type == BTRFS_METADATA_ITEM_KEY)
1073                                 old_val += root->nodesize;
1074                         else
1075                                 old_val += key.offset;
1076                         btrfs_set_super_bytes_used(fs_info->super_copy,
1077                                                    old_val);
1078                 }
1079         }
1080
1081         if (del_nr) {
1082                 ret = btrfs_del_items(trans, root, &path, del_s, del_nr);
1083                 if (ret)
1084                         goto err;
1085         }
1086
1087         if (key.objectid < end) {
1088                 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
1089                         key.objectid += root->sectorsize;
1090                         key.type = BTRFS_EXTENT_ITEM_KEY;
1091                         key.offset = 0;
1092                 }
1093                 btrfs_release_path(&path);
1094                 goto again;
1095         }
1096 err:
1097         btrfs_release_path(&path);
1098         return ret;
1099 }
1100
1101 static int block_group_free_all_extent(struct btrfs_root *root,
1102                                        struct block_group_record *bg)
1103 {
1104         struct btrfs_block_group_cache *cache;
1105         struct btrfs_fs_info *info;
1106         u64 start;
1107         u64 end;
1108
1109         info = root->fs_info;
1110         cache = btrfs_lookup_block_group(info, bg->objectid);
1111         if (!cache)
1112                 return -ENOENT;
1113
1114         start = cache->key.objectid;
1115         end = start + cache->key.offset - 1;
1116
1117         set_extent_bits(&info->block_group_cache, start, end,
1118                         BLOCK_GROUP_DIRTY, GFP_NOFS);
1119         set_extent_dirty(&info->free_space_cache, start, end, GFP_NOFS);
1120
1121         btrfs_set_block_group_used(&cache->item, 0);
1122
1123         return 0;
1124 }
1125
1126 static int remove_chunk_extent_item(struct btrfs_trans_handle *trans,
1127                                     struct recover_control *rc,
1128                                     struct btrfs_root *root)
1129 {
1130         struct chunk_record *chunk;
1131         int ret = 0;
1132
1133         list_for_each_entry(chunk, &rc->good_chunks, list) {
1134                 if (!(chunk->type_flags & BTRFS_BLOCK_GROUP_SYSTEM))
1135                         continue;
1136                 ret = block_group_remove_all_extent_items(trans, root,
1137                                                           chunk->bg_rec);
1138                 if (ret)
1139                         return ret;
1140
1141                 ret = block_group_free_all_extent(root, chunk->bg_rec);
1142                 if (ret)
1143                         return ret;
1144         }
1145         return ret;
1146 }
1147
1148 static int __rebuild_chunk_root(struct btrfs_trans_handle *trans,
1149                                 struct recover_control *rc,
1150                                 struct btrfs_root *root)
1151 {
1152         u64 min_devid = -1;
1153         struct btrfs_device *dev;
1154         struct extent_buffer *cow;
1155         struct btrfs_disk_key disk_key;
1156         int ret = 0;
1157
1158         list_for_each_entry(dev, &rc->fs_devices->devices, dev_list) {
1159                 if (min_devid > dev->devid)
1160                         min_devid = dev->devid;
1161         }
1162         btrfs_set_disk_key_objectid(&disk_key, BTRFS_DEV_ITEMS_OBJECTID);
1163         btrfs_set_disk_key_type(&disk_key, BTRFS_DEV_ITEM_KEY);
1164         btrfs_set_disk_key_offset(&disk_key, min_devid);
1165
1166         cow = btrfs_alloc_free_block(trans, root, root->nodesize,
1167                                      BTRFS_CHUNK_TREE_OBJECTID,
1168                                      &disk_key, 0, 0, 0);
1169         btrfs_set_header_bytenr(cow, cow->start);
1170         btrfs_set_header_generation(cow, trans->transid);
1171         btrfs_set_header_nritems(cow, 0);
1172         btrfs_set_header_level(cow, 0);
1173         btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
1174         btrfs_set_header_owner(cow, BTRFS_CHUNK_TREE_OBJECTID);
1175         write_extent_buffer(cow, root->fs_info->fsid,
1176                         btrfs_header_fsid(), BTRFS_FSID_SIZE);
1177
1178         write_extent_buffer(cow, root->fs_info->chunk_tree_uuid,
1179                         btrfs_header_chunk_tree_uuid(cow),
1180                         BTRFS_UUID_SIZE);
1181
1182         root->node = cow;
1183         btrfs_mark_buffer_dirty(cow);
1184
1185         return ret;
1186 }
1187
1188 static int __rebuild_device_items(struct btrfs_trans_handle *trans,
1189                                   struct recover_control *rc,
1190                                   struct btrfs_root *root)
1191 {
1192         struct btrfs_device *dev;
1193         struct btrfs_key key;
1194         struct btrfs_dev_item dev_item_tmp;
1195         struct btrfs_dev_item *dev_item = &dev_item_tmp;
1196         int ret = 0;
1197
1198         list_for_each_entry(dev, &rc->fs_devices->devices, dev_list) {
1199                 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1200                 key.type = BTRFS_DEV_ITEM_KEY;
1201                 key.offset = dev->devid;
1202
1203                 btrfs_set_stack_device_generation(dev_item, 0);
1204                 btrfs_set_stack_device_type(dev_item, dev->type);
1205                 btrfs_set_stack_device_id(dev_item, dev->devid);
1206                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1207                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1208                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1209                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1210                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1211                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1212                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
1213
1214                 ret = btrfs_insert_item(trans, root, &key,
1215                                         dev_item, sizeof(*dev_item));
1216         }
1217
1218         return ret;
1219 }
1220
1221 static int __insert_chunk_item(struct btrfs_trans_handle *trans,
1222                                 struct chunk_record *chunk_rec,
1223                                 struct btrfs_root *chunk_root)
1224 {
1225         struct btrfs_key key;
1226         struct btrfs_chunk *chunk = NULL;
1227         int ret = 0;
1228
1229         chunk = create_chunk_item(chunk_rec);
1230         if (!chunk)
1231                 return -ENOMEM;
1232         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1233         key.type = BTRFS_CHUNK_ITEM_KEY;
1234         key.offset = chunk_rec->offset;
1235
1236         ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1237                                 btrfs_chunk_item_size(chunk_rec->num_stripes));
1238         free(chunk);
1239         return ret;
1240 }
1241
1242 static int __rebuild_chunk_items(struct btrfs_trans_handle *trans,
1243                                  struct recover_control *rc,
1244                                  struct btrfs_root *root)
1245 {
1246         struct btrfs_root *chunk_root;
1247         struct chunk_record *chunk_rec;
1248         int ret;
1249
1250         chunk_root = root->fs_info->chunk_root;
1251
1252         list_for_each_entry(chunk_rec, &rc->good_chunks, list) {
1253                 ret = __insert_chunk_item(trans, chunk_rec, chunk_root);
1254                 if (ret)
1255                         return ret;
1256         }
1257         list_for_each_entry(chunk_rec, &rc->rebuild_chunks, list) {
1258                 ret = __insert_chunk_item(trans, chunk_rec, chunk_root);
1259                 if (ret)
1260                         return ret;
1261         }
1262         return 0;
1263 }
1264
1265 static int rebuild_chunk_tree(struct btrfs_trans_handle *trans,
1266                               struct recover_control *rc,
1267                               struct btrfs_root *root)
1268 {
1269         int ret = 0;
1270
1271         root = root->fs_info->chunk_root;
1272
1273         ret = __rebuild_chunk_root(trans, rc, root);
1274         if (ret)
1275                 return ret;
1276
1277         ret = __rebuild_device_items(trans, rc, root);
1278         if (ret)
1279                 return ret;
1280
1281         ret = __rebuild_chunk_items(trans, rc, root);
1282
1283         return ret;
1284 }
1285
1286 static int rebuild_sys_array(struct recover_control *rc,
1287                              struct btrfs_root *root)
1288 {
1289         struct btrfs_chunk *chunk;
1290         struct btrfs_key key;
1291         struct chunk_record *chunk_rec;
1292         int ret = 0;
1293         u16 num_stripes;
1294
1295         btrfs_set_super_sys_array_size(root->fs_info->super_copy, 0);
1296
1297         list_for_each_entry(chunk_rec, &rc->good_chunks, list) {
1298                 if (!(chunk_rec->type_flags & BTRFS_BLOCK_GROUP_SYSTEM))
1299                         continue;
1300
1301                 num_stripes = chunk_rec->num_stripes;
1302                 chunk = create_chunk_item(chunk_rec);
1303                 if (!chunk) {
1304                         ret = -ENOMEM;
1305                         break;
1306                 }
1307
1308                 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1309                 key.type = BTRFS_CHUNK_ITEM_KEY;
1310                 key.offset = chunk_rec->offset;
1311
1312                 ret = btrfs_add_system_chunk(NULL, root, &key, chunk,
1313                                 btrfs_chunk_item_size(num_stripes));
1314                 free(chunk);
1315                 if (ret)
1316                         break;
1317         }
1318         return ret;
1319
1320 }
1321
1322 static int calculate_bg_used(struct btrfs_root *extent_root,
1323                              struct chunk_record *chunk_rec,
1324                              struct btrfs_path *path,
1325                              u64 *used)
1326 {
1327         struct extent_buffer *node;
1328         struct btrfs_key found_key;
1329         int slot;
1330         int ret = 0;
1331         u64 used_ret = 0;
1332
1333         while (1) {
1334                 node = path->nodes[0];
1335                 slot = path->slots[0];
1336                 btrfs_item_key_to_cpu(node, &found_key, slot);
1337                 if (found_key.objectid >= chunk_rec->offset + chunk_rec->length)
1338                         break;
1339                 if (found_key.type != BTRFS_METADATA_ITEM_KEY &&
1340                     found_key.type != BTRFS_EXTENT_DATA_KEY)
1341                         goto next;
1342                 if (found_key.type == BTRFS_METADATA_ITEM_KEY)
1343                         used_ret += extent_root->nodesize;
1344                 else
1345                         used_ret += found_key.offset;
1346 next:
1347                 if (slot + 1 < btrfs_header_nritems(node)) {
1348                         slot++;
1349                 } else {
1350                         ret = btrfs_next_leaf(extent_root, path);
1351                         if (ret > 0) {
1352                                 ret = 0;
1353                                 break;
1354                         }
1355                         if (ret < 0)
1356                                 break;
1357                 }
1358         }
1359         if (!ret)
1360                 *used = used_ret;
1361         return ret;
1362 }
1363
1364 static int __insert_block_group(struct btrfs_trans_handle *trans,
1365                                 struct chunk_record *chunk_rec,
1366                                 struct btrfs_root *extent_root,
1367                                 u64 used)
1368 {
1369         struct btrfs_block_group_item bg_item;
1370         struct btrfs_key key;
1371         int ret = 0;
1372
1373         btrfs_set_block_group_used(&bg_item, used);
1374         btrfs_set_block_group_chunk_objectid(&bg_item, used);
1375         btrfs_set_block_group_flags(&bg_item, chunk_rec->type_flags);
1376         key.objectid = chunk_rec->offset;
1377         key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
1378         key.offset = chunk_rec->length;
1379
1380         ret = btrfs_insert_item(trans, extent_root, &key, &bg_item,
1381                                 sizeof(bg_item));
1382         return ret;
1383 }
1384
1385 /*
1386  * Search through the extent tree to rebuild the 'used' member of the block
1387  * group.
1388  * However, since block group and extent item shares the extent tree,
1389  * the extent item may also missing.
1390  * In that case, we fill the 'used' with the length of the block group to
1391  * ensure no write into the block group.
1392  * Btrfsck will hate it but we will inform user to call '--init-extent-tree'
1393  * if possible, or just salvage as much data as possible from the fs.
1394  */
1395 static int rebuild_block_group(struct btrfs_trans_handle *trans,
1396                                struct recover_control *rc,
1397                                struct btrfs_root *root)
1398 {
1399         struct chunk_record *chunk_rec;
1400         struct btrfs_key search_key;
1401         struct btrfs_path *path;
1402         u64 used = 0;
1403         int ret = 0;
1404
1405         if (list_empty(&rc->rebuild_chunks))
1406                 return 0;
1407
1408         path = btrfs_alloc_path();
1409         if (!path)
1410                 return -ENOMEM;
1411         list_for_each_entry(chunk_rec, &rc->rebuild_chunks, list) {
1412                 search_key.objectid = chunk_rec->offset;
1413                 search_key.type = BTRFS_EXTENT_ITEM_KEY;
1414                 search_key.offset = 0;
1415                 ret = btrfs_search_slot(NULL, root->fs_info->extent_root,
1416                                         &search_key, path, 0, 0);
1417                 if (ret < 0)
1418                         goto out;
1419                 ret = calculate_bg_used(root->fs_info->extent_root,
1420                                         chunk_rec, path, &used);
1421                 /*
1422                  * Extent tree is damaged, better to rebuild the whole extent
1423                  * tree. Currently, change the used to chunk's len to prevent
1424                  * write/block reserve happening in that block group.
1425                  */
1426                 if (ret < 0) {
1427                         fprintf(stderr,
1428                                 "Fail to search extent tree for block group: [%llu,%llu]\n",
1429                                 chunk_rec->offset,
1430                                 chunk_rec->offset + chunk_rec->length);
1431                         fprintf(stderr,
1432                                 "Mark the block group full to prevent block rsv problems\n");
1433                         used = chunk_rec->length;
1434                 }
1435                 btrfs_release_path(path);
1436                 ret = __insert_block_group(trans, chunk_rec,
1437                                            root->fs_info->extent_root,
1438                                            used);
1439                 if (ret < 0)
1440                         goto out;
1441         }
1442 out:
1443         btrfs_free_path(path);
1444         return ret;
1445 }
1446
1447 static struct btrfs_root *
1448 open_ctree_with_broken_chunk(struct recover_control *rc)
1449 {
1450         struct btrfs_fs_info *fs_info;
1451         struct btrfs_super_block *disk_super;
1452         struct extent_buffer *eb;
1453         u32 sectorsize;
1454         u32 nodesize;
1455         u32 leafsize;
1456         u32 stripesize;
1457         int ret;
1458
1459         fs_info = btrfs_new_fs_info(1, BTRFS_SUPER_INFO_OFFSET);
1460         if (!fs_info) {
1461                 fprintf(stderr, "Failed to allocate memory for fs_info\n");
1462                 return ERR_PTR(-ENOMEM);
1463         }
1464         fs_info->is_chunk_recover = 1;
1465
1466         fs_info->fs_devices = rc->fs_devices;
1467         ret = btrfs_open_devices(fs_info->fs_devices, O_RDWR);
1468         if (ret)
1469                 goto out;
1470
1471         disk_super = fs_info->super_copy;
1472         ret = btrfs_read_dev_super(fs_info->fs_devices->latest_bdev,
1473                                    disk_super, fs_info->super_bytenr,
1474                                    SBREAD_RECOVER);
1475         if (ret) {
1476                 fprintf(stderr, "No valid btrfs found\n");
1477                 goto out_devices;
1478         }
1479
1480         memcpy(fs_info->fsid, &disk_super->fsid, BTRFS_FSID_SIZE);
1481
1482         ret = btrfs_check_fs_compatibility(disk_super, 1);
1483         if (ret)
1484                 goto out_devices;
1485
1486         nodesize = btrfs_super_nodesize(disk_super);
1487         leafsize = btrfs_super_leafsize(disk_super);
1488         sectorsize = btrfs_super_sectorsize(disk_super);
1489         stripesize = btrfs_super_stripesize(disk_super);
1490
1491         btrfs_setup_root(nodesize, leafsize, sectorsize, stripesize,
1492                      fs_info->chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1493
1494         ret = build_device_maps_by_chunk_records(rc, fs_info->chunk_root);
1495         if (ret)
1496                 goto out_cleanup;
1497
1498         ret = btrfs_setup_all_roots(fs_info, 0, 0);
1499         if (ret)
1500                 goto out_failed;
1501
1502         eb = fs_info->tree_root->node;
1503         read_extent_buffer(eb, fs_info->chunk_tree_uuid,
1504                            btrfs_header_chunk_tree_uuid(eb),
1505                            BTRFS_UUID_SIZE);
1506
1507         return fs_info->fs_root;
1508 out_failed:
1509         btrfs_release_all_roots(fs_info);
1510 out_cleanup:
1511         btrfs_cleanup_all_caches(fs_info);
1512 out_devices:
1513         btrfs_close_devices(fs_info->fs_devices);
1514 out:
1515         btrfs_free_fs_info(fs_info);
1516         return ERR_PTR(ret);
1517 }
1518
1519 static int recover_prepare(struct recover_control *rc, char *path)
1520 {
1521         int ret;
1522         int fd;
1523         struct btrfs_super_block *sb;
1524         char buf[BTRFS_SUPER_INFO_SIZE];
1525         struct btrfs_fs_devices *fs_devices;
1526
1527         ret = 0;
1528         fd = open(path, O_RDONLY);
1529         if (fd < 0) {
1530                 fprintf(stderr, "open %s\n error.\n", path);
1531                 return -1;
1532         }
1533
1534         sb = (struct btrfs_super_block*)buf;
1535         ret = btrfs_read_dev_super(fd, sb, BTRFS_SUPER_INFO_OFFSET,
1536                         SBREAD_RECOVER);
1537         if (ret) {
1538                 fprintf(stderr, "read super block error\n");
1539                 goto out_close_fd;
1540         }
1541
1542         rc->sectorsize = btrfs_super_sectorsize(sb);
1543         rc->nodesize = btrfs_super_nodesize(sb);
1544         rc->generation = btrfs_super_generation(sb);
1545         rc->chunk_root_generation = btrfs_super_chunk_root_generation(sb);
1546         rc->csum_size = btrfs_super_csum_size(sb);
1547
1548         /* if seed, the result of scanning below will be partial */
1549         if (btrfs_super_flags(sb) & BTRFS_SUPER_FLAG_SEEDING) {
1550                 fprintf(stderr, "this device is seed device\n");
1551                 ret = -1;
1552                 goto out_close_fd;
1553         }
1554
1555         ret = btrfs_scan_fs_devices(fd, path, &fs_devices, 0, SBREAD_RECOVER, 0);
1556         if (ret)
1557                 goto out_close_fd;
1558
1559         rc->fs_devices = fs_devices;
1560
1561         if (rc->verbose)
1562                 print_all_devices(&rc->fs_devices->devices);
1563
1564 out_close_fd:
1565         close(fd);
1566         return ret;
1567 }
1568
1569 static int btrfs_get_device_extents(u64 chunk_object,
1570                                     struct list_head *orphan_devexts,
1571                                     struct list_head *ret_list)
1572 {
1573         struct device_extent_record *devext;
1574         struct device_extent_record *next;
1575         int count = 0;
1576
1577         list_for_each_entry_safe(devext, next, orphan_devexts, chunk_list) {
1578                 if (devext->chunk_offset == chunk_object) {
1579                         list_move_tail(&devext->chunk_list, ret_list);
1580                         count++;
1581                 }
1582         }
1583         return count;
1584 }
1585
1586 static int calc_num_stripes(u64 type)
1587 {
1588         if (type & (BTRFS_BLOCK_GROUP_RAID0 |
1589                     BTRFS_BLOCK_GROUP_RAID10 |
1590                     BTRFS_BLOCK_GROUP_RAID5 |
1591                     BTRFS_BLOCK_GROUP_RAID6))
1592                 return 0;
1593         else if (type & (BTRFS_BLOCK_GROUP_RAID1 |
1594                          BTRFS_BLOCK_GROUP_DUP))
1595                 return 2;
1596         else
1597                 return 1;
1598 }
1599
1600 static inline int calc_sub_nstripes(u64 type)
1601 {
1602         if (type & BTRFS_BLOCK_GROUP_RAID10)
1603                 return 2;
1604         else
1605                 return 1;
1606 }
1607
1608 static int btrfs_verify_device_extents(struct block_group_record *bg,
1609                                        struct list_head *devexts, int ndevexts)
1610 {
1611         struct device_extent_record *devext;
1612         u64 stripe_length;
1613         int expected_num_stripes;
1614
1615         expected_num_stripes = calc_num_stripes(bg->flags);
1616         if (expected_num_stripes && expected_num_stripes != ndevexts)
1617                 return 1;
1618
1619         if (check_num_stripes(bg->flags, ndevexts) < 0)
1620                 return 1;
1621
1622         stripe_length = calc_stripe_length(bg->flags, bg->offset, ndevexts);
1623         list_for_each_entry(devext, devexts, chunk_list) {
1624                 if (devext->length != stripe_length)
1625                         return 1;
1626         }
1627         return 0;
1628 }
1629
1630 static int btrfs_rebuild_unordered_chunk_stripes(struct recover_control *rc,
1631                                                  struct chunk_record *chunk)
1632 {
1633         struct device_extent_record *devext;
1634         struct btrfs_device *device;
1635         int i;
1636
1637         devext = list_first_entry(&chunk->dextents, struct device_extent_record,
1638                                   chunk_list);
1639         for (i = 0; i < chunk->num_stripes; i++) {
1640                 chunk->stripes[i].devid = devext->objectid;
1641                 chunk->stripes[i].offset = devext->offset;
1642                 device = btrfs_find_device_by_devid(rc->fs_devices,
1643                                                     devext->objectid,
1644                                                     0);
1645                 if (!device)
1646                         return -ENOENT;
1647                 BUG_ON(btrfs_find_device_by_devid(rc->fs_devices,
1648                                                   devext->objectid,
1649                                                   1));
1650                 memcpy(chunk->stripes[i].dev_uuid, device->uuid,
1651                        BTRFS_UUID_SIZE);
1652                 devext = list_next_entry(devext, chunk_list);
1653         }
1654         return 0;
1655 }
1656
1657 static int btrfs_calc_stripe_index(struct chunk_record *chunk, u64 logical)
1658 {
1659         u64 offset = logical - chunk->offset;
1660         int stripe_nr;
1661         int nr_data_stripes;
1662         int index;
1663
1664         stripe_nr = offset / chunk->stripe_len;
1665         if (chunk->type_flags & BTRFS_BLOCK_GROUP_RAID0) {
1666                 index = stripe_nr % chunk->num_stripes;
1667         } else if (chunk->type_flags & BTRFS_BLOCK_GROUP_RAID10) {
1668                 index = stripe_nr % (chunk->num_stripes / chunk->sub_stripes);
1669                 index *= chunk->sub_stripes;
1670         } else if (chunk->type_flags & BTRFS_BLOCK_GROUP_RAID5) {
1671                 nr_data_stripes = chunk->num_stripes - 1;
1672                 index = stripe_nr % nr_data_stripes;
1673                 stripe_nr /= nr_data_stripes;
1674                 index = (index + stripe_nr) % chunk->num_stripes;
1675         } else if (chunk->type_flags & BTRFS_BLOCK_GROUP_RAID6) {
1676                 nr_data_stripes = chunk->num_stripes - 2;
1677                 index = stripe_nr % nr_data_stripes;
1678                 stripe_nr /= nr_data_stripes;
1679                 index = (index + stripe_nr) % chunk->num_stripes;
1680         } else {
1681                 return -1;
1682         }
1683         return index;
1684 }
1685
1686 /* calc the logical offset which is the start of the next stripe. */
1687 static inline u64 btrfs_next_stripe_logical_offset(struct chunk_record *chunk,
1688                                                    u64 logical)
1689 {
1690         u64 offset = logical - chunk->offset;
1691
1692         offset /= chunk->stripe_len;
1693         offset *= chunk->stripe_len;
1694         offset += chunk->stripe_len;
1695
1696         return offset + chunk->offset;
1697 }
1698
1699 static int is_extent_record_in_device_extent(struct extent_record *er,
1700                                              struct device_extent_record *dext,
1701                                              int *mirror)
1702 {
1703         int i;
1704
1705         for (i = 0; i < er->nmirrors; i++) {
1706                 if (er->devices[i]->devid == dext->objectid &&
1707                     er->offsets[i] >= dext->offset &&
1708                     er->offsets[i] < dext->offset + dext->length) {
1709                         *mirror = i;
1710                         return 1;
1711                 }
1712         }
1713         return 0;
1714 }
1715
1716 static int
1717 btrfs_rebuild_ordered_meta_chunk_stripes(struct recover_control *rc,
1718                                          struct chunk_record *chunk)
1719 {
1720         u64 start = chunk->offset;
1721         u64 end = chunk->offset + chunk->length;
1722         struct cache_extent *cache;
1723         struct extent_record *er;
1724         struct device_extent_record *devext;
1725         struct device_extent_record *next;
1726         struct btrfs_device *device;
1727         LIST_HEAD(devexts);
1728         int index;
1729         int mirror;
1730         int ret;
1731
1732         cache = lookup_cache_extent(&rc->eb_cache,
1733                                     start, chunk->length);
1734         if (!cache) {
1735                 /* No used space, we can reorder the stripes freely. */
1736                 ret = btrfs_rebuild_unordered_chunk_stripes(rc, chunk);
1737                 return ret;
1738         }
1739
1740         list_splice_init(&chunk->dextents, &devexts);
1741 again:
1742         er = container_of(cache, struct extent_record, cache);
1743         index = btrfs_calc_stripe_index(chunk, er->cache.start);
1744         BUG_ON(index == -1);
1745         if (chunk->stripes[index].devid)
1746                 goto next;
1747         list_for_each_entry_safe(devext, next, &devexts, chunk_list) {
1748                 if (is_extent_record_in_device_extent(er, devext, &mirror)) {
1749                         chunk->stripes[index].devid = devext->objectid;
1750                         chunk->stripes[index].offset = devext->offset;
1751                         memcpy(chunk->stripes[index].dev_uuid,
1752                                er->devices[mirror]->uuid,
1753                                BTRFS_UUID_SIZE);
1754                         index++;
1755                         list_move(&devext->chunk_list, &chunk->dextents);
1756                 }
1757         }
1758 next:
1759         start = btrfs_next_stripe_logical_offset(chunk, er->cache.start);
1760         if (start >= end)
1761                 goto no_extent_record;
1762
1763         cache = lookup_cache_extent(&rc->eb_cache, start, end - start);
1764         if (cache)
1765                 goto again;
1766 no_extent_record:
1767         if (list_empty(&devexts))
1768                 return 0;
1769
1770         if (chunk->type_flags & (BTRFS_BLOCK_GROUP_RAID5 |
1771                                  BTRFS_BLOCK_GROUP_RAID6)) {
1772                 /* Fixme: try to recover the order by the parity block. */
1773                 list_splice_tail(&devexts, &chunk->dextents);
1774                 return -EINVAL;
1775         }
1776
1777         /* There is no data on the lost stripes, we can reorder them freely. */
1778         for (index = 0; index < chunk->num_stripes; index++) {
1779                 if (chunk->stripes[index].devid)
1780                         continue;
1781
1782                 devext = list_first_entry(&devexts,
1783                                           struct device_extent_record,
1784                                            chunk_list);
1785                 list_move(&devext->chunk_list, &chunk->dextents);
1786
1787                 chunk->stripes[index].devid = devext->objectid;
1788                 chunk->stripes[index].offset = devext->offset;
1789                 device = btrfs_find_device_by_devid(rc->fs_devices,
1790                                                     devext->objectid,
1791                                                     0);
1792                 if (!device) {
1793                         list_splice_tail(&devexts, &chunk->dextents);
1794                         return -EINVAL;
1795                 }
1796                 BUG_ON(btrfs_find_device_by_devid(rc->fs_devices,
1797                                                   devext->objectid,
1798                                                   1));
1799                 memcpy(chunk->stripes[index].dev_uuid, device->uuid,
1800                        BTRFS_UUID_SIZE);
1801         }
1802         return 0;
1803 }
1804
1805 #define BTRFS_ORDERED_RAID      (BTRFS_BLOCK_GROUP_RAID0 |      \
1806                                  BTRFS_BLOCK_GROUP_RAID10 |     \
1807                                  BTRFS_BLOCK_GROUP_RAID5 |      \
1808                                  BTRFS_BLOCK_GROUP_RAID6)
1809
1810 static int btrfs_rebuild_chunk_stripes(struct recover_control *rc,
1811                                        struct chunk_record *chunk)
1812 {
1813         int ret;
1814
1815         /*
1816          * All the data in the system metadata chunk will be dropped,
1817          * so we need not guarantee that the data is right or not, that
1818          * is we can reorder the stripes in the system metadata chunk.
1819          */
1820         if ((chunk->type_flags & BTRFS_BLOCK_GROUP_METADATA) &&
1821             (chunk->type_flags & BTRFS_ORDERED_RAID))
1822                 ret =btrfs_rebuild_ordered_meta_chunk_stripes(rc, chunk);
1823         else if ((chunk->type_flags & BTRFS_BLOCK_GROUP_DATA) &&
1824                  (chunk->type_flags & BTRFS_ORDERED_RAID))
1825                 ret = 1;        /* Be handled after the fs is opened. */
1826         else
1827                 ret = btrfs_rebuild_unordered_chunk_stripes(rc, chunk);
1828
1829         return ret;
1830 }
1831
1832 static int next_csum(struct btrfs_root *root,
1833                      struct extent_buffer **leaf,
1834                      struct btrfs_path *path,
1835                      int *slot,
1836                      u64 *csum_offset,
1837                      u32 *tree_csum,
1838                      u64 end,
1839                      struct btrfs_key *key)
1840 {
1841         int ret = 0;
1842         struct btrfs_root *csum_root = root->fs_info->csum_root;
1843         struct btrfs_csum_item *csum_item;
1844         u32 blocksize = root->sectorsize;
1845         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
1846         int csums_in_item = btrfs_item_size_nr(*leaf, *slot) / csum_size;
1847
1848         if (*csum_offset >= csums_in_item) {
1849                 ++(*slot);
1850                 *csum_offset = 0;
1851                 if (*slot >= btrfs_header_nritems(*leaf)) {
1852                         ret = btrfs_next_leaf(csum_root, path);
1853                         if (ret < 0)
1854                                 return -1;
1855                         else if (ret > 0)
1856                                 return 1;
1857                         *leaf = path->nodes[0];
1858                         *slot = path->slots[0];
1859                 }
1860                 btrfs_item_key_to_cpu(*leaf, key, *slot);
1861         }
1862
1863         if (key->offset + (*csum_offset) * blocksize >= end)
1864                 return 2;
1865         csum_item = btrfs_item_ptr(*leaf, *slot, struct btrfs_csum_item);
1866         csum_item = (struct btrfs_csum_item *)((unsigned char *)csum_item
1867                                              + (*csum_offset) * csum_size);
1868         read_extent_buffer(*leaf, tree_csum,
1869                           (unsigned long)csum_item, csum_size);
1870         return ret;
1871 }
1872
1873 static u64 calc_data_offset(struct btrfs_key *key,
1874                             struct chunk_record *chunk,
1875                             u64 dev_offset,
1876                             u64 csum_offset,
1877                             u32 blocksize)
1878 {
1879         u64 data_offset;
1880         int logical_stripe_nr;
1881         int dev_stripe_nr;
1882         int nr_data_stripes;
1883
1884         data_offset = key->offset + csum_offset * blocksize - chunk->offset;
1885         nr_data_stripes = chunk->num_stripes;
1886
1887         if (chunk->type_flags & BTRFS_BLOCK_GROUP_RAID5)
1888                 nr_data_stripes -= 1;
1889         else if (chunk->type_flags & BTRFS_BLOCK_GROUP_RAID6)
1890                 nr_data_stripes -= 2;
1891
1892         logical_stripe_nr = data_offset / chunk->stripe_len;
1893         dev_stripe_nr = logical_stripe_nr / nr_data_stripes;
1894
1895         data_offset -= logical_stripe_nr * chunk->stripe_len;
1896         data_offset += dev_stripe_nr * chunk->stripe_len;
1897
1898         return dev_offset + data_offset;
1899 }
1900
1901 static int check_one_csum(int fd, u64 start, u32 len, u32 tree_csum)
1902 {
1903         char *data;
1904         int ret = 0;
1905         u32 csum_result = ~(u32)0;
1906
1907         data = malloc(len);
1908         if (!data)
1909                 return -1;
1910         ret = pread64(fd, data, len, start);
1911         if (ret < 0 || ret != len) {
1912                 ret = -1;
1913                 goto out;
1914         }
1915         ret = 0;
1916         csum_result = btrfs_csum_data(NULL, data, csum_result, len);
1917         btrfs_csum_final(csum_result, (u8 *)&csum_result);
1918         if (csum_result != tree_csum)
1919                 ret = 1;
1920 out:
1921         free(data);
1922         return ret;
1923 }
1924
1925 static u64 item_end_offset(struct btrfs_root *root, struct btrfs_key *key,
1926                            struct extent_buffer *leaf, int slot) {
1927         u32 blocksize = root->sectorsize;
1928         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
1929
1930         u64 offset = btrfs_item_size_nr(leaf, slot);
1931         offset /= csum_size;
1932         offset *= blocksize;
1933         offset += key->offset;
1934
1935         return offset;
1936 }
1937
1938 static int insert_stripe(struct list_head *devexts,
1939                          struct recover_control *rc,
1940                          struct chunk_record *chunk,
1941                          int index) {
1942         struct device_extent_record *devext;
1943         struct btrfs_device *dev;
1944
1945         devext = list_entry(devexts->next, struct device_extent_record,
1946                             chunk_list);
1947         dev = btrfs_find_device_by_devid(rc->fs_devices, devext->objectid,
1948                                         0);
1949         if (!dev)
1950                 return -ENOENT;
1951         if (btrfs_find_device_by_devid(rc->fs_devices, devext->objectid, 1)) {
1952                 error("unexpected: found another device with id %llu",
1953                                 (unsigned long long)devext->objectid);
1954                 return -EINVAL;
1955         }
1956
1957         chunk->stripes[index].devid = devext->objectid;
1958         chunk->stripes[index].offset = devext->offset;
1959         memcpy(chunk->stripes[index].dev_uuid, dev->uuid, BTRFS_UUID_SIZE);
1960
1961         list_move(&devext->chunk_list, &chunk->dextents);
1962
1963         return 0;
1964 }
1965
1966 static inline int count_devext_records(struct list_head *record_list)
1967 {
1968         int num_of_records = 0;
1969         struct device_extent_record *devext;
1970
1971         list_for_each_entry(devext, record_list, chunk_list)
1972                 num_of_records++;
1973
1974         return num_of_records;
1975 }
1976
1977 static int fill_chunk_up(struct chunk_record *chunk, struct list_head *devexts,
1978                          struct recover_control *rc)
1979 {
1980         int ret = 0;
1981         int i;
1982
1983         for (i = 0; i < chunk->num_stripes; i++) {
1984                 if (!chunk->stripes[i].devid) {
1985                         ret = insert_stripe(devexts, rc, chunk, i);
1986                         if (ret)
1987                                 break;
1988                 }
1989         }
1990
1991         return ret;
1992 }
1993
1994 #define EQUAL_STRIPE (1 << 0)
1995
1996 static int rebuild_raid_data_chunk_stripes(struct recover_control *rc,
1997                                            struct btrfs_root *root,
1998                                            struct chunk_record *chunk,
1999                                            u8 *flags)
2000 {
2001         int i;
2002         int ret = 0;
2003         int slot;
2004         struct btrfs_path path;
2005         struct btrfs_key prev_key;
2006         struct btrfs_key key;
2007         struct btrfs_root *csum_root;
2008         struct extent_buffer *leaf;
2009         struct device_extent_record *devext;
2010         struct device_extent_record *next;
2011         struct btrfs_device *dev;
2012         u64 start = chunk->offset;
2013         u64 end = start + chunk->stripe_len;
2014         u64 chunk_end = chunk->offset + chunk->length;
2015         u64 csum_offset = 0;
2016         u64 data_offset;
2017         u32 blocksize = root->sectorsize;
2018         u32 tree_csum;
2019         int index = 0;
2020         int num_unordered = 0;
2021         LIST_HEAD(unordered);
2022         LIST_HEAD(candidates);
2023
2024         csum_root = root->fs_info->csum_root;
2025         btrfs_init_path(&path);
2026         list_splice_init(&chunk->dextents, &candidates);
2027 again:
2028         if (list_is_last(candidates.next, &candidates))
2029                 goto out;
2030
2031         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
2032         key.type = BTRFS_EXTENT_CSUM_KEY;
2033         key.offset = start;
2034
2035         ret = btrfs_search_slot(NULL, csum_root, &key, &path, 0, 0);
2036         if (ret < 0) {
2037                 fprintf(stderr, "Search csum failed(%d)\n", ret);
2038                 goto fail_out;
2039         }
2040         leaf = path.nodes[0];
2041         slot = path.slots[0];
2042         if (ret > 0) {
2043                 if (slot >= btrfs_header_nritems(leaf)) {
2044                         ret = btrfs_next_leaf(csum_root, &path);
2045                         if (ret < 0) {
2046                                 fprintf(stderr,
2047                                         "Walk tree failed(%d)\n", ret);
2048                                 goto fail_out;
2049                         } else if (ret > 0) {
2050                                 slot = btrfs_header_nritems(leaf) - 1;
2051                                 btrfs_item_key_to_cpu(leaf, &key, slot);
2052                                 if (item_end_offset(root, &key, leaf, slot)
2053                                                                 > start) {
2054                                         csum_offset = start - key.offset;
2055                                         csum_offset /= blocksize;
2056                                         goto next_csum;
2057                                 }
2058                                 goto next_stripe;
2059                         }
2060                         leaf = path.nodes[0];
2061                         slot = path.slots[0];
2062                 }
2063                 btrfs_item_key_to_cpu(leaf, &key, slot);
2064                 ret = btrfs_previous_item(csum_root, &path, 0,
2065                                           BTRFS_EXTENT_CSUM_KEY);
2066                 if (ret < 0)
2067                         goto fail_out;
2068                 else if (ret > 0) {
2069                         if (key.offset >= end)
2070                                 goto next_stripe;
2071                         else
2072                                 goto next_csum;
2073                 }
2074                 leaf = path.nodes[0];
2075                 slot = path.slots[0];
2076
2077                 btrfs_item_key_to_cpu(leaf, &prev_key, slot);
2078                 if (item_end_offset(root, &prev_key, leaf, slot) > start) {
2079                         csum_offset = start - prev_key.offset;
2080                         csum_offset /= blocksize;
2081                         btrfs_item_key_to_cpu(leaf, &key, slot);
2082                 } else {
2083                         if (key.offset >= end)
2084                                 goto next_stripe;
2085                 }
2086
2087                 if (key.offset + csum_offset * blocksize > chunk_end)
2088                         goto out;
2089         }
2090 next_csum:
2091         ret = next_csum(root, &leaf, &path, &slot, &csum_offset, &tree_csum,
2092                         end, &key);
2093         if (ret < 0) {
2094                 fprintf(stderr, "Fetch csum failed\n");
2095                 goto fail_out;
2096         } else if (ret == 1) {
2097                 if (!(*flags & EQUAL_STRIPE))
2098                         *flags |= EQUAL_STRIPE;
2099                 goto out;
2100         } else if (ret == 2)
2101                 goto next_stripe;
2102
2103         list_for_each_entry_safe(devext, next, &candidates, chunk_list) {
2104                 data_offset = calc_data_offset(&key, chunk, devext->offset,
2105                                                csum_offset, blocksize);
2106                 dev = btrfs_find_device_by_devid(rc->fs_devices,
2107                                                  devext->objectid, 0);
2108                 if (!dev) {
2109                         ret = 1;
2110                         goto fail_out;
2111                 }
2112                 BUG_ON(btrfs_find_device_by_devid(rc->fs_devices,
2113                                                   devext->objectid, 1));
2114
2115                 ret = check_one_csum(dev->fd, data_offset, blocksize,
2116                                      tree_csum);
2117                 if (ret < 0)
2118                         goto fail_out;
2119                 else if (ret > 0)
2120                         list_move(&devext->chunk_list, &unordered);
2121         }
2122
2123         if (list_empty(&candidates)) {
2124                 num_unordered = count_devext_records(&unordered);
2125                 if (chunk->type_flags & BTRFS_BLOCK_GROUP_RAID6
2126                                         && num_unordered == 2) {
2127                         btrfs_release_path(&path);
2128                         ret = fill_chunk_up(chunk, &unordered, rc);
2129                         return ret;
2130                 }
2131
2132                 goto next_stripe;
2133         }
2134
2135         if (list_is_last(candidates.next, &candidates)) {
2136                 index = btrfs_calc_stripe_index(chunk,
2137                         key.offset + csum_offset * blocksize);
2138                 BUG_ON(index == -1);
2139                 if (chunk->stripes[index].devid)
2140                         goto next_stripe;
2141                 ret = insert_stripe(&candidates, rc, chunk, index);
2142                 if (ret)
2143                         goto fail_out;
2144         } else {
2145                 csum_offset++;
2146                 goto next_csum;
2147         }
2148 next_stripe:
2149         start = btrfs_next_stripe_logical_offset(chunk, start);
2150         end = min(start + chunk->stripe_len, chunk_end);
2151         list_splice_init(&unordered, &candidates);
2152         btrfs_release_path(&path);
2153         csum_offset = 0;
2154         if (end < chunk_end)
2155                 goto again;
2156 out:
2157         ret = 0;
2158         list_splice_init(&candidates, &unordered);
2159         num_unordered = count_devext_records(&unordered);
2160         if (num_unordered == 1) {
2161                 for (i = 0; i < chunk->num_stripes; i++) {
2162                         if (!chunk->stripes[i].devid) {
2163                                 index = i;
2164                                 break;
2165                         }
2166                 }
2167                 ret = insert_stripe(&unordered, rc, chunk, index);
2168                 if (ret)
2169                         goto fail_out;
2170         } else {
2171                 if ((num_unordered == 2 && chunk->type_flags
2172                         & BTRFS_BLOCK_GROUP_RAID5)
2173                  || (num_unordered == 3 && chunk->type_flags
2174                         & BTRFS_BLOCK_GROUP_RAID6)) {
2175                         ret = fill_chunk_up(chunk, &unordered, rc);
2176                 }
2177         }
2178 fail_out:
2179         ret = !!ret || (list_empty(&unordered) ? 0 : 1);
2180         list_splice_init(&candidates, &chunk->dextents);
2181         list_splice_init(&unordered, &chunk->dextents);
2182         btrfs_release_path(&path);
2183
2184         return ret;
2185 }
2186
2187 static int btrfs_rebuild_ordered_data_chunk_stripes(struct recover_control *rc,
2188                                            struct btrfs_root *root)
2189 {
2190         struct chunk_record *chunk;
2191         struct chunk_record *next;
2192         int ret = 0;
2193         int err;
2194         u8 flags;
2195
2196         list_for_each_entry_safe(chunk, next, &rc->unrepaired_chunks, list) {
2197                 if ((chunk->type_flags & BTRFS_BLOCK_GROUP_DATA)
2198                  && (chunk->type_flags & BTRFS_ORDERED_RAID)) {
2199                         flags = 0;
2200                         err = rebuild_raid_data_chunk_stripes(rc, root, chunk,
2201                                                               &flags);
2202                         if (err) {
2203                                 list_move(&chunk->list, &rc->bad_chunks);
2204                                 if (flags & EQUAL_STRIPE)
2205                                         fprintf(stderr,
2206                         "Failure: too many equal stripes in chunk[%llu %llu]\n",
2207                                                 chunk->offset, chunk->length);
2208                                 if (!ret)
2209                                         ret = err;
2210                         } else
2211                                 list_move(&chunk->list, &rc->good_chunks);
2212                 }
2213         }
2214         return ret;
2215 }
2216
2217 static int btrfs_recover_chunks(struct recover_control *rc)
2218 {
2219         struct chunk_record *chunk;
2220         struct block_group_record *bg;
2221         struct block_group_record *next;
2222         LIST_HEAD(new_chunks);
2223         LIST_HEAD(devexts);
2224         int nstripes;
2225         int ret;
2226
2227         /* create the chunk by block group */
2228         list_for_each_entry_safe(bg, next, &rc->bg.block_groups, list) {
2229                 nstripes = btrfs_get_device_extents(bg->objectid,
2230                                                     &rc->devext.no_chunk_orphans,
2231                                                     &devexts);
2232                 chunk = calloc(1, btrfs_chunk_record_size(nstripes));
2233                 if (!chunk)
2234                         return -ENOMEM;
2235                 INIT_LIST_HEAD(&chunk->dextents);
2236                 chunk->bg_rec = bg;
2237                 chunk->cache.start = bg->objectid;
2238                 chunk->cache.size = bg->offset;
2239                 chunk->objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2240                 chunk->type = BTRFS_CHUNK_ITEM_KEY;
2241                 chunk->offset = bg->objectid;
2242                 chunk->generation = bg->generation;
2243                 chunk->length = bg->offset;
2244                 chunk->owner = BTRFS_CHUNK_TREE_OBJECTID;
2245                 chunk->stripe_len = BTRFS_STRIPE_LEN;
2246                 chunk->type_flags = bg->flags;
2247                 chunk->io_width = BTRFS_STRIPE_LEN;
2248                 chunk->io_align = BTRFS_STRIPE_LEN;
2249                 chunk->sector_size = rc->sectorsize;
2250                 chunk->sub_stripes = calc_sub_nstripes(bg->flags);
2251
2252                 ret = insert_cache_extent(&rc->chunk, &chunk->cache);
2253                 if (ret == -EEXIST) {
2254                         error("duplicate entry in cache start %llu size %llu",
2255                                         (unsigned long long)chunk->cache.start,
2256                                         (unsigned long long)chunk->cache.size);
2257                         free(chunk);
2258                         return ret;
2259                 }
2260                 BUG_ON(ret);
2261
2262                 list_del_init(&bg->list);
2263                 if (!nstripes) {
2264                         list_add_tail(&chunk->list, &rc->bad_chunks);
2265                         continue;
2266                 }
2267
2268                 list_splice_init(&devexts, &chunk->dextents);
2269
2270                 ret = btrfs_verify_device_extents(bg, &devexts, nstripes);
2271                 if (ret) {
2272                         list_add_tail(&chunk->list, &rc->bad_chunks);
2273                         continue;
2274                 }
2275
2276                 chunk->num_stripes = nstripes;
2277                 ret = btrfs_rebuild_chunk_stripes(rc, chunk);
2278                 if (ret > 0)
2279                         list_add_tail(&chunk->list, &rc->unrepaired_chunks);
2280                 else if (ret < 0)
2281                         list_add_tail(&chunk->list, &rc->bad_chunks);
2282                 else
2283                         list_add_tail(&chunk->list, &rc->good_chunks);
2284         }
2285         /*
2286          * Don't worry about the lost orphan device extents, they don't
2287          * have its chunk and block group, they must be the old ones that
2288          * we have dropped.
2289          */
2290         return 0;
2291 }
2292
2293 static inline int is_chunk_overlap(struct chunk_record *chunk1,
2294                                    struct chunk_record *chunk2)
2295 {
2296         if (chunk1->offset >= chunk2->offset + chunk2->length ||
2297             chunk1->offset + chunk1->length <= chunk2->offset)
2298                 return 0;
2299         return 1;
2300 }
2301
2302 /* Move invalid(overlap with good chunks) rebuild chunks to bad chunk list */
2303 static void validate_rebuild_chunks(struct recover_control *rc)
2304 {
2305         struct chunk_record *good;
2306         struct chunk_record *rebuild;
2307         struct chunk_record *tmp;
2308
2309         list_for_each_entry_safe(rebuild, tmp, &rc->rebuild_chunks, list) {
2310                 list_for_each_entry(good, &rc->good_chunks, list) {
2311                         if (is_chunk_overlap(rebuild, good)) {
2312                                 list_move_tail(&rebuild->list,
2313                                                &rc->bad_chunks);
2314                                 break;
2315                         }
2316                 }
2317         }
2318 }
2319
2320 /*
2321  * Return 0 when successful, < 0 on error and > 0 if aborted by user
2322  */
2323 int btrfs_recover_chunk_tree(char *path, int verbose, int yes)
2324 {
2325         int ret = 0;
2326         struct btrfs_root *root = NULL;
2327         struct btrfs_trans_handle *trans;
2328         struct recover_control rc;
2329
2330         init_recover_control(&rc, verbose, yes);
2331
2332         ret = recover_prepare(&rc, path);
2333         if (ret) {
2334                 fprintf(stderr, "recover prepare error\n");
2335                 return ret;
2336         }
2337
2338         ret = scan_devices(&rc);
2339         if (ret) {
2340                 fprintf(stderr, "scan chunk headers error\n");
2341                 goto fail_rc;
2342         }
2343
2344         if (cache_tree_empty(&rc.chunk) &&
2345             cache_tree_empty(&rc.bg.tree) &&
2346             cache_tree_empty(&rc.devext.tree)) {
2347                 fprintf(stderr, "no recoverable chunk\n");
2348                 goto fail_rc;
2349         }
2350
2351         print_scan_result(&rc);
2352
2353         ret = check_chunks(&rc.chunk, &rc.bg, &rc.devext, &rc.good_chunks,
2354                            &rc.bad_chunks, &rc.rebuild_chunks, 1);
2355         if (ret) {
2356                 if (!list_empty(&rc.bg.block_groups) ||
2357                     !list_empty(&rc.devext.no_chunk_orphans)) {
2358                         ret = btrfs_recover_chunks(&rc);
2359                         if (ret)
2360                                 goto fail_rc;
2361                 }
2362         } else {
2363                 print_check_result(&rc);
2364                 printf("Check chunks successfully with no orphans\n");
2365                 goto fail_rc;
2366         }
2367         validate_rebuild_chunks(&rc);
2368         print_check_result(&rc);
2369
2370         root = open_ctree_with_broken_chunk(&rc);
2371         if (IS_ERR(root)) {
2372                 fprintf(stderr, "open with broken chunk error\n");
2373                 ret = PTR_ERR(root);
2374                 goto fail_rc;
2375         }
2376
2377         ret = check_all_chunks_by_metadata(&rc, root);
2378         if (ret) {
2379                 fprintf(stderr, "The chunks in memory can not match the metadata of the fs. Repair failed.\n");
2380                 goto fail_close_ctree;
2381         }
2382
2383         ret = btrfs_rebuild_ordered_data_chunk_stripes(&rc, root);
2384         if (ret) {
2385                 fprintf(stderr, "Failed to rebuild ordered chunk stripes.\n");
2386                 goto fail_close_ctree;
2387         }
2388
2389         if (!rc.yes) {
2390                 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?");
2391                 if (!ret) {
2392                         ret = 1;
2393                         goto fail_close_ctree;
2394                 }
2395         }
2396
2397         trans = btrfs_start_transaction(root, 1);
2398         ret = remove_chunk_extent_item(trans, &rc, root);
2399         BUG_ON(ret);
2400
2401         ret = rebuild_chunk_tree(trans, &rc, root);
2402         BUG_ON(ret);
2403
2404         ret = rebuild_sys_array(&rc, root);
2405         BUG_ON(ret);
2406
2407         ret = rebuild_block_group(trans, &rc, root);
2408         if (ret) {
2409                 printf("Fail to rebuild block groups.\n");
2410                 printf("Recommend to run 'btrfs check --init-extent-tree <dev>' after recovery\n");
2411         }
2412
2413         btrfs_commit_transaction(trans, root);
2414 fail_close_ctree:
2415         close_ctree(root);
2416 fail_rc:
2417         free_recover_control(&rc);
2418         return ret;
2419 }