btrfs-image.c

   1 /*
   2  * Copyright (C) 2008 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <pthread.h>
  20 #include <stdio.h>
  21 #include <stdlib.h>
  22 #include <sys/types.h>
  23 #include <sys/stat.h>
  24 #include <fcntl.h>
  25 #include <unistd.h>
  26 #include <dirent.h>
  27 #include <zlib.h>
  28 #include "kerncompat.h"
  29 #include "crc32c.h"
  30 #include "ctree.h"
  31 #include "disk-io.h"
  32 #include "transaction.h"
  33 #include "utils.h"
  34 #include "volumes.h"
  35 #include "extent_io.h"
  36
  37 #define HEADER_MAGIC            0xbd5c25e27295668bULL
  38 #define MAX_PENDING_SIZE        (256 * 1024)
  39 #define BLOCK_SIZE              1024
  40 #define BLOCK_MASK              (BLOCK_SIZE - 1)
  41
  42 #define COMPRESS_NONE           0
  43 #define COMPRESS_ZLIB           1
  44
  45 struct meta_cluster_item {
  46         __le64 bytenr;
  47         __le32 size;
  48 } __attribute__ ((__packed__));
  49
  50 struct meta_cluster_header {
  51         __le64 magic;
  52         __le64 bytenr;
  53         __le32 nritems;
  54         u8 compress;
  55 } __attribute__ ((__packed__));
  56
  57 /* cluster header + index items + buffers */
  58 struct meta_cluster {
  59         struct meta_cluster_header header;
  60         struct meta_cluster_item items[];
  61 } __attribute__ ((__packed__));
  62
  63 #define ITEMS_PER_CLUSTER ((BLOCK_SIZE - sizeof(struct meta_cluster)) / \
  64                            sizeof(struct meta_cluster_item))
  65
  66 struct fs_chunk {
  67         u64 logical;
  68         u64 physical;
  69         u64 bytes;
  70         struct rb_node l;
  71         struct rb_node p;
  72         struct list_head list;
  73 };
  74
  75 struct async_work {
  76         struct list_head list;
  77         struct list_head ordered;
  78         u64 start;
  79         u64 size;
  80         u8 *buffer;
  81         size_t bufsize;
  82         int error;
  83 };
  84
  85 struct metadump_struct {
  86         struct btrfs_root *root;
  87         FILE *out;
  88
  89         struct meta_cluster *cluster;
  90
  91         pthread_t *threads;
  92         size_t num_threads;
  93         pthread_mutex_t mutex;
  94         pthread_cond_t cond;
  95         struct rb_root name_tree;
  96
  97         struct list_head list;
  98         struct list_head ordered;
  99         size_t num_items;
 100         size_t num_ready;
 101
 102         u64 pending_start;
 103         u64 pending_size;
 104
 105         int compress_level;
 106         int done;
 107         int data;
 108         int sanitize_names;
 109
 110         int error;
 111 };
 112
 113 struct name {
 114         struct rb_node n;
 115         char *val;
 116         char *sub;
 117         u32 len;
 118 };
 119
 120 struct mdrestore_struct {
 121         FILE *in;
 122         FILE *out;
 123
 124         pthread_t *threads;
 125         size_t num_threads;
 126         pthread_mutex_t mutex;
 127         pthread_cond_t cond;
 128
 129         struct rb_root chunk_tree;
 130         struct rb_root physical_tree;
 131         struct list_head list;
 132         struct list_head overlapping_chunks;
 133         size_t num_items;
 134         u32 leafsize;
 135         u64 devid;
 136         u64 last_physical_offset;
 137         u8 uuid[BTRFS_UUID_SIZE];
 138         u8 fsid[BTRFS_FSID_SIZE];
 139
 140         int compress_method;
 141         int done;
 142         int error;
 143         int old_restore;
 144         int fixup_offset;
 145         int multi_devices;
 146         int clear_space_cache;
 147         struct btrfs_fs_info *info;
 148 };
 149
 150 static void print_usage(void) __attribute__((noreturn));
 151 static int search_for_chunk_blocks(struct mdrestore_struct *mdres,
 152                                    u64 search, u64 cluster_bytenr);
 153 static struct extent_buffer *alloc_dummy_eb(u64 bytenr, u32 size);
 154
 155 static void csum_block(u8 *buf, size_t len)
 156 {
 157         char result[BTRFS_CRC32_SIZE];
 158         u32 crc = ~(u32)0;
 159         crc = crc32c(crc, buf + BTRFS_CSUM_SIZE, len - BTRFS_CSUM_SIZE);
 160         btrfs_csum_final(crc, result);
 161         memcpy(buf, result, BTRFS_CRC32_SIZE);
 162 }
 163
 164 static int has_name(struct btrfs_key *key)
 165 {
 166         switch (key->type) {
 167         case BTRFS_DIR_ITEM_KEY:
 168         case BTRFS_DIR_INDEX_KEY:
 169         case BTRFS_INODE_REF_KEY:
 170         case BTRFS_INODE_EXTREF_KEY:
 171         case BTRFS_XATTR_ITEM_KEY:
 172                 return 1;
 173         default:
 174                 break;
 175         }
 176
 177         return 0;
 178 }
 179
 180 static char *generate_garbage(u32 name_len)
 181 {
 182         char *buf = malloc(name_len);
 183         int i;
 184
 185         if (!buf)
 186                 return NULL;
 187
 188         for (i = 0; i < name_len; i++) {
 189                 char c = rand() % 94 + 33;
 190
 191                 if (c == '/')
 192                         c++;
 193                 buf[i] = c;
 194         }
 195
 196         return buf;
 197 }
 198
 199 static int name_cmp(struct rb_node *a, struct rb_node *b, int fuzz)
 200 {
 201         struct name *entry = rb_entry(a, struct name, n);
 202         struct name *ins = rb_entry(b, struct name, n);
 203         u32 len;
 204
 205         len = min(ins->len, entry->len);
 206         return memcmp(ins->val, entry->val, len);
 207 }
 208
 209 static int chunk_cmp(struct rb_node *a, struct rb_node *b, int fuzz)
 210 {
 211         struct fs_chunk *entry = rb_entry(a, struct fs_chunk, l);
 212         struct fs_chunk *ins = rb_entry(b, struct fs_chunk, l);
 213
 214         if (fuzz && ins->logical >= entry->logical &&
 215             ins->logical < entry->logical + entry->bytes)
 216                 return 0;
 217
 218         if (ins->logical < entry->logical)
 219                 return -1;
 220         else if (ins->logical > entry->logical)
 221                 return 1;
 222         return 0;
 223 }
 224
 225 static int physical_cmp(struct rb_node *a, struct rb_node *b, int fuzz)
 226 {
 227         struct fs_chunk *entry = rb_entry(a, struct fs_chunk, p);
 228         struct fs_chunk *ins = rb_entry(b, struct fs_chunk, p);
 229
 230         if (fuzz && ins->physical >= entry->physical &&
 231             ins->physical < entry->physical + entry->bytes)
 232                 return 0;
 233
 234         if (fuzz && entry->physical >= ins->physical &&
 235             entry->physical < ins->physical + ins->bytes)
 236                 return 0;
 237
 238         if (ins->physical < entry->physical)
 239                 return -1;
 240         else if (ins->physical > entry->physical)
 241                 return 1;
 242         return 0;
 243 }
 244
 245 static void tree_insert(struct rb_root *root, struct rb_node *ins,
 246                         int (*cmp)(struct rb_node *a, struct rb_node *b,
 247                                    int fuzz))
 248 {
 249         struct rb_node ** p = &root->rb_node;
 250         struct rb_node * parent = NULL;
 251         int dir;
 252
 253         while(*p) {
 254                 parent = *p;
 255
 256                 dir = cmp(*p, ins, 1);
 257                 if (dir < 0)
 258                         p = &(*p)->rb_left;
 259                 else if (dir > 0)
 260                         p = &(*p)->rb_right;
 261                 else
 262                         BUG();
 263         }
 264
 265         rb_link_node(ins, parent, p);
 266         rb_insert_color(ins, root);
 267 }
 268
 269 static struct rb_node *tree_search(struct rb_root *root,
 270                                    struct rb_node *search,
 271                                    int (*cmp)(struct rb_node *a,
 272                                               struct rb_node *b, int fuzz),
 273                                    int fuzz)
 274 {
 275         struct rb_node *n = root->rb_node;
 276         int dir;
 277
 278         while (n) {
 279                 dir = cmp(n, search, fuzz);
 280                 if (dir < 0)
 281                         n = n->rb_left;
 282                 else if (dir > 0)
 283                         n = n->rb_right;
 284                 else
 285                         return n;
 286         }
 287
 288         return NULL;
 289 }
 290
 291 static u64 logical_to_physical(struct mdrestore_struct *mdres, u64 logical, u64 *size)
 292 {
 293         struct fs_chunk *fs_chunk;
 294         struct rb_node *entry;
 295         struct fs_chunk search;
 296         u64 offset;
 297
 298         if (logical == BTRFS_SUPER_INFO_OFFSET)
 299                 return logical;
 300
 301         search.logical = logical;
 302         entry = tree_search(&mdres->chunk_tree, &search.l, chunk_cmp, 1);
 303         if (!entry) {
 304                 if (mdres->in != stdin)
 305                         printf("Couldn't find a chunk, using logical\n");
 306                 return logical;
 307         }
 308         fs_chunk = rb_entry(entry, struct fs_chunk, l);
 309         if (fs_chunk->logical > logical || fs_chunk->logical + fs_chunk->bytes < logical)
 310                 BUG();
 311         offset = search.logical - fs_chunk->logical;
 312
 313         *size = min(*size, fs_chunk->bytes + fs_chunk->logical - logical);
 314         return fs_chunk->physical + offset;
 315 }
 316
 317
 318 static char *find_collision(struct metadump_struct *md, char *name,
 319                             u32 name_len)
 320 {
 321         struct name *val;
 322         struct rb_node *entry;
 323         struct name tmp;
 324         unsigned long checksum;
 325         int found = 0;
 326         int i;
 327
 328         tmp.val = name;
 329         tmp.len = name_len;
 330         entry = tree_search(&md->name_tree, &tmp.n, name_cmp, 0);
 331         if (entry) {
 332                 val = rb_entry(entry, struct name, n);
 333                 free(name);
 334                 return val->sub;
 335         }
 336
 337         val = malloc(sizeof(struct name));
 338         if (!val) {
 339                 fprintf(stderr, "Couldn't sanitize name, enomem\n");
 340                 free(name);
 341                 return NULL;
 342         }
 343
 344         memset(val, 0, sizeof(*val));
 345
 346         val->val = name;
 347         val->len = name_len;
 348         val->sub = malloc(name_len);
 349         if (!val->sub) {
 350                 fprintf(stderr, "Couldn't sanitize name, enomem\n");
 351                 free(val);
 352                 free(name);
 353                 return NULL;
 354         }
 355
 356         checksum = crc32c(~1, val->val, name_len);
 357         memset(val->sub, ' ', name_len);
 358         i = 0;
 359         while (1) {
 360                 if (crc32c(~1, val->sub, name_len) == checksum &&
 361                     memcmp(val->sub, val->val, val->len)) {
 362                         found = 1;
 363                         break;
 364                 }
 365
 366                 if (val->sub[i] == 127) {
 367                         do {
 368                                 i++;
 369                                 if (i >= name_len)
 370                                         break;
 371                         } while (val->sub[i] == 127);
 372
 373                         if (i >= name_len)
 374                                 break;
 375                         val->sub[i]++;
 376                         if (val->sub[i] == '/')
 377                                 val->sub[i]++;
 378                         memset(val->sub, ' ', i);
 379                         i = 0;
 380                         continue;
 381                 } else {
 382                         val->sub[i]++;
 383                         if (val->sub[i] == '/')
 384                                 val->sub[i]++;
 385                 }
 386         }
 387
 388         if (!found) {
 389                 fprintf(stderr, "Couldn't find a collision for '%.*s', "
 390                         "generating normal garbage, it won't match indexes\n",
 391                         val->len, val->val);
 392                 for (i = 0; i < name_len; i++) {
 393                         char c = rand() % 94 + 33;
 394
 395                         if (c == '/')
 396                                 c++;
 397                         val->sub[i] = c;
 398                 }
 399         }
 400
 401         tree_insert(&md->name_tree, &val->n, name_cmp);
 402         return val->sub;
 403 }
 404
 405 static void sanitize_dir_item(struct metadump_struct *md, struct extent_buffer *eb,
 406                               int slot)
 407 {
 408         struct btrfs_dir_item *dir_item;
 409         char *buf;
 410         char *garbage;
 411         unsigned long name_ptr;
 412         u32 total_len;
 413         u32 cur = 0;
 414         u32 this_len;
 415         u32 name_len;
 416         int free_garbage = (md->sanitize_names == 1);
 417
 418         dir_item = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
 419         total_len = btrfs_item_size_nr(eb, slot);
 420         while (cur < total_len) {
 421                 this_len = sizeof(*dir_item) +
 422                         btrfs_dir_name_len(eb, dir_item) +
 423                         btrfs_dir_data_len(eb, dir_item);
 424                 name_ptr = (unsigned long)(dir_item + 1);
 425                 name_len = btrfs_dir_name_len(eb, dir_item);
 426
 427                 if (md->sanitize_names > 1) {
 428                         buf = malloc(name_len);
 429                         if (!buf) {
 430                                 fprintf(stderr, "Couldn't sanitize name, "
 431                                         "enomem\n");
 432                                 return;
 433                         }
 434                         read_extent_buffer(eb, buf, name_ptr, name_len);
 435                         garbage = find_collision(md, buf, name_len);
 436                 } else {
 437                         garbage = generate_garbage(name_len);
 438                 }
 439                 if (!garbage) {
 440                         fprintf(stderr, "Couldn't sanitize name, enomem\n");
 441                         return;
 442                 }
 443                 write_extent_buffer(eb, garbage, name_ptr, name_len);
 444                 cur += this_len;
 445                 dir_item = (struct btrfs_dir_item *)((char *)dir_item +
 446                                                      this_len);
 447                 if (free_garbage)
 448                         free(garbage);
 449         }
 450 }
 451
 452 static void sanitize_inode_ref(struct metadump_struct *md,
 453                                struct extent_buffer *eb, int slot, int ext)
 454 {
 455         struct btrfs_inode_extref *extref;
 456         struct btrfs_inode_ref *ref;
 457         char *garbage, *buf;
 458         unsigned long ptr;
 459         unsigned long name_ptr;
 460         u32 item_size;
 461         u32 cur_offset = 0;
 462         int len;
 463         int free_garbage = (md->sanitize_names == 1);
 464
 465         item_size = btrfs_item_size_nr(eb, slot);
 466         ptr = btrfs_item_ptr_offset(eb, slot);
 467         while (cur_offset < item_size) {
 468                 if (ext) {
 469                         extref = (struct btrfs_inode_extref *)(ptr +
 470                                                                cur_offset);
 471                         name_ptr = (unsigned long)(&extref->name);
 472                         len = btrfs_inode_extref_name_len(eb, extref);
 473                         cur_offset += sizeof(*extref);
 474                 } else {
 475                         ref = (struct btrfs_inode_ref *)(ptr + cur_offset);
 476                         len = btrfs_inode_ref_name_len(eb, ref);
 477                         name_ptr = (unsigned long)(ref + 1);
 478                         cur_offset += sizeof(*ref);
 479                 }
 480                 cur_offset += len;
 481
 482                 if (md->sanitize_names > 1) {
 483                         buf = malloc(len);
 484                         if (!buf) {
 485                                 fprintf(stderr, "Couldn't sanitize name, "
 486                                         "enomem\n");
 487                                 return;
 488                         }
 489                         read_extent_buffer(eb, buf, name_ptr, len);
 490                         garbage = find_collision(md, buf, len);
 491                 } else {
 492                         garbage = generate_garbage(len);
 493                 }
 494
 495                 if (!garbage) {
 496                         fprintf(stderr, "Couldn't sanitize name, enomem\n");
 497                         return;
 498                 }
 499                 write_extent_buffer(eb, garbage, name_ptr, len);
 500                 if (free_garbage)
 501                         free(garbage);
 502         }
 503 }
 504
 505 static void sanitize_xattr(struct metadump_struct *md,
 506                            struct extent_buffer *eb, int slot)
 507 {
 508         struct btrfs_dir_item *dir_item;
 509         unsigned long data_ptr;
 510         u32 data_len;
 511
 512         dir_item = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
 513         data_len = btrfs_dir_data_len(eb, dir_item);
 514
 515         data_ptr = (unsigned long)((char *)(dir_item + 1) +
 516                                    btrfs_dir_name_len(eb, dir_item));
 517         memset_extent_buffer(eb, 0, data_ptr, data_len);
 518 }
 519
 520 static void sanitize_name(struct metadump_struct *md, u8 *dst,
 521                           struct extent_buffer *src, struct btrfs_key *key,
 522                           int slot)
 523 {
 524         struct extent_buffer *eb;
 525
 526         eb = alloc_dummy_eb(src->start, src->len);
 527         if (!eb) {
 528                 fprintf(stderr, "Couldn't sanitize name, no memory\n");
 529                 return;
 530         }
 531
 532         memcpy(eb->data, dst, eb->len);
 533
 534         switch (key->type) {
 535         case BTRFS_DIR_ITEM_KEY:
 536         case BTRFS_DIR_INDEX_KEY:
 537                 sanitize_dir_item(md, eb, slot);
 538                 break;
 539         case BTRFS_INODE_REF_KEY:
 540                 sanitize_inode_ref(md, eb, slot, 0);
 541                 break;
 542         case BTRFS_INODE_EXTREF_KEY:
 543                 sanitize_inode_ref(md, eb, slot, 1);
 544                 break;
 545         case BTRFS_XATTR_ITEM_KEY:
 546                 sanitize_xattr(md, eb, slot);
 547                 break;
 548         default:
 549                 break;
 550         }
 551
 552         memcpy(dst, eb->data, eb->len);
 553         free(eb);
 554 }
 555
 556 /*
 557  * zero inline extents and csum items
 558  */
 559 static void zero_items(struct metadump_struct *md, u8 *dst,
 560                        struct extent_buffer *src)
 561 {
 562         struct btrfs_file_extent_item *fi;
 563         struct btrfs_item *item;
 564         struct btrfs_key key;
 565         u32 nritems = btrfs_header_nritems(src);
 566         size_t size;
 567         unsigned long ptr;
 568         int i, extent_type;
 569
 570         for (i = 0; i < nritems; i++) {
 571                 item = btrfs_item_nr(i);
 572                 btrfs_item_key_to_cpu(src, &key, i);
 573                 if (key.type == BTRFS_CSUM_ITEM_KEY) {
 574                         size = btrfs_item_size_nr(src, i);
 575                         memset(dst + btrfs_leaf_data(src) +
 576                                btrfs_item_offset_nr(src, i), 0, size);
 577                         continue;
 578                 }
 579
 580                 if (md->sanitize_names && has_name(&key)) {
 581                         sanitize_name(md, dst, src, &key, i);
 582                         continue;
 583                 }
 584
 585                 if (key.type != BTRFS_EXTENT_DATA_KEY)
 586                         continue;
 587
 588                 fi = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
 589                 extent_type = btrfs_file_extent_type(src, fi);
 590                 if (extent_type != BTRFS_FILE_EXTENT_INLINE)
 591                         continue;
 592
 593                 ptr = btrfs_file_extent_inline_start(fi);
 594                 size = btrfs_file_extent_inline_item_len(src, item);
 595                 memset(dst + ptr, 0, size);
 596         }
 597 }
 598
 599 /*
 600  * copy buffer and zero useless data in the buffer
 601  */
 602 static void copy_buffer(struct metadump_struct *md, u8 *dst,
 603                         struct extent_buffer *src)
 604 {
 605         int level;
 606         size_t size;
 607         u32 nritems;
 608
 609         memcpy(dst, src->data, src->len);
 610         if (src->start == BTRFS_SUPER_INFO_OFFSET)
 611                 return;
 612
 613         level = btrfs_header_level(src);
 614         nritems = btrfs_header_nritems(src);
 615
 616         if (nritems == 0) {
 617                 size = sizeof(struct btrfs_header);
 618                 memset(dst + size, 0, src->len - size);
 619         } else if (level == 0) {
 620                 size = btrfs_leaf_data(src) +
 621                         btrfs_item_offset_nr(src, nritems - 1) -
 622                         btrfs_item_nr_offset(nritems);
 623                 memset(dst + btrfs_item_nr_offset(nritems), 0, size);
 624                 zero_items(md, dst, src);
 625         } else {
 626                 size = offsetof(struct btrfs_node, ptrs) +
 627                         sizeof(struct btrfs_key_ptr) * nritems;
 628                 memset(dst + size, 0, src->len - size);
 629         }
 630         csum_block(dst, src->len);
 631 }
 632
 633 static void *dump_worker(void *data)
 634 {
 635         struct metadump_struct *md = (struct metadump_struct *)data;
 636         struct async_work *async;
 637         int ret;
 638
 639         while (1) {
 640                 pthread_mutex_lock(&md->mutex);
 641                 while (list_empty(&md->list)) {
 642                         if (md->done) {
 643                                 pthread_mutex_unlock(&md->mutex);
 644                                 goto out;
 645                         }
 646                         pthread_cond_wait(&md->cond, &md->mutex);
 647                 }
 648                 async = list_entry(md->list.next, struct async_work, list);
 649                 list_del_init(&async->list);
 650                 pthread_mutex_unlock(&md->mutex);
 651
 652                 if (md->compress_level > 0) {
 653                         u8 *orig = async->buffer;
 654
 655                         async->bufsize = compressBound(async->size);
 656                         async->buffer = malloc(async->bufsize);
 657                         if (!async->buffer) {
 658                                 fprintf(stderr, "Error allocing buffer\n");
 659                                 pthread_mutex_lock(&md->mutex);
 660                                 if (!md->error)
 661                                         md->error = -ENOMEM;
 662                                 pthread_mutex_unlock(&md->mutex);
 663                                 pthread_exit(NULL);
 664                         }
 665
 666                         ret = compress2(async->buffer,
 667                                          (unsigned long *)&async->bufsize,
 668                                          orig, async->size, md->compress_level);
 669
 670                         if (ret != Z_OK)
 671                                 async->error = 1;
 672
 673                         free(orig);
 674                 }
 675
 676                 pthread_mutex_lock(&md->mutex);
 677                 md->num_ready++;
 678                 pthread_mutex_unlock(&md->mutex);
 679         }
 680 out:
 681         pthread_exit(NULL);
 682 }
 683
 684 static void meta_cluster_init(struct metadump_struct *md, u64 start)
 685 {
 686         struct meta_cluster_header *header;
 687
 688         md->num_items = 0;
 689         md->num_ready = 0;
 690         header = &md->cluster->header;
 691         header->magic = cpu_to_le64(HEADER_MAGIC);
 692         header->bytenr = cpu_to_le64(start);
 693         header->nritems = cpu_to_le32(0);
 694         header->compress = md->compress_level > 0 ?
 695                            COMPRESS_ZLIB : COMPRESS_NONE;
 696 }
 697
 698 static void metadump_destroy(struct metadump_struct *md, int num_threads)
 699 {
 700         int i;
 701         struct rb_node *n;
 702
 703         pthread_mutex_lock(&md->mutex);
 704         md->done = 1;
 705         pthread_cond_broadcast(&md->cond);
 706         pthread_mutex_unlock(&md->mutex);
 707
 708         for (i = 0; i < num_threads; i++)
 709                 pthread_join(md->threads[i], NULL);
 710
 711         pthread_cond_destroy(&md->cond);
 712         pthread_mutex_destroy(&md->mutex);
 713
 714         while ((n = rb_first(&md->name_tree))) {
 715                 struct name *name;
 716
 717                 name = rb_entry(n, struct name, n);
 718                 rb_erase(n, &md->name_tree);
 719                 free(name->val);
 720                 free(name->sub);
 721                 free(name);
 722         }
 723         free(md->threads);
 724         free(md->cluster);
 725 }
 726
 727 static int metadump_init(struct metadump_struct *md, struct btrfs_root *root,
 728                          FILE *out, int num_threads, int compress_level,
 729                          int sanitize_names)
 730 {
 731         int i, ret = 0;
 732
 733         memset(md, 0, sizeof(*md));
 734         pthread_cond_init(&md->cond, NULL);
 735         pthread_mutex_init(&md->mutex, NULL);
 736         INIT_LIST_HEAD(&md->list);
 737         INIT_LIST_HEAD(&md->ordered);
 738         md->root = root;
 739         md->out = out;
 740         md->pending_start = (u64)-1;
 741         md->compress_level = compress_level;
 742         md->cluster = calloc(1, BLOCK_SIZE);
 743         md->sanitize_names = sanitize_names;
 744         if (sanitize_names > 1)
 745                 crc32c_optimization_init();
 746
 747         if (!md->cluster) {
 748                 pthread_cond_destroy(&md->cond);
 749                 pthread_mutex_destroy(&md->mutex);
 750                 return -ENOMEM;
 751         }
 752
 753         meta_cluster_init(md, 0);
 754         if (!num_threads)
 755                 return 0;
 756
 757         md->name_tree.rb_node = NULL;
 758         md->num_threads = num_threads;
 759         md->threads = calloc(num_threads, sizeof(pthread_t));
 760         if (!md->threads) {
 761                 free(md->cluster);
 762                 pthread_cond_destroy(&md->cond);
 763                 pthread_mutex_destroy(&md->mutex);
 764                 return -ENOMEM;
 765         }
 766
 767         for (i = 0; i < num_threads; i++) {
 768                 ret = pthread_create(md->threads + i, NULL, dump_worker, md);
 769                 if (ret)
 770                         break;
 771         }
 772
 773         if (ret)
 774                 metadump_destroy(md, i + 1);
 775
 776         return ret;
 777 }
 778
 779 static int write_zero(FILE *out, size_t size)
 780 {
 781         static char zero[BLOCK_SIZE];
 782         return fwrite(zero, size, 1, out);
 783 }
 784
 785 static int write_buffers(struct metadump_struct *md, u64 *next)
 786 {
 787         struct meta_cluster_header *header = &md->cluster->header;
 788         struct meta_cluster_item *item;
 789         struct async_work *async;
 790         u64 bytenr = 0;
 791         u32 nritems = 0;
 792         int ret;
 793         int err = 0;
 794
 795         if (list_empty(&md->ordered))
 796                 goto out;
 797
 798         /* wait until all buffers are compressed */
 799         while (!err && md->num_items > md->num_ready) {
 800                 struct timespec ts = {
 801                         .tv_sec = 0,
 802                         .tv_nsec = 10000000,
 803                 };
 804                 pthread_mutex_unlock(&md->mutex);
 805                 nanosleep(&ts, NULL);
 806                 pthread_mutex_lock(&md->mutex);
 807                 err = md->error;
 808         }
 809
 810         if (err) {
 811                 fprintf(stderr, "One of the threads errored out %s\n",
 812                                 strerror(err));
 813                 goto out;
 814         }
 815
 816         /* setup and write index block */
 817         list_for_each_entry(async, &md->ordered, ordered) {
 818                 item = md->cluster->items + nritems;
 819                 item->bytenr = cpu_to_le64(async->start);
 820                 item->size = cpu_to_le32(async->bufsize);
 821                 nritems++;
 822         }
 823         header->nritems = cpu_to_le32(nritems);
 824
 825         ret = fwrite(md->cluster, BLOCK_SIZE, 1, md->out);
 826         if (ret != 1) {
 827                 fprintf(stderr, "Error writing out cluster: %d\n", errno);
 828                 return -EIO;
 829         }
 830
 831         /* write buffers */
 832         bytenr += le64_to_cpu(header->bytenr) + BLOCK_SIZE;
 833         while (!list_empty(&md->ordered)) {
 834                 async = list_entry(md->ordered.next, struct async_work,
 835                                    ordered);
 836                 list_del_init(&async->ordered);
 837
 838                 bytenr += async->bufsize;
 839                 if (!err)
 840                         ret = fwrite(async->buffer, async->bufsize, 1,
 841                                      md->out);
 842                 if (ret != 1) {
 843                         err = -EIO;
 844                         ret = 0;
 845                         fprintf(stderr, "Error writing out cluster: %d\n",
 846                                 errno);
 847                 }
 848
 849                 free(async->buffer);
 850                 free(async);
 851         }
 852
 853         /* zero unused space in the last block */
 854         if (!err && bytenr & BLOCK_MASK) {
 855                 size_t size = BLOCK_SIZE - (bytenr & BLOCK_MASK);
 856
 857                 bytenr += size;
 858                 ret = write_zero(md->out, size);
 859                 if (ret != 1) {
 860                         fprintf(stderr, "Error zeroing out buffer: %d\n",
 861                                 errno);
 862                         err = -EIO;
 863                 }
 864         }
 865 out:
 866         *next = bytenr;
 867         return err;
 868 }
 869
 870 static int read_data_extent(struct metadump_struct *md,
 871                             struct async_work *async)
 872 {
 873         struct btrfs_multi_bio *multi = NULL;
 874         struct btrfs_device *device;
 875         u64 bytes_left = async->size;
 876         u64 logical = async->start;
 877         u64 offset = 0;
 878         u64 bytenr;
 879         u64 read_len;
 880         ssize_t done;
 881         int fd;
 882         int ret;
 883
 884         while (bytes_left) {
 885                 read_len = bytes_left;
 886                 ret = btrfs_map_block(&md->root->fs_info->mapping_tree, READ,
 887                                       logical, &read_len, &multi, 0, NULL);
 888                 if (ret) {
 889                         fprintf(stderr, "Couldn't map data block %d\n", ret);
 890                         return ret;
 891                 }
 892
 893                 device = multi->stripes[0].dev;
 894
 895                 if (device->fd == 0) {
 896                         fprintf(stderr,
 897                                 "Device we need to read from is not open\n");
 898                         free(multi);
 899                         return -EIO;
 900                 }
 901                 fd = device->fd;
 902                 bytenr = multi->stripes[0].physical;
 903                 free(multi);
 904
 905                 read_len = min(read_len, bytes_left);
 906                 done = pread64(fd, async->buffer+offset, read_len, bytenr);
 907                 if (done < read_len) {
 908                         if (done < 0)
 909                                 fprintf(stderr, "Error reading extent %d\n",
 910                                         errno);
 911                         else
 912                                 fprintf(stderr, "Short read\n");
 913                         return -EIO;
 914                 }
 915
 916                 bytes_left -= done;
 917                 offset += done;
 918                 logical += done;
 919         }
 920
 921         return 0;
 922 }
 923
 924 static int get_dev_fd(struct btrfs_root *root)
 925 {
 926         struct btrfs_device *dev;
 927
 928         dev = list_first_entry(&root->fs_info->fs_devices->devices,
 929                                struct btrfs_device, dev_list);
 930         return dev->fd;
 931 }
 932
 933 static int flush_pending(struct metadump_struct *md, int done)
 934 {
 935         struct async_work *async = NULL;
 936         struct extent_buffer *eb;
 937         u64 blocksize = md->root->nodesize;
 938         u64 start;
 939         u64 size;
 940         size_t offset;
 941         int ret = 0;
 942
 943         if (md->pending_size) {
 944                 async = calloc(1, sizeof(*async));
 945                 if (!async)
 946                         return -ENOMEM;
 947
 948                 async->start = md->pending_start;
 949                 async->size = md->pending_size;
 950                 async->bufsize = async->size;
 951                 async->buffer = malloc(async->bufsize);
 952                 if (!async->buffer) {
 953                         free(async);
 954                         return -ENOMEM;
 955                 }
 956                 offset = 0;
 957                 start = async->start;
 958                 size = async->size;
 959
 960                 if (md->data) {
 961                         ret = read_data_extent(md, async);
 962                         if (ret) {
 963                                 free(async->buffer);
 964                                 free(async);
 965                                 return ret;
 966                         }
 967                 }
 968
 969                 /*
 970                  * Balance can make the mapping not cover the super block, so
 971                  * just copy directly from one of the devices.
 972                  */
 973                 if (start == BTRFS_SUPER_INFO_OFFSET) {
 974                         int fd = get_dev_fd(md->root);
 975
 976                         ret = pread64(fd, async->buffer, size, start);
 977                         if (ret < size) {
 978                                 free(async->buffer);
 979                                 free(async);
 980                                 fprintf(stderr, "Error reading superblock\n");
 981                                 return -EIO;
 982                         }
 983                         size = 0;
 984                         ret = 0;
 985                 }
 986
 987                 while (!md->data && size > 0) {
 988                         u64 this_read = min(blocksize, size);
 989                         eb = read_tree_block(md->root, start, this_read, 0);
 990                         if (!extent_buffer_uptodate(eb)) {
 991                                 free(async->buffer);
 992                                 free(async);
 993                                 fprintf(stderr,
 994                                         "Error reading metadata block\n");
 995                                 return -EIO;
 996                         }
 997                         copy_buffer(md, async->buffer + offset, eb);
 998                         free_extent_buffer(eb);
 999                         start += this_read;
1000                         offset += this_read;
1001                         size -= this_read;
1002                 }
1003
1004                 md->pending_start = (u64)-1;
1005                 md->pending_size = 0;
1006         } else if (!done) {
1007                 return 0;
1008         }
1009
1010         pthread_mutex_lock(&md->mutex);
1011         if (async) {
1012                 list_add_tail(&async->ordered, &md->ordered);
1013                 md->num_items++;
1014                 if (md->compress_level > 0) {
1015                         list_add_tail(&async->list, &md->list);
1016                         pthread_cond_signal(&md->cond);
1017                 } else {
1018                         md->num_ready++;
1019                 }
1020         }
1021         if (md->num_items >= ITEMS_PER_CLUSTER || done) {
1022                 ret = write_buffers(md, &start);
1023                 if (ret)
1024                         fprintf(stderr, "Error writing buffers %d\n",
1025                                 errno);
1026                 else
1027                         meta_cluster_init(md, start);
1028         }
1029         pthread_mutex_unlock(&md->mutex);
1030         return ret;
1031 }
1032
1033 static int add_extent(u64 start, u64 size, struct metadump_struct *md,
1034                       int data)
1035 {
1036         int ret;
1037         if (md->data != data ||
1038             md->pending_size + size > MAX_PENDING_SIZE ||
1039             md->pending_start + md->pending_size != start) {
1040                 ret = flush_pending(md, 0);
1041                 if (ret)
1042                         return ret;
1043                 md->pending_start = start;
1044         }
1045         readahead_tree_block(md->root, start, size, 0);
1046         md->pending_size += size;
1047         md->data = data;
1048         return 0;
1049 }
1050
1051 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1052 static int is_tree_block(struct btrfs_root *extent_root,
1053                          struct btrfs_path *path, u64 bytenr)
1054 {
1055         struct extent_buffer *leaf;
1056         struct btrfs_key key;
1057         u64 ref_objectid;
1058         int ret;
1059
1060         leaf = path->nodes[0];
1061         while (1) {
1062                 struct btrfs_extent_ref_v0 *ref_item;
1063                 path->slots[0]++;
1064                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1065                         ret = btrfs_next_leaf(extent_root, path);
1066                         if (ret < 0)
1067                                 return ret;
1068                         if (ret > 0)
1069                                 break;
1070                         leaf = path->nodes[0];
1071                 }
1072                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1073                 if (key.objectid != bytenr)
1074                         break;
1075                 if (key.type != BTRFS_EXTENT_REF_V0_KEY)
1076                         continue;
1077                 ref_item = btrfs_item_ptr(leaf, path->slots[0],
1078                                           struct btrfs_extent_ref_v0);
1079                 ref_objectid = btrfs_ref_objectid_v0(leaf, ref_item);
1080                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID)
1081                         return 1;
1082                 break;
1083         }
1084         return 0;
1085 }
1086 #endif
1087
1088 static int copy_tree_blocks(struct btrfs_root *root, struct extent_buffer *eb,
1089                             struct metadump_struct *metadump, int root_tree)
1090 {
1091         struct extent_buffer *tmp;
1092         struct btrfs_root_item *ri;
1093         struct btrfs_key key;
1094         u64 bytenr;
1095         int level;
1096         int nritems = 0;
1097         int i = 0;
1098         int ret;
1099
1100         ret = add_extent(btrfs_header_bytenr(eb), root->leafsize, metadump, 0);
1101         if (ret) {
1102                 fprintf(stderr, "Error adding metadata block\n");
1103                 return ret;
1104         }
1105
1106         if (btrfs_header_level(eb) == 0 && !root_tree)
1107                 return 0;
1108
1109         level = btrfs_header_level(eb);
1110         nritems = btrfs_header_nritems(eb);
1111         for (i = 0; i < nritems; i++) {
1112                 if (level == 0) {
1113                         btrfs_item_key_to_cpu(eb, &key, i);
1114                         if (key.type != BTRFS_ROOT_ITEM_KEY)
1115                                 continue;
1116                         ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
1117                         bytenr = btrfs_disk_root_bytenr(eb, ri);
1118                         tmp = read_tree_block(root, bytenr, root->leafsize, 0);
1119                         if (!extent_buffer_uptodate(tmp)) {
1120                                 fprintf(stderr,
1121                                         "Error reading log root block\n");
1122                                 return -EIO;
1123                         }
1124                         ret = copy_tree_blocks(root, tmp, metadump, 0);
1125                         free_extent_buffer(tmp);
1126                         if (ret)
1127                                 return ret;
1128                 } else {
1129                         bytenr = btrfs_node_blockptr(eb, i);
1130                         tmp = read_tree_block(root, bytenr, root->leafsize, 0);
1131                         if (!extent_buffer_uptodate(tmp)) {
1132                                 fprintf(stderr, "Error reading log block\n");
1133                                 return -EIO;
1134                         }
1135                         ret = copy_tree_blocks(root, tmp, metadump, root_tree);
1136                         free_extent_buffer(tmp);
1137                         if (ret)
1138                                 return ret;
1139                 }
1140         }
1141
1142         return 0;
1143 }
1144
1145 static int copy_log_trees(struct btrfs_root *root,
1146                           struct metadump_struct *metadump,
1147                           struct btrfs_path *path)
1148 {
1149         u64 blocknr = btrfs_super_log_root(root->fs_info->super_copy);
1150
1151         if (blocknr == 0)
1152                 return 0;
1153
1154         if (!root->fs_info->log_root_tree ||
1155             !root->fs_info->log_root_tree->node) {
1156                 fprintf(stderr, "Error copying tree log, it wasn't setup\n");
1157                 return -EIO;
1158         }
1159
1160         return copy_tree_blocks(root, root->fs_info->log_root_tree->node,
1161                                 metadump, 1);
1162 }
1163
1164 static int copy_space_cache(struct btrfs_root *root,
1165                             struct metadump_struct *metadump,
1166                             struct btrfs_path *path)
1167 {
1168         struct extent_buffer *leaf;
1169         struct btrfs_file_extent_item *fi;
1170         struct btrfs_key key;
1171         u64 bytenr, num_bytes;
1172         int ret;
1173
1174         root = root->fs_info->tree_root;
1175
1176         key.objectid = 0;
1177         key.type = BTRFS_EXTENT_DATA_KEY;
1178         key.offset = 0;
1179
1180         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1181         if (ret < 0) {
1182                 fprintf(stderr, "Error searching for free space inode %d\n",
1183                         ret);
1184                 return ret;
1185         }
1186
1187         leaf = path->nodes[0];
1188
1189         while (1) {
1190                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1191                         ret = btrfs_next_leaf(root, path);
1192                         if (ret < 0) {
1193                                 fprintf(stderr, "Error going to next leaf "
1194                                         "%d\n", ret);
1195                                 return ret;
1196                         }
1197                         if (ret > 0)
1198                                 break;
1199                         leaf = path->nodes[0];
1200                 }
1201
1202                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1203                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
1204                         path->slots[0]++;
1205                         continue;
1206                 }
1207
1208                 fi = btrfs_item_ptr(leaf, path->slots[0],
1209                                     struct btrfs_file_extent_item);
1210                 if (btrfs_file_extent_type(leaf, fi) !=
1211                     BTRFS_FILE_EXTENT_REG) {
1212                         path->slots[0]++;
1213                         continue;
1214                 }
1215
1216                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1217                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1218                 ret = add_extent(bytenr, num_bytes, metadump, 1);
1219                 if (ret) {
1220                         fprintf(stderr, "Error adding space cache blocks %d\n",
1221                                 ret);
1222                         btrfs_release_path(path);
1223                         return ret;
1224                 }
1225                 path->slots[0]++;
1226         }
1227
1228         return 0;
1229 }
1230
1231 static int copy_from_extent_tree(struct metadump_struct *metadump,
1232                                  struct btrfs_path *path)
1233 {
1234         struct btrfs_root *extent_root;
1235         struct extent_buffer *leaf;
1236         struct btrfs_extent_item *ei;
1237         struct btrfs_key key;
1238         u64 bytenr;
1239         u64 num_bytes;
1240         int ret;
1241
1242         extent_root = metadump->root->fs_info->extent_root;
1243         bytenr = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
1244         key.objectid = bytenr;
1245         key.type = BTRFS_EXTENT_ITEM_KEY;
1246         key.offset = 0;
1247
1248         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1249         if (ret < 0) {
1250                 fprintf(stderr, "Error searching extent root %d\n", ret);
1251                 return ret;
1252         }
1253         ret = 0;
1254
1255         leaf = path->nodes[0];
1256
1257         while (1) {
1258                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1259                         ret = btrfs_next_leaf(extent_root, path);
1260                         if (ret < 0) {
1261                                 fprintf(stderr, "Error going to next leaf %d"
1262                                         "\n", ret);
1263                                 break;
1264                         }
1265                         if (ret > 0) {
1266                                 ret = 0;
1267                                 break;
1268                         }
1269                         leaf = path->nodes[0];
1270                 }
1271
1272                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1273                 if (key.objectid < bytenr ||
1274                     (key.type != BTRFS_EXTENT_ITEM_KEY &&
1275                      key.type != BTRFS_METADATA_ITEM_KEY)) {
1276                         path->slots[0]++;
1277                         continue;
1278                 }
1279
1280                 bytenr = key.objectid;
1281                 if (key.type == BTRFS_METADATA_ITEM_KEY)
1282                         num_bytes = extent_root->leafsize;
1283                 else
1284                         num_bytes = key.offset;
1285
1286                 if (btrfs_item_size_nr(leaf, path->slots[0]) > sizeof(*ei)) {
1287                         ei = btrfs_item_ptr(leaf, path->slots[0],
1288                                             struct btrfs_extent_item);
1289                         if (btrfs_extent_flags(leaf, ei) &
1290                             BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1291                                 ret = add_extent(bytenr, num_bytes, metadump,
1292                                                  0);
1293                                 if (ret) {
1294                                         fprintf(stderr, "Error adding block "
1295                                                 "%d\n", ret);
1296                                         break;
1297                                 }
1298                         }
1299                 } else {
1300 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1301                         ret = is_tree_block(extent_root, path, bytenr);
1302                         if (ret < 0) {
1303                                 fprintf(stderr, "Error checking tree block "
1304                                         "%d\n", ret);
1305                                 break;
1306                         }
1307
1308                         if (ret) {
1309                                 ret = add_extent(bytenr, num_bytes, metadump,
1310                                                  0);
1311                                 if (ret) {
1312                                         fprintf(stderr, "Error adding block "
1313                                                 "%d\n", ret);
1314                                         break;
1315                                 }
1316                         }
1317                         ret = 0;
1318 #else
1319                         fprintf(stderr, "Either extent tree corruption or "
1320                                 "you haven't built with V0 support\n");
1321                         ret = -EIO;
1322                         break;
1323 #endif
1324                 }
1325                 bytenr += num_bytes;
1326         }
1327
1328         btrfs_release_path(path);
1329
1330         return ret;
1331 }
1332
1333 static int create_metadump(const char *input, FILE *out, int num_threads,
1334                            int compress_level, int sanitize, int walk_trees)
1335 {
1336         struct btrfs_root *root;
1337         struct btrfs_path *path = NULL;
1338         struct metadump_struct metadump;
1339         int ret;
1340         int err = 0;
1341
1342         root = open_ctree(input, 0, 0);
1343         if (!root) {
1344                 fprintf(stderr, "Open ctree failed\n");
1345                 return -EIO;
1346         }
1347
1348         BUG_ON(root->nodesize != root->leafsize);
1349
1350         ret = metadump_init(&metadump, root, out, num_threads,
1351                             compress_level, sanitize);
1352         if (ret) {
1353                 fprintf(stderr, "Error initing metadump %d\n", ret);
1354                 close_ctree(root);
1355                 return ret;
1356         }
1357
1358         ret = add_extent(BTRFS_SUPER_INFO_OFFSET, BTRFS_SUPER_INFO_SIZE,
1359                         &metadump, 0);
1360         if (ret) {
1361                 fprintf(stderr, "Error adding metadata %d\n", ret);
1362                 err = ret;
1363                 goto out;
1364         }
1365
1366         path = btrfs_alloc_path();
1367         if (!path) {
1368                 fprintf(stderr, "Out of memory allocing path\n");
1369                 err = -ENOMEM;
1370                 goto out;
1371         }
1372
1373         if (walk_trees) {
1374                 ret = copy_tree_blocks(root, root->fs_info->chunk_root->node,
1375                                        &metadump, 1);
1376                 if (ret) {
1377                         err = ret;
1378                         goto out;
1379                 }
1380
1381                 ret = copy_tree_blocks(root, root->fs_info->tree_root->node,
1382                                        &metadump, 1);
1383                 if (ret) {
1384                         err = ret;
1385                         goto out;
1386                 }
1387         } else {
1388                 ret = copy_from_extent_tree(&metadump, path);
1389                 if (ret) {
1390                         err = ret;
1391                         goto out;
1392                 }
1393         }
1394
1395         ret = copy_log_trees(root, &metadump, path);
1396         if (ret) {
1397                 err = ret;
1398                 goto out;
1399         }
1400
1401         ret = copy_space_cache(root, &metadump, path);
1402 out:
1403         ret = flush_pending(&metadump, 1);
1404         if (ret) {
1405                 if (!err)
1406                         err = ret;
1407                 fprintf(stderr, "Error flushing pending %d\n", ret);
1408         }
1409
1410         metadump_destroy(&metadump, num_threads);
1411
1412         btrfs_free_path(path);
1413         ret = close_ctree(root);
1414         return err ? err : ret;
1415 }
1416
1417 static void update_super_old(u8 *buffer)
1418 {
1419         struct btrfs_super_block *super = (struct btrfs_super_block *)buffer;
1420         struct btrfs_chunk *chunk;
1421         struct btrfs_disk_key *key;
1422         u32 sectorsize = btrfs_super_sectorsize(super);
1423         u64 flags = btrfs_super_flags(super);
1424
1425         flags |= BTRFS_SUPER_FLAG_METADUMP;
1426         btrfs_set_super_flags(super, flags);
1427
1428         key = (struct btrfs_disk_key *)(super->sys_chunk_array);
1429         chunk = (struct btrfs_chunk *)(super->sys_chunk_array +
1430                                        sizeof(struct btrfs_disk_key));
1431
1432         btrfs_set_disk_key_objectid(key, BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1433         btrfs_set_disk_key_type(key, BTRFS_CHUNK_ITEM_KEY);
1434         btrfs_set_disk_key_offset(key, 0);
1435
1436         btrfs_set_stack_chunk_length(chunk, (u64)-1);
1437         btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
1438         btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
1439         btrfs_set_stack_chunk_type(chunk, BTRFS_BLOCK_GROUP_SYSTEM);
1440         btrfs_set_stack_chunk_io_align(chunk, sectorsize);
1441         btrfs_set_stack_chunk_io_width(chunk, sectorsize);
1442         btrfs_set_stack_chunk_sector_size(chunk, sectorsize);
1443         btrfs_set_stack_chunk_num_stripes(chunk, 1);
1444         btrfs_set_stack_chunk_sub_stripes(chunk, 0);
1445         chunk->stripe.devid = super->dev_item.devid;
1446         btrfs_set_stack_stripe_offset(&chunk->stripe, 0);
1447         memcpy(chunk->stripe.dev_uuid, super->dev_item.uuid, BTRFS_UUID_SIZE);
1448         btrfs_set_super_sys_array_size(super, sizeof(*key) + sizeof(*chunk));
1449         csum_block(buffer, BTRFS_SUPER_INFO_SIZE);
1450 }
1451
1452 static int update_super(struct mdrestore_struct *mdres, u8 *buffer)
1453 {
1454         struct btrfs_super_block *super = (struct btrfs_super_block *)buffer;
1455         struct btrfs_chunk *chunk;
1456         struct btrfs_disk_key *disk_key;
1457         struct btrfs_key key;
1458         u64 flags = btrfs_super_flags(super);
1459         u32 new_array_size = 0;
1460         u32 array_size;
1461         u32 cur = 0;
1462         u8 *ptr, *write_ptr;
1463         int old_num_stripes;
1464
1465         write_ptr = ptr = super->sys_chunk_array;
1466         array_size = btrfs_super_sys_array_size(super);
1467
1468         while (cur < array_size) {
1469                 disk_key = (struct btrfs_disk_key *)ptr;
1470                 btrfs_disk_key_to_cpu(&key, disk_key);
1471
1472                 new_array_size += sizeof(*disk_key);
1473                 memmove(write_ptr, ptr, sizeof(*disk_key));
1474
1475                 write_ptr += sizeof(*disk_key);
1476                 ptr += sizeof(*disk_key);
1477                 cur += sizeof(*disk_key);
1478
1479                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1480                         u64 physical, size = 0;
1481
1482                         chunk = (struct btrfs_chunk *)ptr;
1483                         old_num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1484                         chunk = (struct btrfs_chunk *)write_ptr;
1485
1486                         memmove(write_ptr, ptr, sizeof(*chunk));
1487                         btrfs_set_stack_chunk_num_stripes(chunk, 1);
1488                         btrfs_set_stack_chunk_sub_stripes(chunk, 0);
1489                         btrfs_set_stack_chunk_type(chunk,
1490                                                    BTRFS_BLOCK_GROUP_SYSTEM);
1491                         btrfs_set_stack_stripe_devid(&chunk->stripe,
1492                                                      super->dev_item.devid);
1493                         physical = logical_to_physical(mdres, key.offset,
1494                                                        &size);
1495                         if (size != (u64)-1)
1496                                 btrfs_set_stack_stripe_offset(&chunk->stripe,
1497                                                               physical);
1498                         memcpy(chunk->stripe.dev_uuid, super->dev_item.uuid,
1499                                BTRFS_UUID_SIZE);
1500                         new_array_size += sizeof(*chunk);
1501                 } else {
1502                         fprintf(stderr, "Bogus key in the sys chunk array "
1503                                 "%d\n", key.type);
1504                         return -EIO;
1505                 }
1506                 write_ptr += sizeof(*chunk);
1507                 ptr += btrfs_chunk_item_size(old_num_stripes);
1508                 cur += btrfs_chunk_item_size(old_num_stripes);
1509         }
1510
1511         if (mdres->clear_space_cache)
1512                 btrfs_set_super_cache_generation(super, 0);
1513
1514         flags |= BTRFS_SUPER_FLAG_METADUMP_V2;
1515         btrfs_set_super_flags(super, flags);
1516         btrfs_set_super_sys_array_size(super, new_array_size);
1517         csum_block(buffer, BTRFS_SUPER_INFO_SIZE);
1518
1519         return 0;
1520 }
1521
1522 static struct extent_buffer *alloc_dummy_eb(u64 bytenr, u32 size)
1523 {
1524         struct extent_buffer *eb;
1525
1526         eb = malloc(sizeof(struct extent_buffer) + size);
1527         if (!eb)
1528                 return NULL;
1529         memset(eb, 0, sizeof(struct extent_buffer) + size);
1530
1531         eb->start = bytenr;
1532         eb->len = size;
1533         return eb;
1534 }
1535
1536 static void truncate_item(struct extent_buffer *eb, int slot, u32 new_size)
1537 {
1538         struct btrfs_item *item;
1539         u32 nritems;
1540         u32 old_size;
1541         u32 old_data_start;
1542         u32 size_diff;
1543         u32 data_end;
1544         int i;
1545
1546         old_size = btrfs_item_size_nr(eb, slot);
1547         if (old_size == new_size)
1548                 return;
1549
1550         nritems = btrfs_header_nritems(eb);
1551         data_end = btrfs_item_offset_nr(eb, nritems - 1);
1552
1553         old_data_start = btrfs_item_offset_nr(eb, slot);
1554         size_diff = old_size - new_size;
1555
1556         for (i = slot; i < nritems; i++) {
1557                 u32 ioff;
1558                 item = btrfs_item_nr(i);
1559                 ioff = btrfs_item_offset(eb, item);
1560                 btrfs_set_item_offset(eb, item, ioff + size_diff);
1561         }
1562
1563         memmove_extent_buffer(eb, btrfs_leaf_data(eb) + data_end + size_diff,
1564                               btrfs_leaf_data(eb) + data_end,
1565                               old_data_start + new_size - data_end);
1566         item = btrfs_item_nr(slot);
1567         btrfs_set_item_size(eb, item, new_size);
1568 }
1569
1570 static int fixup_chunk_tree_block(struct mdrestore_struct *mdres,
1571                                   struct async_work *async, u8 *buffer,
1572                                   size_t size)
1573 {
1574         struct extent_buffer *eb;
1575         size_t size_left = size;
1576         u64 bytenr = async->start;
1577         int i;
1578
1579         if (size_left % mdres->leafsize)
1580                 return 0;
1581
1582         eb = alloc_dummy_eb(bytenr, mdres->leafsize);
1583         if (!eb)
1584                 return -ENOMEM;
1585
1586         while (size_left) {
1587                 eb->start = bytenr;
1588                 memcpy(eb->data, buffer, mdres->leafsize);
1589
1590                 if (btrfs_header_bytenr(eb) != bytenr)
1591                         break;
1592                 if (memcmp(mdres->fsid,
1593                            eb->data + offsetof(struct btrfs_header, fsid),
1594                            BTRFS_FSID_SIZE))
1595                         break;
1596
1597                 if (btrfs_header_owner(eb) != BTRFS_CHUNK_TREE_OBJECTID)
1598                         goto next;
1599
1600                 if (btrfs_header_level(eb) != 0)
1601                         goto next;
1602
1603                 for (i = 0; i < btrfs_header_nritems(eb); i++) {
1604                         struct btrfs_chunk chunk;
1605                         struct btrfs_key key;
1606                         u64 type, physical, size = (u64)-1;
1607
1608                         btrfs_item_key_to_cpu(eb, &key, i);
1609                         if (key.type != BTRFS_CHUNK_ITEM_KEY)
1610                                 continue;
1611                         truncate_item(eb, i, sizeof(chunk));
1612                         read_extent_buffer(eb, &chunk,
1613                                            btrfs_item_ptr_offset(eb, i),
1614                                            sizeof(chunk));
1615
1616                         size = 0;
1617                         physical = logical_to_physical(mdres, key.offset,
1618                                                        &size);
1619
1620                         /* Zero out the RAID profile */
1621                         type = btrfs_stack_chunk_type(&chunk);
1622                         type &= (BTRFS_BLOCK_GROUP_DATA |
1623                                  BTRFS_BLOCK_GROUP_SYSTEM |
1624                                  BTRFS_BLOCK_GROUP_METADATA |
1625                                  BTRFS_BLOCK_GROUP_DUP);
1626                         btrfs_set_stack_chunk_type(&chunk, type);
1627
1628                         btrfs_set_stack_chunk_num_stripes(&chunk, 1);
1629                         btrfs_set_stack_chunk_sub_stripes(&chunk, 0);
1630                         btrfs_set_stack_stripe_devid(&chunk.stripe, mdres->devid);
1631                         if (size != (u64)-1)
1632                                 btrfs_set_stack_stripe_offset(&chunk.stripe,
1633                                                               physical);
1634                         memcpy(chunk.stripe.dev_uuid, mdres->uuid,
1635                                BTRFS_UUID_SIZE);
1636                         write_extent_buffer(eb, &chunk,
1637                                             btrfs_item_ptr_offset(eb, i),
1638                                             sizeof(chunk));
1639                 }
1640                 memcpy(buffer, eb->data, eb->len);
1641                 csum_block(buffer, eb->len);
1642 next:
1643                 size_left -= mdres->leafsize;
1644                 buffer += mdres->leafsize;
1645                 bytenr += mdres->leafsize;
1646         }
1647
1648         free(eb);
1649         return 0;
1650 }
1651
1652 static void write_backup_supers(int fd, u8 *buf)
1653 {
1654         struct btrfs_super_block *super = (struct btrfs_super_block *)buf;
1655         struct stat st;
1656         u64 size;
1657         u64 bytenr;
1658         int i;
1659         int ret;
1660
1661         if (fstat(fd, &st)) {
1662                 fprintf(stderr, "Couldn't stat restore point, won't be able "
1663                         "to write backup supers: %d\n", errno);
1664                 return;
1665         }
1666
1667         size = btrfs_device_size(fd, &st);
1668
1669         for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1670                 bytenr = btrfs_sb_offset(i);
1671                 if (bytenr + BTRFS_SUPER_INFO_SIZE > size)
1672                         break;
1673                 btrfs_set_super_bytenr(super, bytenr);
1674                 csum_block(buf, BTRFS_SUPER_INFO_SIZE);
1675                 ret = pwrite64(fd, buf, BTRFS_SUPER_INFO_SIZE, bytenr);
1676                 if (ret < BTRFS_SUPER_INFO_SIZE) {
1677                         if (ret < 0)
1678                                 fprintf(stderr, "Problem writing out backup "
1679                                         "super block %d, err %d\n", i, errno);
1680                         else
1681                                 fprintf(stderr, "Short write writing out "
1682                                         "backup super block\n");
1683                         break;
1684                 }
1685         }
1686 }
1687
1688 static void *restore_worker(void *data)
1689 {
1690         struct mdrestore_struct *mdres = (struct mdrestore_struct *)data;
1691         struct async_work *async;
1692         size_t size;
1693         u8 *buffer;
1694         u8 *outbuf;
1695         int outfd;
1696         int ret;
1697         int compress_size = MAX_PENDING_SIZE * 4;
1698
1699         outfd = fileno(mdres->out);
1700         buffer = malloc(compress_size);
1701         if (!buffer) {
1702                 fprintf(stderr, "Error allocing buffer\n");
1703                 pthread_mutex_lock(&mdres->mutex);
1704                 if (!mdres->error)
1705                         mdres->error = -ENOMEM;
1706                 pthread_mutex_unlock(&mdres->mutex);
1707                 pthread_exit(NULL);
1708         }
1709
1710         while (1) {
1711                 u64 bytenr;
1712                 off_t offset = 0;
1713                 int err = 0;
1714
1715                 pthread_mutex_lock(&mdres->mutex);
1716                 while (!mdres->leafsize || list_empty(&mdres->list)) {
1717                         if (mdres->done) {
1718                                 pthread_mutex_unlock(&mdres->mutex);
1719                                 goto out;
1720                         }
1721                         pthread_cond_wait(&mdres->cond, &mdres->mutex);
1722                 }
1723                 async = list_entry(mdres->list.next, struct async_work, list);
1724                 list_del_init(&async->list);
1725                 pthread_mutex_unlock(&mdres->mutex);
1726
1727                 if (mdres->compress_method == COMPRESS_ZLIB) {
1728                         size = compress_size;
1729                         ret = uncompress(buffer, (unsigned long *)&size,
1730                                          async->buffer, async->bufsize);
1731                         if (ret != Z_OK) {
1732                                 fprintf(stderr, "Error decompressing %d\n",
1733                                         ret);
1734                                 err = -EIO;
1735                         }
1736                         outbuf = buffer;
1737                 } else {
1738                         outbuf = async->buffer;
1739                         size = async->bufsize;
1740                 }
1741
1742                 if (!mdres->multi_devices) {
1743                         if (async->start == BTRFS_SUPER_INFO_OFFSET) {
1744                                 if (mdres->old_restore) {
1745                                         update_super_old(outbuf);
1746                                 } else {
1747                                         ret = update_super(mdres, outbuf);
1748                                         if (ret)
1749                                                 err = ret;
1750                                 }
1751                         } else if (!mdres->old_restore) {
1752                                 ret = fixup_chunk_tree_block(mdres, async, outbuf, size);
1753                                 if (ret)
1754                                         err = ret;
1755                         }
1756                 }
1757
1758                 if (!mdres->fixup_offset) {
1759                         while (size) {
1760                                 u64 chunk_size = size;
1761                                 if (!mdres->multi_devices && !mdres->old_restore)
1762                                         bytenr = logical_to_physical(mdres,
1763                                                                      async->start + offset,
1764                                                                      &chunk_size);
1765                                 else
1766                                         bytenr = async->start + offset;
1767
1768                                 ret = pwrite64(outfd, outbuf+offset, chunk_size,
1769                                                bytenr);
1770                                 if (ret != chunk_size) {
1771                                         if (ret < 0) {
1772                                                 fprintf(stderr, "Error writing to "
1773                                                         "device %d\n", errno);
1774                                                 err = errno;
1775                                                 break;
1776                                         } else {
1777                                                 fprintf(stderr, "Short write\n");
1778                                                 err = -EIO;
1779                                                 break;
1780                                         }
1781                                 }
1782                                 size -= chunk_size;
1783                                 offset += chunk_size;
1784                         }
1785                 } else if (async->start != BTRFS_SUPER_INFO_OFFSET) {
1786                         ret = write_data_to_disk(mdres->info, outbuf, async->start, size, 0);
1787                         if (ret) {
1788                                 printk("Error write data\n");
1789                                 exit(1);
1790                         }
1791                 }
1792
1793
1794                 /* backup super blocks are already there at fixup_offset stage */
1795                 if (!mdres->multi_devices && async->start == BTRFS_SUPER_INFO_OFFSET)
1796                         write_backup_supers(outfd, outbuf);
1797
1798                 pthread_mutex_lock(&mdres->mutex);
1799                 if (err && !mdres->error)
1800                         mdres->error = err;
1801                 mdres->num_items--;
1802                 pthread_mutex_unlock(&mdres->mutex);
1803
1804                 free(async->buffer);
1805                 free(async);
1806         }
1807 out:
1808         free(buffer);
1809         pthread_exit(NULL);
1810 }
1811
1812 static void mdrestore_destroy(struct mdrestore_struct *mdres, int num_threads)
1813 {
1814         struct rb_node *n;
1815         int i;
1816
1817         while ((n = rb_first(&mdres->chunk_tree))) {
1818                 struct fs_chunk *entry;
1819
1820                 entry = rb_entry(n, struct fs_chunk, l);
1821                 rb_erase(n, &mdres->chunk_tree);
1822                 rb_erase(&entry->p, &mdres->physical_tree);
1823                 free(entry);
1824         }
1825         pthread_mutex_lock(&mdres->mutex);
1826         mdres->done = 1;
1827         pthread_cond_broadcast(&mdres->cond);
1828         pthread_mutex_unlock(&mdres->mutex);
1829
1830         for (i = 0; i < num_threads; i++)
1831                 pthread_join(mdres->threads[i], NULL);
1832
1833         pthread_cond_destroy(&mdres->cond);
1834         pthread_mutex_destroy(&mdres->mutex);
1835         free(mdres->threads);
1836 }
1837
1838 static int mdrestore_init(struct mdrestore_struct *mdres,
1839                           FILE *in, FILE *out, int old_restore,
1840                           int num_threads, int fixup_offset,
1841                           struct btrfs_fs_info *info, int multi_devices)
1842 {
1843         int i, ret = 0;
1844
1845         memset(mdres, 0, sizeof(*mdres));
1846         pthread_cond_init(&mdres->cond, NULL);
1847         pthread_mutex_init(&mdres->mutex, NULL);
1848         INIT_LIST_HEAD(&mdres->list);
1849         INIT_LIST_HEAD(&mdres->overlapping_chunks);
1850         mdres->in = in;
1851         mdres->out = out;
1852         mdres->old_restore = old_restore;
1853         mdres->chunk_tree.rb_node = NULL;
1854         mdres->fixup_offset = fixup_offset;
1855         mdres->info = info;
1856         mdres->multi_devices = multi_devices;
1857         mdres->clear_space_cache = 0;
1858         mdres->last_physical_offset = 0;
1859
1860         if (!num_threads)
1861                 return 0;
1862
1863         mdres->num_threads = num_threads;
1864         mdres->threads = calloc(num_threads, sizeof(pthread_t));
1865         if (!mdres->threads)
1866                 return -ENOMEM;
1867         for (i = 0; i < num_threads; i++) {
1868                 ret = pthread_create(mdres->threads + i, NULL, restore_worker,
1869                                      mdres);
1870                 if (ret)
1871                         break;
1872         }
1873         if (ret)
1874                 mdrestore_destroy(mdres, i + 1);
1875         return ret;
1876 }
1877
1878 static int fill_mdres_info(struct mdrestore_struct *mdres,
1879                            struct async_work *async)
1880 {
1881         struct btrfs_super_block *super;
1882         u8 *buffer = NULL;
1883         u8 *outbuf;
1884         int ret;
1885
1886         /* We've already been initialized */
1887         if (mdres->leafsize)
1888                 return 0;
1889
1890         if (mdres->compress_method == COMPRESS_ZLIB) {
1891                 size_t size = MAX_PENDING_SIZE * 2;
1892
1893                 buffer = malloc(MAX_PENDING_SIZE * 2);
1894                 if (!buffer)
1895                         return -ENOMEM;
1896                 ret = uncompress(buffer, (unsigned long *)&size,
1897                                  async->buffer, async->bufsize);
1898                 if (ret != Z_OK) {
1899                         fprintf(stderr, "Error decompressing %d\n", ret);
1900                         free(buffer);
1901                         return -EIO;
1902                 }
1903                 outbuf = buffer;
1904         } else {
1905                 outbuf = async->buffer;
1906         }
1907
1908         super = (struct btrfs_super_block *)outbuf;
1909         mdres->leafsize = btrfs_super_leafsize(super);
1910         memcpy(mdres->fsid, super->fsid, BTRFS_FSID_SIZE);
1911         memcpy(mdres->uuid, super->dev_item.uuid,
1912                        BTRFS_UUID_SIZE);
1913         mdres->devid = le64_to_cpu(super->dev_item.devid);
1914         free(buffer);
1915         return 0;
1916 }
1917
1918 static int add_cluster(struct meta_cluster *cluster,
1919                        struct mdrestore_struct *mdres, u64 *next)
1920 {
1921         struct meta_cluster_item *item;
1922         struct meta_cluster_header *header = &cluster->header;
1923         struct async_work *async;
1924         u64 bytenr;
1925         u32 i, nritems;
1926         int ret;
1927
1928         mdres->compress_method = header->compress;
1929
1930         bytenr = le64_to_cpu(header->bytenr) + BLOCK_SIZE;
1931         nritems = le32_to_cpu(header->nritems);
1932         for (i = 0; i < nritems; i++) {
1933                 item = &cluster->items[i];
1934                 async = calloc(1, sizeof(*async));
1935                 if (!async) {
1936                         fprintf(stderr, "Error allocating async\n");
1937                         return -ENOMEM;
1938                 }
1939                 async->start = le64_to_cpu(item->bytenr);
1940                 async->bufsize = le32_to_cpu(item->size);
1941                 async->buffer = malloc(async->bufsize);
1942                 if (!async->buffer) {
1943                         fprintf(stderr, "Error allocing async buffer\n");
1944                         free(async);
1945                         return -ENOMEM;
1946                 }
1947                 ret = fread(async->buffer, async->bufsize, 1, mdres->in);
1948                 if (ret != 1) {
1949                         fprintf(stderr, "Error reading buffer %d\n", errno);
1950                         free(async->buffer);
1951                         free(async);
1952                         return -EIO;
1953                 }
1954                 bytenr += async->bufsize;
1955
1956                 pthread_mutex_lock(&mdres->mutex);
1957                 if (async->start == BTRFS_SUPER_INFO_OFFSET) {
1958                         ret = fill_mdres_info(mdres, async);
1959                         if (ret) {
1960                                 fprintf(stderr, "Error setting up restore\n");
1961                                 pthread_mutex_unlock(&mdres->mutex);
1962                                 free(async->buffer);
1963                                 free(async);
1964                                 return ret;
1965                         }
1966                 }
1967                 list_add_tail(&async->list, &mdres->list);
1968                 mdres->num_items++;
1969                 pthread_cond_signal(&mdres->cond);
1970                 pthread_mutex_unlock(&mdres->mutex);
1971         }
1972         if (bytenr & BLOCK_MASK) {
1973                 char buffer[BLOCK_MASK];
1974                 size_t size = BLOCK_SIZE - (bytenr & BLOCK_MASK);
1975
1976                 bytenr += size;
1977                 ret = fread(buffer, size, 1, mdres->in);
1978                 if (ret != 1) {
1979                         fprintf(stderr, "Error reading in buffer %d\n", errno);
1980                         return -EIO;
1981                 }
1982         }
1983         *next = bytenr;
1984         return 0;
1985 }
1986
1987 static int wait_for_worker(struct mdrestore_struct *mdres)
1988 {
1989         int ret = 0;
1990
1991         pthread_mutex_lock(&mdres->mutex);
1992         ret = mdres->error;
1993         while (!ret && mdres->num_items > 0) {
1994                 struct timespec ts = {
1995                         .tv_sec = 0,
1996                         .tv_nsec = 10000000,
1997                 };
1998                 pthread_mutex_unlock(&mdres->mutex);
1999                 nanosleep(&ts, NULL);
2000                 pthread_mutex_lock(&mdres->mutex);
2001                 ret = mdres->error;
2002         }
2003         pthread_mutex_unlock(&mdres->mutex);
2004         return ret;
2005 }
2006
2007 static int read_chunk_block(struct mdrestore_struct *mdres, u8 *buffer,
2008                             u64 bytenr, u64 item_bytenr, u32 bufsize,
2009                             u64 cluster_bytenr)
2010 {
2011         struct extent_buffer *eb;
2012         int ret = 0;
2013         int i;
2014
2015         eb = alloc_dummy_eb(bytenr, mdres->leafsize);
2016         if (!eb) {
2017                 ret = -ENOMEM;
2018                 goto out;
2019         }
2020
2021         while (item_bytenr != bytenr) {
2022                 buffer += mdres->leafsize;
2023                 item_bytenr += mdres->leafsize;
2024         }
2025
2026         memcpy(eb->data, buffer, mdres->leafsize);
2027         if (btrfs_header_bytenr(eb) != bytenr) {
2028                 fprintf(stderr, "Eb bytenr doesn't match found bytenr\n");
2029                 ret = -EIO;
2030                 goto out;
2031         }
2032
2033         if (memcmp(mdres->fsid, eb->data + offsetof(struct btrfs_header, fsid),
2034                    BTRFS_FSID_SIZE)) {
2035                 fprintf(stderr, "Fsid doesn't match\n");
2036                 ret = -EIO;
2037                 goto out;
2038         }
2039
2040         if (btrfs_header_owner(eb) != BTRFS_CHUNK_TREE_OBJECTID) {
2041                 fprintf(stderr, "Does not belong to the chunk tree\n");
2042                 ret = -EIO;
2043                 goto out;
2044         }
2045
2046         for (i = 0; i < btrfs_header_nritems(eb); i++) {
2047                 struct btrfs_chunk chunk;
2048                 struct fs_chunk *fs_chunk;
2049                 struct btrfs_key key;
2050
2051                 if (btrfs_header_level(eb)) {
2052                         u64 blockptr = btrfs_node_blockptr(eb, i);
2053
2054                         ret = search_for_chunk_blocks(mdres, blockptr,
2055                                                       cluster_bytenr);
2056                         if (ret)
2057                                 break;
2058                         continue;
2059                 }
2060
2061                 /* Yay a leaf!  We loves leafs! */
2062                 btrfs_item_key_to_cpu(eb, &key, i);
2063                 if (key.type != BTRFS_CHUNK_ITEM_KEY)
2064                         continue;
2065
2066                 fs_chunk = malloc(sizeof(struct fs_chunk));
2067                 if (!fs_chunk) {
2068                         fprintf(stderr, "Erorr allocating chunk\n");
2069                         ret = -ENOMEM;
2070                         break;
2071                 }
2072                 memset(fs_chunk, 0, sizeof(*fs_chunk));
2073                 read_extent_buffer(eb, &chunk, btrfs_item_ptr_offset(eb, i),
2074                                    sizeof(chunk));
2075
2076                 fs_chunk->logical = key.offset;
2077                 fs_chunk->physical = btrfs_stack_stripe_offset(&chunk.stripe);
2078                 fs_chunk->bytes = btrfs_stack_chunk_length(&chunk);
2079                 INIT_LIST_HEAD(&fs_chunk->list);
2080                 if (tree_search(&mdres->physical_tree, &fs_chunk->p,
2081                                 physical_cmp, 1) != NULL)
2082                         list_add(&fs_chunk->list, &mdres->overlapping_chunks);
2083                 else
2084                         tree_insert(&mdres->physical_tree, &fs_chunk->p,
2085                                     physical_cmp);
2086                 if (fs_chunk->physical + fs_chunk->bytes >
2087                     mdres->last_physical_offset)
2088                         mdres->last_physical_offset = fs_chunk->physical +
2089                                 fs_chunk->bytes;
2090                 tree_insert(&mdres->chunk_tree, &fs_chunk->l, chunk_cmp);
2091         }
2092 out:
2093         free(eb);
2094         return ret;
2095 }
2096
2097 /* If you have to ask you aren't worthy */
2098 static int search_for_chunk_blocks(struct mdrestore_struct *mdres,
2099                                    u64 search, u64 cluster_bytenr)
2100 {
2101         struct meta_cluster *cluster;
2102         struct meta_cluster_header *header;
2103         struct meta_cluster_item *item;
2104         u64 current_cluster = cluster_bytenr, bytenr;
2105         u64 item_bytenr;
2106         u32 bufsize, nritems, i;
2107         u32 max_size = MAX_PENDING_SIZE * 2;
2108         u8 *buffer, *tmp = NULL;
2109         int ret = 0;
2110
2111         cluster = malloc(BLOCK_SIZE);
2112         if (!cluster) {
2113                 fprintf(stderr, "Error allocating cluster\n");
2114                 return -ENOMEM;
2115         }
2116
2117         buffer = malloc(max_size);
2118         if (!buffer) {
2119                 fprintf(stderr, "Error allocing buffer\n");
2120                 free(cluster);
2121                 return -ENOMEM;
2122         }
2123
2124         if (mdres->compress_method == COMPRESS_ZLIB) {
2125                 tmp = malloc(max_size);
2126                 if (!tmp) {
2127                         fprintf(stderr, "Error allocing tmp buffer\n");
2128                         free(cluster);
2129                         free(buffer);
2130                         return -ENOMEM;
2131                 }
2132         }
2133
2134         bytenr = current_cluster;
2135         while (1) {
2136                 if (fseek(mdres->in, current_cluster, SEEK_SET)) {
2137                         fprintf(stderr, "Error seeking: %d\n", errno);
2138                         ret = -EIO;
2139                         break;
2140                 }
2141
2142                 ret = fread(cluster, BLOCK_SIZE, 1, mdres->in);
2143                 if (ret == 0) {
2144                         if (cluster_bytenr != 0) {
2145                                 cluster_bytenr = 0;
2146                                 current_cluster = 0;
2147                                 bytenr = 0;
2148                                 continue;
2149                         }
2150                         printf("ok this is where we screwed up?\n");
2151                         ret = -EIO;
2152                         break;
2153                 } else if (ret < 0) {
2154                         fprintf(stderr, "Error reading image\n");
2155                         break;
2156                 }
2157                 ret = 0;
2158
2159                 header = &cluster->header;
2160                 if (le64_to_cpu(header->magic) != HEADER_MAGIC ||
2161                     le64_to_cpu(header->bytenr) != current_cluster) {
2162                         fprintf(stderr, "bad header in metadump image\n");
2163                         ret = -EIO;
2164                         break;
2165                 }
2166
2167                 bytenr += BLOCK_SIZE;
2168                 nritems = le32_to_cpu(header->nritems);
2169                 for (i = 0; i < nritems; i++) {
2170                         size_t size;
2171
2172                         item = &cluster->items[i];
2173                         bufsize = le32_to_cpu(item->size);
2174                         item_bytenr = le64_to_cpu(item->bytenr);
2175
2176                         if (bufsize > max_size) {
2177                                 fprintf(stderr, "item %u size %u too big\n",
2178                                         i, bufsize);
2179                                 ret = -EIO;
2180                                 break;
2181                         }
2182
2183                         if (mdres->compress_method == COMPRESS_ZLIB) {
2184                                 ret = fread(tmp, bufsize, 1, mdres->in);
2185                                 if (ret != 1) {
2186                                         fprintf(stderr, "Error reading: %d\n",
2187                                                 errno);
2188                                         ret = -EIO;
2189                                         break;
2190                                 }
2191
2192                                 size = max_size;
2193                                 ret = uncompress(buffer,
2194                                                  (unsigned long *)&size, tmp,
2195                                                  bufsize);
2196                                 if (ret != Z_OK) {
2197                                         fprintf(stderr, "Error decompressing "
2198                                                 "%d\n", ret);
2199                                         ret = -EIO;
2200                                         break;
2201                                 }
2202                         } else {
2203                                 ret = fread(buffer, bufsize, 1, mdres->in);
2204                                 if (ret != 1) {
2205                                         fprintf(stderr, "Error reading: %d\n",
2206                                                 errno);
2207                                         ret = -EIO;
2208                                         break;
2209                                 }
2210                                 size = bufsize;
2211                         }
2212                         ret = 0;
2213
2214                         if (item_bytenr <= search &&
2215                             item_bytenr + size > search) {
2216                                 ret = read_chunk_block(mdres, buffer, search,
2217                                                        item_bytenr, size,
2218                                                        current_cluster);
2219                                 if (!ret)
2220                                         ret = 1;
2221                                 break;
2222                         }
2223                         bytenr += bufsize;
2224                 }
2225                 if (ret) {
2226                         if (ret > 0)
2227                                 ret = 0;
2228                         break;
2229                 }
2230                 if (bytenr & BLOCK_MASK)
2231                         bytenr += BLOCK_SIZE - (bytenr & BLOCK_MASK);
2232                 current_cluster = bytenr;
2233         }
2234
2235         free(tmp);
2236         free(buffer);
2237         free(cluster);
2238         return ret;
2239 }
2240
2241 static int build_chunk_tree(struct mdrestore_struct *mdres,
2242                             struct meta_cluster *cluster)
2243 {
2244         struct btrfs_super_block *super;
2245         struct meta_cluster_header *header;
2246         struct meta_cluster_item *item = NULL;
2247         u64 chunk_root_bytenr = 0;
2248         u32 i, nritems;
2249         u64 bytenr = 0;
2250         u8 *buffer;
2251         int ret;
2252
2253         /* We can't seek with stdin so don't bother doing this */
2254         if (mdres->in == stdin)
2255                 return 0;
2256
2257         ret = fread(cluster, BLOCK_SIZE, 1, mdres->in);
2258         if (ret <= 0) {
2259                 fprintf(stderr, "Error reading in cluster: %d\n", errno);
2260                 return -EIO;
2261         }
2262         ret = 0;
2263
2264         header = &cluster->header;
2265         if (le64_to_cpu(header->magic) != HEADER_MAGIC ||
2266             le64_to_cpu(header->bytenr) != 0) {
2267                 fprintf(stderr, "bad header in metadump image\n");
2268                 return -EIO;
2269         }
2270
2271         bytenr += BLOCK_SIZE;
2272         mdres->compress_method = header->compress;
2273         nritems = le32_to_cpu(header->nritems);
2274         for (i = 0; i < nritems; i++) {
2275                 item = &cluster->items[i];
2276
2277                 if (le64_to_cpu(item->bytenr) == BTRFS_SUPER_INFO_OFFSET)
2278                         break;
2279                 bytenr += le32_to_cpu(item->size);
2280                 if (fseek(mdres->in, le32_to_cpu(item->size), SEEK_CUR)) {
2281                         fprintf(stderr, "Error seeking: %d\n", errno);
2282                         return -EIO;
2283                 }
2284         }
2285
2286         if (!item || le64_to_cpu(item->bytenr) != BTRFS_SUPER_INFO_OFFSET) {
2287                 fprintf(stderr, "Huh, didn't find the super?\n");
2288                 return -EINVAL;
2289         }
2290
2291         buffer = malloc(le32_to_cpu(item->size));
2292         if (!buffer) {
2293                 fprintf(stderr, "Error allocing buffer\n");
2294                 return -ENOMEM;
2295         }
2296
2297         ret = fread(buffer, le32_to_cpu(item->size), 1, mdres->in);
2298         if (ret != 1) {
2299                 fprintf(stderr, "Error reading buffer: %d\n", errno);
2300                 free(buffer);
2301                 return -EIO;
2302         }
2303
2304         if (mdres->compress_method == COMPRESS_ZLIB) {
2305                 size_t size = MAX_PENDING_SIZE * 2;
2306                 u8 *tmp;
2307
2308                 tmp = malloc(MAX_PENDING_SIZE * 2);
2309                 if (!tmp) {
2310                         free(buffer);
2311                         return -ENOMEM;
2312                 }
2313                 ret = uncompress(tmp, (unsigned long *)&size,
2314                                  buffer, le32_to_cpu(item->size));
2315                 if (ret != Z_OK) {
2316                         fprintf(stderr, "Error decompressing %d\n", ret);
2317                         free(buffer);
2318                         free(tmp);
2319                         return -EIO;
2320                 }
2321                 free(buffer);
2322                 buffer = tmp;
2323         }
2324
2325         pthread_mutex_lock(&mdres->mutex);
2326         super = (struct btrfs_super_block *)buffer;
2327         chunk_root_bytenr = btrfs_super_chunk_root(super);
2328         mdres->leafsize = btrfs_super_leafsize(super);
2329         memcpy(mdres->fsid, super->fsid, BTRFS_FSID_SIZE);
2330         memcpy(mdres->uuid, super->dev_item.uuid,
2331                        BTRFS_UUID_SIZE);
2332         mdres->devid = le64_to_cpu(super->dev_item.devid);
2333         free(buffer);
2334         pthread_mutex_unlock(&mdres->mutex);
2335
2336         return search_for_chunk_blocks(mdres, chunk_root_bytenr, 0);
2337 }
2338
2339 static int range_contains_super(u64 physical, u64 bytes)
2340 {
2341         u64 super_bytenr;
2342         int i;
2343
2344         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
2345                 super_bytenr = btrfs_sb_offset(i);
2346                 if (super_bytenr >= physical &&
2347                     super_bytenr < physical + bytes)
2348                         return 1;
2349         }
2350
2351         return 0;
2352 }
2353
2354 static void remap_overlapping_chunks(struct mdrestore_struct *mdres)
2355 {
2356         struct fs_chunk *fs_chunk;
2357
2358         while (!list_empty(&mdres->overlapping_chunks)) {
2359                 fs_chunk = list_first_entry(&mdres->overlapping_chunks,
2360                                             struct fs_chunk, list);
2361                 list_del_init(&fs_chunk->list);
2362                 if (range_contains_super(fs_chunk->physical,
2363                                          fs_chunk->bytes)) {
2364                         fprintf(stderr, "Remapping a chunk that had a super "
2365                                 "mirror inside of it, clearing space cache "
2366                                 "so we don't end up with corruption\n");
2367                         mdres->clear_space_cache = 1;
2368                 }
2369                 fs_chunk->physical = mdres->last_physical_offset;
2370                 tree_insert(&mdres->physical_tree, &fs_chunk->p, physical_cmp);
2371                 mdres->last_physical_offset += fs_chunk->bytes;
2372         }
2373 }
2374
2375 static int __restore_metadump(const char *input, FILE *out, int old_restore,
2376                               int num_threads, int fixup_offset,
2377                               const char *target, int multi_devices)
2378 {
2379         struct meta_cluster *cluster = NULL;
2380         struct meta_cluster_header *header;
2381         struct mdrestore_struct mdrestore;
2382         struct btrfs_fs_info *info = NULL;
2383         u64 bytenr = 0;
2384         FILE *in = NULL;
2385         int ret = 0;
2386
2387         if (!strcmp(input, "-")) {
2388                 in = stdin;
2389         } else {
2390                 in = fopen(input, "r");
2391                 if (!in) {
2392                         perror("unable to open metadump image");
2393                         return 1;
2394                 }
2395         }
2396
2397         /* NOTE: open with write mode */
2398         if (fixup_offset) {
2399                 BUG_ON(!target);
2400                 info = open_ctree_fs_info(target, 0, 0,
2401                                           OPEN_CTREE_WRITES |
2402                                           OPEN_CTREE_RESTORE |
2403                                           OPEN_CTREE_PARTIAL);
2404                 if (!info) {
2405                         fprintf(stderr, "%s: open ctree failed\n", __func__);
2406                         ret = -EIO;
2407                         goto failed_open;
2408                 }
2409         }
2410
2411         cluster = malloc(BLOCK_SIZE);
2412         if (!cluster) {
2413                 fprintf(stderr, "Error allocating cluster\n");
2414                 ret = -ENOMEM;
2415                 goto failed_info;
2416         }
2417
2418         ret = mdrestore_init(&mdrestore, in, out, old_restore, num_threads,
2419                              fixup_offset, info, multi_devices);
2420         if (ret) {
2421                 fprintf(stderr, "Error initing mdrestore %d\n", ret);
2422                 goto failed_cluster;
2423         }
2424
2425         if (!multi_devices && !old_restore) {
2426                 ret = build_chunk_tree(&mdrestore, cluster);
2427                 if (ret)
2428                         goto out;
2429                 if (!list_empty(&mdrestore.overlapping_chunks))
2430                         remap_overlapping_chunks(&mdrestore);
2431         }
2432
2433         if (in != stdin && fseek(in, 0, SEEK_SET)) {
2434                 fprintf(stderr, "Error seeking %d\n", errno);
2435                 goto out;
2436         }
2437
2438         while (!mdrestore.error) {
2439                 ret = fread(cluster, BLOCK_SIZE, 1, in);
2440                 if (!ret)
2441                         break;
2442
2443                 header = &cluster->header;
2444                 if (le64_to_cpu(header->magic) != HEADER_MAGIC ||
2445                     le64_to_cpu(header->bytenr) != bytenr) {
2446                         fprintf(stderr, "bad header in metadump image\n");
2447                         ret = -EIO;
2448                         break;
2449                 }
2450                 ret = add_cluster(cluster, &mdrestore, &bytenr);
2451                 if (ret) {
2452                         fprintf(stderr, "Error adding cluster\n");
2453                         break;
2454                 }
2455         }
2456         ret = wait_for_worker(&mdrestore);
2457 out:
2458         mdrestore_destroy(&mdrestore, num_threads);
2459 failed_cluster:
2460         free(cluster);
2461 failed_info:
2462         if (fixup_offset && info)
2463                 close_ctree(info->chunk_root);
2464 failed_open:
2465         if (in != stdin)
2466                 fclose(in);
2467         return ret;
2468 }
2469
2470 static int restore_metadump(const char *input, FILE *out, int old_restore,
2471                             int num_threads, int multi_devices)
2472 {
2473         return __restore_metadump(input, out, old_restore, num_threads, 0, NULL,
2474                                   multi_devices);
2475 }
2476
2477 static int fixup_metadump(const char *input, FILE *out, int num_threads,
2478                           const char *target)
2479 {
2480         return __restore_metadump(input, out, 0, num_threads, 1, target, 1);
2481 }
2482
2483 static int update_disk_super_on_device(struct btrfs_fs_info *info,
2484                                        const char *other_dev, u64 cur_devid)
2485 {
2486         struct btrfs_key key;
2487         struct extent_buffer *leaf;
2488         struct btrfs_path path;
2489         struct btrfs_dev_item *dev_item;
2490         struct btrfs_super_block *disk_super;
2491         char dev_uuid[BTRFS_UUID_SIZE];
2492         char fs_uuid[BTRFS_UUID_SIZE];
2493         u64 devid, type, io_align, io_width;
2494         u64 sector_size, total_bytes, bytes_used;
2495         char *buf;
2496         int fp;
2497         int ret;
2498
2499         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2500         key.type = BTRFS_DEV_ITEM_KEY;
2501         key.offset = cur_devid;
2502
2503         btrfs_init_path(&path);
2504         ret = btrfs_search_slot(NULL, info->chunk_root, &key, &path, 0, 0);
2505         if (ret) {
2506                 fprintf(stderr, "search key fails\n");
2507                 exit(1);
2508         }
2509
2510         leaf = path.nodes[0];
2511         dev_item = btrfs_item_ptr(leaf, path.slots[0],
2512                                   struct btrfs_dev_item);
2513
2514         devid = btrfs_device_id(leaf, dev_item);
2515         if (devid != cur_devid) {
2516                 printk("devid %llu mismatch with %llu\n", devid, cur_devid);
2517                 exit(1);
2518         }
2519
2520         type = btrfs_device_type(leaf, dev_item);
2521         io_align = btrfs_device_io_align(leaf, dev_item);
2522         io_width = btrfs_device_io_width(leaf, dev_item);
2523         sector_size = btrfs_device_sector_size(leaf, dev_item);
2524         total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2525         bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2526         read_extent_buffer(leaf, dev_uuid, (unsigned long)btrfs_device_uuid(dev_item), BTRFS_UUID_SIZE);
2527         read_extent_buffer(leaf, fs_uuid, (unsigned long)btrfs_device_fsid(dev_item), BTRFS_UUID_SIZE);
2528
2529         btrfs_release_path(&path);
2530
2531         printk("update disk super on %s devid=%llu\n", other_dev, devid);
2532
2533         /* update other devices' super block */
2534         fp = open(other_dev, O_CREAT | O_RDWR, 0600);
2535         if (fp < 0) {
2536                 fprintf(stderr, "could not open %s\n", other_dev);
2537                 exit(1);
2538         }
2539
2540         buf = malloc(BTRFS_SUPER_INFO_SIZE);
2541         if (!buf) {
2542                 ret = -ENOMEM;
2543                 close(fp);
2544                 return ret;
2545         }
2546
2547         memcpy(buf, info->super_copy, BTRFS_SUPER_INFO_SIZE);
2548
2549         disk_super = (struct btrfs_super_block *)buf;
2550         dev_item = &disk_super->dev_item;
2551
2552         btrfs_set_stack_device_type(dev_item, type);
2553         btrfs_set_stack_device_id(dev_item, devid);
2554         btrfs_set_stack_device_total_bytes(dev_item, total_bytes);
2555         btrfs_set_stack_device_bytes_used(dev_item, bytes_used);
2556         btrfs_set_stack_device_io_align(dev_item, io_align);
2557         btrfs_set_stack_device_io_width(dev_item, io_width);
2558         btrfs_set_stack_device_sector_size(dev_item, sector_size);
2559         memcpy(dev_item->uuid, dev_uuid, BTRFS_UUID_SIZE);
2560         memcpy(dev_item->fsid, fs_uuid, BTRFS_UUID_SIZE);
2561         csum_block((u8 *)buf, BTRFS_SUPER_INFO_SIZE);
2562
2563         ret = pwrite64(fp, buf, BTRFS_SUPER_INFO_SIZE, BTRFS_SUPER_INFO_OFFSET);
2564         if (ret != BTRFS_SUPER_INFO_SIZE) {
2565                 ret = -EIO;
2566                 goto out;
2567         }
2568
2569         write_backup_supers(fp, (u8 *)buf);
2570
2571 out:
2572         free(buf);
2573         close(fp);
2574         return 0;
2575 }
2576
2577 static void print_usage(void)
2578 {
2579         fprintf(stderr, "usage: btrfs-image [options] source target\n");
2580         fprintf(stderr, "\t-r      \trestore metadump image\n");
2581         fprintf(stderr, "\t-c value\tcompression level (0 ~ 9)\n");
2582         fprintf(stderr, "\t-t value\tnumber of threads (1 ~ 32)\n");
2583         fprintf(stderr, "\t-o      \tdon't mess with the chunk tree when restoring\n");
2584         fprintf(stderr, "\t-s      \tsanitize file names, use once to just use garbage, use twice if you want crc collisions\n");
2585         fprintf(stderr, "\t-w      \twalk all trees instead of using extent tree, do this if your extent tree is broken\n");
2586         fprintf(stderr, "\t-m      \trestore for multiple devices\n");
2587         fprintf(stderr, "\n");
2588         fprintf(stderr, "\tIn the dump mode, source is the btrfs device and target is the output file (use '-' for stdout).\n");
2589         fprintf(stderr, "\tIn the restore mode, source is the dumped image and target is the btrfs device/file.\n");
2590         exit(1);
2591 }
2592
2593 int main(int argc, char *argv[])
2594 {
2595         char *source;
2596         char *target;
2597         u64 num_threads = 1;
2598         u64 compress_level = 0;
2599         int create = 1;
2600         int old_restore = 0;
2601         int walk_trees = 0;
2602         int multi_devices = 0;
2603         int ret;
2604         int sanitize = 0;
2605         int dev_cnt = 0;
2606         int usage_error = 0;
2607         FILE *out;
2608
2609         while (1) {
2610                 int c = getopt(argc, argv, "rc:t:oswm");
2611                 if (c < 0)
2612                         break;
2613                 switch (c) {
2614                 case 'r':
2615                         create = 0;
2616                         break;
2617                 case 't':
2618                         num_threads = arg_strtou64(optarg);
2619                         if (num_threads > 32)
2620                                 print_usage();
2621                         break;
2622                 case 'c':
2623                         compress_level = arg_strtou64(optarg);
2624                         if (compress_level > 9)
2625                                 print_usage();
2626                         break;
2627                 case 'o':
2628                         old_restore = 1;
2629                         break;
2630                 case 's':
2631                         sanitize++;
2632                         break;
2633                 case 'w':
2634                         walk_trees = 1;
2635                         break;
2636                 case 'm':
2637                         create = 0;
2638                         multi_devices = 1;
2639                         break;
2640                 default:
2641                         print_usage();
2642                 }
2643         }
2644
2645         argc = argc - optind;
2646         set_argv0(argv);
2647         if (check_argc_min(argc, 2))
2648                 print_usage();
2649
2650         dev_cnt = argc - 1;
2651
2652         if (create) {
2653                 if (old_restore) {
2654                         fprintf(stderr, "Usage error: create and restore cannot be used at the same time\n");
2655                         usage_error++;
2656                 }
2657         } else {
2658                 if (walk_trees || sanitize || compress_level) {
2659                         fprintf(stderr, "Usage error: use -w, -s, -c options for restore makes no sense\n");
2660                         usage_error++;
2661                 }
2662                 if (multi_devices && dev_cnt < 2) {
2663                         fprintf(stderr, "Usage error: not enough devices specified for -m option\n");
2664                         usage_error++;
2665                 }
2666                 if (!multi_devices && dev_cnt != 1) {
2667                         fprintf(stderr, "Usage error: accepts only 1 device without -m option\n");
2668                         usage_error++;
2669                 }
2670         }
2671
2672         if (usage_error)
2673                 print_usage();
2674
2675         source = argv[optind];
2676         target = argv[optind + 1];
2677
2678         if (create && !strcmp(target, "-")) {
2679                 out = stdout;
2680         } else {
2681                 out = fopen(target, "w+");
2682                 if (!out) {
2683                         perror("unable to create target file");
2684                         exit(1);
2685                 }
2686         }
2687
2688         if (num_threads == 1 && compress_level > 0) {
2689                 num_threads = sysconf(_SC_NPROCESSORS_ONLN);
2690                 if (num_threads <= 0)
2691                         num_threads = 1;
2692         }
2693
2694         if (create) {
2695                 ret = check_mounted(source);
2696                 if (ret < 0) {
2697                         fprintf(stderr, "Could not check mount status: %s\n",
2698                                 strerror(-ret));
2699                         exit(1);
2700                 } else if (ret)
2701                         fprintf(stderr,
2702                 "WARNING: The device is mounted. Make sure the filesystem is quiescent.\n");
2703
2704                 ret = create_metadump(source, out, num_threads,
2705                                       compress_level, sanitize, walk_trees);
2706         } else {
2707                 ret = restore_metadump(source, out, old_restore, num_threads,
2708                                        multi_devices);
2709         }
2710         if (ret) {
2711                 printk("%s failed (%s)\n", (create) ? "create" : "restore",
2712                        strerror(errno));
2713                 goto out;
2714         }
2715
2716          /* extended support for multiple devices */
2717         if (!create && multi_devices) {
2718                 struct btrfs_fs_info *info;
2719                 u64 total_devs;
2720                 int i;
2721
2722                 info = open_ctree_fs_info(target, 0, 0,
2723                                           OPEN_CTREE_PARTIAL |
2724                                           OPEN_CTREE_RESTORE);
2725                 if (!info) {
2726                         int e = errno;
2727                         fprintf(stderr, "unable to open %s error = %s\n",
2728                                 target, strerror(e));
2729                         return 1;
2730                 }
2731
2732                 total_devs = btrfs_super_num_devices(info->super_copy);
2733                 if (total_devs != dev_cnt) {
2734                         printk("it needs %llu devices but has only %d\n",
2735                                 total_devs, dev_cnt);
2736                         close_ctree(info->chunk_root);
2737                         goto out;
2738                 }
2739
2740                 /* update super block on other disks */
2741                 for (i = 2; i <= dev_cnt; i++) {
2742                         ret = update_disk_super_on_device(info,
2743                                         argv[optind + i], (u64)i);
2744                         if (ret) {
2745                                 printk("update disk super failed devid=%d (error=%d)\n",
2746                                         i, ret);
2747                                 close_ctree(info->chunk_root);
2748                                 exit(1);
2749                         }
2750                 }
2751
2752                 close_ctree(info->chunk_root);
2753
2754                 /* fix metadata block to map correct chunk */
2755                 ret = fixup_metadump(source, out, 1, target);
2756                 if (ret) {
2757                         fprintf(stderr, "fix metadump failed (error=%d)\n",
2758                                 ret);
2759                         exit(1);
2760                 }
2761         }
2762
2763 out:
2764         if (out == stdout) {
2765                 fflush(out);
2766         } else {
2767                 fclose(out);
2768                 if (ret && create) {
2769                         int unlink_ret;
2770
2771                         unlink_ret = unlink(target);
2772                         if (unlink_ret)
2773                                 fprintf(stderr,
2774                                         "unlink output file failed : %s\n",
2775                                         strerror(errno));
2776                 }
2777         }
2778
2779         return !!ret;
2780 }