chunk-recover.c

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