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 <getopt.h>
  29
  30 #include "kerncompat.h"
  31 #include "crc32c.h"
  32 #include "ctree.h"
  33 #include "disk-io.h"
  34 #include "transaction.h"
  35 #include "utils.h"
  36 #include "volumes.h"
  37 #include "extent_io.h"
  38
  39 #define HEADER_MAGIC            0xbd5c25e27295668bULL
  40 #define MAX_PENDING_SIZE        (256 * 1024)
  41 #define BLOCK_SIZE              1024
  42 #define BLOCK_MASK              (BLOCK_SIZE - 1)
  43
  44 #define COMPRESS_NONE           0
  45 #define COMPRESS_ZLIB           1
  46
  47 struct meta_cluster_item {
  48         __le64 bytenr;
  49         __le32 size;
  50 } __attribute__ ((__packed__));
  51
  52 struct meta_cluster_header {
  53         __le64 magic;
  54         __le64 bytenr;
  55         __le32 nritems;
  56         u8 compress;
  57 } __attribute__ ((__packed__));
  58
  59 /* cluster header + index items + buffers */
  60 struct meta_cluster {
  61         struct meta_cluster_header header;
  62         struct meta_cluster_item items[];
  63 } __attribute__ ((__packed__));
  64
  65 #define ITEMS_PER_CLUSTER ((BLOCK_SIZE - sizeof(struct meta_cluster)) / \
  66                            sizeof(struct meta_cluster_item))
  67
  68 struct fs_chunk {
  69         u64 logical;
  70         u64 physical;
  71         /*
  72          * physical_dup only store additonal physical for BTRFS_BLOCK_GROUP_DUP
  73          * currently restore only support single and DUP
  74          * TODO: modify this structure and the function related to this
  75          * structure for support RAID*
  76          */
  77         u64 physical_dup;
  78         u64 bytes;
  79         struct rb_node l;
  80         struct rb_node p;
  81         struct list_head list;
  82 };
  83
  84 struct async_work {
  85         struct list_head list;
  86         struct list_head ordered;
  87         u64 start;
  88         u64 size;
  89         u8 *buffer;
  90         size_t bufsize;
  91         int error;
  92 };
  93
  94 struct metadump_struct {
  95         struct btrfs_root *root;
  96         FILE *out;
  97
  98         struct meta_cluster *cluster;
  99
 100         pthread_t *threads;
 101         size_t num_threads;
 102         pthread_mutex_t mutex;
 103         pthread_cond_t cond;
 104         struct rb_root name_tree;
 105
 106         struct list_head list;
 107         struct list_head ordered;
 108         size_t num_items;
 109         size_t num_ready;
 110
 111         u64 pending_start;
 112         u64 pending_size;
 113
 114         int compress_level;
 115         int done;
 116         int data;
 117         int sanitize_names;
 118
 119         int error;
 120 };
 121
 122 struct name {
 123         struct rb_node n;
 124         char *val;
 125         char *sub;
 126         u32 len;
 127 };
 128
 129 struct mdrestore_struct {
 130         FILE *in;
 131         FILE *out;
 132
 133         pthread_t *threads;
 134         size_t num_threads;
 135         pthread_mutex_t mutex;
 136         pthread_cond_t cond;
 137
 138         struct rb_root chunk_tree;
 139         struct rb_root physical_tree;
 140         struct list_head list;
 141         struct list_head overlapping_chunks;
 142         size_t num_items;
 143         u32 nodesize;
 144         u64 devid;
 145         u64 alloced_chunks;
 146         u64 last_physical_offset;
 147         u8 uuid[BTRFS_UUID_SIZE];
 148         u8 fsid[BTRFS_FSID_SIZE];
 149
 150         int compress_method;
 151         int done;
 152         int error;
 153         int old_restore;
 154         int fixup_offset;
 155         int multi_devices;
 156         int clear_space_cache;
 157         struct btrfs_fs_info *info;
 158 };
 159
 160 static int search_for_chunk_blocks(struct mdrestore_struct *mdres,
 161                                    u64 search, u64 cluster_bytenr);
 162 static struct extent_buffer *alloc_dummy_eb(u64 bytenr, u32 size);
 163
 164 static void csum_block(u8 *buf, size_t len)
 165 {
 166         u8 result[BTRFS_CRC32_SIZE];
 167         u32 crc = ~(u32)0;
 168         crc = crc32c(crc, buf + BTRFS_CSUM_SIZE, len - BTRFS_CSUM_SIZE);
 169         btrfs_csum_final(crc, result);
 170         memcpy(buf, result, BTRFS_CRC32_SIZE);
 171 }
 172
 173 static int has_name(struct btrfs_key *key)
 174 {
 175         switch (key->type) {
 176         case BTRFS_DIR_ITEM_KEY:
 177         case BTRFS_DIR_INDEX_KEY:
 178         case BTRFS_INODE_REF_KEY:
 179         case BTRFS_INODE_EXTREF_KEY:
 180         case BTRFS_XATTR_ITEM_KEY:
 181                 return 1;
 182         default:
 183                 break;
 184         }
 185
 186         return 0;
 187 }
 188
 189 static char *generate_garbage(u32 name_len)
 190 {
 191         char *buf = malloc(name_len);
 192         int i;
 193
 194         if (!buf)
 195                 return NULL;
 196
 197         for (i = 0; i < name_len; i++) {
 198                 char c = rand_range(94) + 33;
 199
 200                 if (c == '/')
 201                         c++;
 202                 buf[i] = c;
 203         }
 204
 205         return buf;
 206 }
 207
 208 static int name_cmp(struct rb_node *a, struct rb_node *b, int fuzz)
 209 {
 210         struct name *entry = rb_entry(a, struct name, n);
 211         struct name *ins = rb_entry(b, struct name, n);
 212         u32 len;
 213
 214         len = min(ins->len, entry->len);
 215         return memcmp(ins->val, entry->val, len);
 216 }
 217
 218 static int chunk_cmp(struct rb_node *a, struct rb_node *b, int fuzz)
 219 {
 220         struct fs_chunk *entry = rb_entry(a, struct fs_chunk, l);
 221         struct fs_chunk *ins = rb_entry(b, struct fs_chunk, l);
 222
 223         if (fuzz && ins->logical >= entry->logical &&
 224             ins->logical < entry->logical + entry->bytes)
 225                 return 0;
 226
 227         if (ins->logical < entry->logical)
 228                 return -1;
 229         else if (ins->logical > entry->logical)
 230                 return 1;
 231         return 0;
 232 }
 233
 234 static int physical_cmp(struct rb_node *a, struct rb_node *b, int fuzz)
 235 {
 236         struct fs_chunk *entry = rb_entry(a, struct fs_chunk, p);
 237         struct fs_chunk *ins = rb_entry(b, struct fs_chunk, p);
 238
 239         if (fuzz && ins->physical >= entry->physical &&
 240             ins->physical < entry->physical + entry->bytes)
 241                 return 0;
 242
 243         if (fuzz && entry->physical >= ins->physical &&
 244             entry->physical < ins->physical + ins->bytes)
 245                 return 0;
 246
 247         if (ins->physical < entry->physical)
 248                 return -1;
 249         else if (ins->physical > entry->physical)
 250                 return 1;
 251         return 0;
 252 }
 253
 254 static void tree_insert(struct rb_root *root, struct rb_node *ins,
 255                         int (*cmp)(struct rb_node *a, struct rb_node *b,
 256                                    int fuzz))
 257 {
 258         struct rb_node ** p = &root->rb_node;
 259         struct rb_node * parent = NULL;
 260         int dir;
 261
 262         while(*p) {
 263                 parent = *p;
 264
 265                 dir = cmp(*p, ins, 1);
 266                 if (dir < 0)
 267                         p = &(*p)->rb_left;
 268                 else if (dir > 0)
 269                         p = &(*p)->rb_right;
 270                 else
 271                         BUG();
 272         }
 273
 274         rb_link_node(ins, parent, p);
 275         rb_insert_color(ins, root);
 276 }
 277
 278 static struct rb_node *tree_search(struct rb_root *root,
 279                                    struct rb_node *search,
 280                                    int (*cmp)(struct rb_node *a,
 281                                               struct rb_node *b, int fuzz),
 282                                    int fuzz)
 283 {
 284         struct rb_node *n = root->rb_node;
 285         int dir;
 286
 287         while (n) {
 288                 dir = cmp(n, search, fuzz);
 289                 if (dir < 0)
 290                         n = n->rb_left;
 291                 else if (dir > 0)
 292                         n = n->rb_right;
 293                 else
 294                         return n;
 295         }
 296
 297         return NULL;
 298 }
 299
 300 static u64 logical_to_physical(struct mdrestore_struct *mdres, u64 logical,
 301                                u64 *size, u64 *physical_dup)
 302 {
 303         struct fs_chunk *fs_chunk;
 304         struct rb_node *entry;
 305         struct fs_chunk search;
 306         u64 offset;
 307
 308         if (logical == BTRFS_SUPER_INFO_OFFSET)
 309                 return logical;
 310
 311         search.logical = logical;
 312         entry = tree_search(&mdres->chunk_tree, &search.l, chunk_cmp, 1);
 313         if (!entry) {
 314                 if (mdres->in != stdin)
 315                         warning("cannot find a chunk, using logical");
 316                 return logical;
 317         }
 318         fs_chunk = rb_entry(entry, struct fs_chunk, l);
 319         if (fs_chunk->logical > logical || fs_chunk->logical + fs_chunk->bytes < logical)
 320                 BUG();
 321         offset = search.logical - fs_chunk->logical;
 322
 323         if (physical_dup) {
 324                 /* Only in dup case, physical_dup is not equal to 0 */
 325                 if (fs_chunk->physical_dup)
 326                         *physical_dup = fs_chunk->physical_dup + offset;
 327                 else
 328                         *physical_dup = 0;
 329         }
 330
 331         *size = min(*size, fs_chunk->bytes + fs_chunk->logical - logical);
 332         return fs_chunk->physical + offset;
 333 }
 334
 335
 336 static char *find_collision(struct metadump_struct *md, char *name,
 337                             u32 name_len)
 338 {
 339         struct name *val;
 340         struct rb_node *entry;
 341         struct name tmp;
 342         unsigned long checksum;
 343         int found = 0;
 344         int i;
 345
 346         tmp.val = name;
 347         tmp.len = name_len;
 348         entry = tree_search(&md->name_tree, &tmp.n, name_cmp, 0);
 349         if (entry) {
 350                 val = rb_entry(entry, struct name, n);
 351                 free(name);
 352                 return val->sub;
 353         }
 354
 355         val = malloc(sizeof(struct name));
 356         if (!val) {
 357                 error("cannot sanitize name, not enough memory");
 358                 free(name);
 359                 return NULL;
 360         }
 361
 362         memset(val, 0, sizeof(*val));
 363
 364         val->val = name;
 365         val->len = name_len;
 366         val->sub = malloc(name_len);
 367         if (!val->sub) {
 368                 error("cannot sanitize name, not enough memory");
 369                 free(val);
 370                 free(name);
 371                 return NULL;
 372         }
 373
 374         checksum = crc32c(~1, val->val, name_len);
 375         memset(val->sub, ' ', name_len);
 376         i = 0;
 377         while (1) {
 378                 if (crc32c(~1, val->sub, name_len) == checksum &&
 379                     memcmp(val->sub, val->val, val->len)) {
 380                         found = 1;
 381                         break;
 382                 }
 383
 384                 if (val->sub[i] == 127) {
 385                         do {
 386                                 i++;
 387                                 if (i >= name_len)
 388                                         break;
 389                         } while (val->sub[i] == 127);
 390
 391                         if (i >= name_len)
 392                                 break;
 393                         val->sub[i]++;
 394                         if (val->sub[i] == '/')
 395                                 val->sub[i]++;
 396                         memset(val->sub, ' ', i);
 397                         i = 0;
 398                         continue;
 399                 } else {
 400                         val->sub[i]++;
 401                         if (val->sub[i] == '/')
 402                                 val->sub[i]++;
 403                 }
 404         }
 405
 406         if (!found) {
 407                 warning(
 408 "cannot find a hash collision for '%.*s', generating garbage, it won't match indexes",
 409                         val->len, val->val);
 410                 for (i = 0; i < name_len; i++) {
 411                         char c = rand_range(94) + 33;
 412
 413                         if (c == '/')
 414                                 c++;
 415                         val->sub[i] = c;
 416                 }
 417         }
 418
 419         tree_insert(&md->name_tree, &val->n, name_cmp);
 420         return val->sub;
 421 }
 422
 423 static void sanitize_dir_item(struct metadump_struct *md, struct extent_buffer *eb,
 424                               int slot)
 425 {
 426         struct btrfs_dir_item *dir_item;
 427         char *buf;
 428         char *garbage;
 429         unsigned long name_ptr;
 430         u32 total_len;
 431         u32 cur = 0;
 432         u32 this_len;
 433         u32 name_len;
 434         int free_garbage = (md->sanitize_names == 1);
 435
 436         dir_item = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
 437         total_len = btrfs_item_size_nr(eb, slot);
 438         while (cur < total_len) {
 439                 this_len = sizeof(*dir_item) +
 440                         btrfs_dir_name_len(eb, dir_item) +
 441                         btrfs_dir_data_len(eb, dir_item);
 442                 name_ptr = (unsigned long)(dir_item + 1);
 443                 name_len = btrfs_dir_name_len(eb, dir_item);
 444
 445                 if (md->sanitize_names > 1) {
 446                         buf = malloc(name_len);
 447                         if (!buf) {
 448                                 error("cannot sanitize name, not enough memory");
 449                                 return;
 450                         }
 451                         read_extent_buffer(eb, buf, name_ptr, name_len);
 452                         garbage = find_collision(md, buf, name_len);
 453                 } else {
 454                         garbage = generate_garbage(name_len);
 455                 }
 456                 if (!garbage) {
 457                         error("cannot sanitize name, not enough memory");
 458                         return;
 459                 }
 460                 write_extent_buffer(eb, garbage, name_ptr, name_len);
 461                 cur += this_len;
 462                 dir_item = (struct btrfs_dir_item *)((char *)dir_item +
 463                                                      this_len);
 464                 if (free_garbage)
 465                         free(garbage);
 466         }
 467 }
 468
 469 static void sanitize_inode_ref(struct metadump_struct *md,
 470                                struct extent_buffer *eb, int slot, int ext)
 471 {
 472         struct btrfs_inode_extref *extref;
 473         struct btrfs_inode_ref *ref;
 474         char *garbage, *buf;
 475         unsigned long ptr;
 476         unsigned long name_ptr;
 477         u32 item_size;
 478         u32 cur_offset = 0;
 479         int len;
 480         int free_garbage = (md->sanitize_names == 1);
 481
 482         item_size = btrfs_item_size_nr(eb, slot);
 483         ptr = btrfs_item_ptr_offset(eb, slot);
 484         while (cur_offset < item_size) {
 485                 if (ext) {
 486                         extref = (struct btrfs_inode_extref *)(ptr +
 487                                                                cur_offset);
 488                         name_ptr = (unsigned long)(&extref->name);
 489                         len = btrfs_inode_extref_name_len(eb, extref);
 490                         cur_offset += sizeof(*extref);
 491                 } else {
 492                         ref = (struct btrfs_inode_ref *)(ptr + cur_offset);
 493                         len = btrfs_inode_ref_name_len(eb, ref);
 494                         name_ptr = (unsigned long)(ref + 1);
 495                         cur_offset += sizeof(*ref);
 496                 }
 497                 cur_offset += len;
 498
 499                 if (md->sanitize_names > 1) {
 500                         buf = malloc(len);
 501                         if (!buf) {
 502                                 error("cannot sanitize name, not enough memory");
 503                                 return;
 504                         }
 505                         read_extent_buffer(eb, buf, name_ptr, len);
 506                         garbage = find_collision(md, buf, len);
 507                 } else {
 508                         garbage = generate_garbage(len);
 509                 }
 510
 511                 if (!garbage) {
 512                         error("cannot sanitize name, not enough memory");
 513                         return;
 514                 }
 515                 write_extent_buffer(eb, garbage, name_ptr, len);
 516                 if (free_garbage)
 517                         free(garbage);
 518         }
 519 }
 520
 521 static void sanitize_xattr(struct metadump_struct *md,
 522                            struct extent_buffer *eb, int slot)
 523 {
 524         struct btrfs_dir_item *dir_item;
 525         unsigned long data_ptr;
 526         u32 data_len;
 527
 528         dir_item = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
 529         data_len = btrfs_dir_data_len(eb, dir_item);
 530
 531         data_ptr = (unsigned long)((char *)(dir_item + 1) +
 532                                    btrfs_dir_name_len(eb, dir_item));
 533         memset_extent_buffer(eb, 0, data_ptr, data_len);
 534 }
 535
 536 static void sanitize_name(struct metadump_struct *md, u8 *dst,
 537                           struct extent_buffer *src, struct btrfs_key *key,
 538                           int slot)
 539 {
 540         struct extent_buffer *eb;
 541
 542         eb = alloc_dummy_eb(src->start, src->len);
 543         if (!eb) {
 544                 error("cannot sanitize name, not enough memory");
 545                 return;
 546         }
 547
 548         memcpy(eb->data, dst, eb->len);
 549
 550         switch (key->type) {
 551         case BTRFS_DIR_ITEM_KEY:
 552         case BTRFS_DIR_INDEX_KEY:
 553                 sanitize_dir_item(md, eb, slot);
 554                 break;
 555         case BTRFS_INODE_REF_KEY:
 556                 sanitize_inode_ref(md, eb, slot, 0);
 557                 break;
 558         case BTRFS_INODE_EXTREF_KEY:
 559                 sanitize_inode_ref(md, eb, slot, 1);
 560                 break;
 561         case BTRFS_XATTR_ITEM_KEY:
 562                 sanitize_xattr(md, eb, slot);
 563                 break;
 564         default:
 565                 break;
 566         }
 567
 568         memcpy(dst, eb->data, eb->len);
 569         free(eb);
 570 }
 571
 572 /*
 573  * zero inline extents and csum items
 574  */
 575 static void zero_items(struct metadump_struct *md, u8 *dst,
 576                        struct extent_buffer *src)
 577 {
 578         struct btrfs_file_extent_item *fi;
 579         struct btrfs_item *item;
 580         struct btrfs_key key;
 581         u32 nritems = btrfs_header_nritems(src);
 582         size_t size;
 583         unsigned long ptr;
 584         int i, extent_type;
 585
 586         for (i = 0; i < nritems; i++) {
 587                 item = btrfs_item_nr(i);
 588                 btrfs_item_key_to_cpu(src, &key, i);
 589                 if (key.type == BTRFS_CSUM_ITEM_KEY) {
 590                         size = btrfs_item_size_nr(src, i);
 591                         memset(dst + btrfs_leaf_data(src) +
 592                                btrfs_item_offset_nr(src, i), 0, size);
 593                         continue;
 594                 }
 595
 596                 if (md->sanitize_names && has_name(&key)) {
 597                         sanitize_name(md, dst, src, &key, i);
 598                         continue;
 599                 }
 600
 601                 if (key.type != BTRFS_EXTENT_DATA_KEY)
 602                         continue;
 603
 604                 fi = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
 605                 extent_type = btrfs_file_extent_type(src, fi);
 606                 if (extent_type != BTRFS_FILE_EXTENT_INLINE)
 607                         continue;
 608
 609                 ptr = btrfs_file_extent_inline_start(fi);
 610                 size = btrfs_file_extent_inline_item_len(src, item);
 611                 memset(dst + ptr, 0, size);
 612         }
 613 }
 614
 615 /*
 616  * copy buffer and zero useless data in the buffer
 617  */
 618 static void copy_buffer(struct metadump_struct *md, u8 *dst,
 619                         struct extent_buffer *src)
 620 {
 621         int level;
 622         size_t size;
 623         u32 nritems;
 624
 625         memcpy(dst, src->data, src->len);
 626         if (src->start == BTRFS_SUPER_INFO_OFFSET)
 627                 return;
 628
 629         level = btrfs_header_level(src);
 630         nritems = btrfs_header_nritems(src);
 631
 632         if (nritems == 0) {
 633                 size = sizeof(struct btrfs_header);
 634                 memset(dst + size, 0, src->len - size);
 635         } else if (level == 0) {
 636                 size = btrfs_leaf_data(src) +
 637                         btrfs_item_offset_nr(src, nritems - 1) -
 638                         btrfs_item_nr_offset(nritems);
 639                 memset(dst + btrfs_item_nr_offset(nritems), 0, size);
 640                 zero_items(md, dst, src);
 641         } else {
 642                 size = offsetof(struct btrfs_node, ptrs) +
 643                         sizeof(struct btrfs_key_ptr) * nritems;
 644                 memset(dst + size, 0, src->len - size);
 645         }
 646         csum_block(dst, src->len);
 647 }
 648
 649 static void *dump_worker(void *data)
 650 {
 651         struct metadump_struct *md = (struct metadump_struct *)data;
 652         struct async_work *async;
 653         int ret;
 654
 655         while (1) {
 656                 pthread_mutex_lock(&md->mutex);
 657                 while (list_empty(&md->list)) {
 658                         if (md->done) {
 659                                 pthread_mutex_unlock(&md->mutex);
 660                                 goto out;
 661                         }
 662                         pthread_cond_wait(&md->cond, &md->mutex);
 663                 }
 664                 async = list_entry(md->list.next, struct async_work, list);
 665                 list_del_init(&async->list);
 666                 pthread_mutex_unlock(&md->mutex);
 667
 668                 if (md->compress_level > 0) {
 669                         u8 *orig = async->buffer;
 670
 671                         async->bufsize = compressBound(async->size);
 672                         async->buffer = malloc(async->bufsize);
 673                         if (!async->buffer) {
 674                                 error("not enough memory for async buffer");
 675                                 pthread_mutex_lock(&md->mutex);
 676                                 if (!md->error)
 677                                         md->error = -ENOMEM;
 678                                 pthread_mutex_unlock(&md->mutex);
 679                                 pthread_exit(NULL);
 680                         }
 681
 682                         ret = compress2(async->buffer,
 683                                          (unsigned long *)&async->bufsize,
 684                                          orig, async->size, md->compress_level);
 685
 686                         if (ret != Z_OK)
 687                                 async->error = 1;
 688
 689                         free(orig);
 690                 }
 691
 692                 pthread_mutex_lock(&md->mutex);
 693                 md->num_ready++;
 694                 pthread_mutex_unlock(&md->mutex);
 695         }
 696 out:
 697         pthread_exit(NULL);
 698 }
 699
 700 static void meta_cluster_init(struct metadump_struct *md, u64 start)
 701 {
 702         struct meta_cluster_header *header;
 703
 704         md->num_items = 0;
 705         md->num_ready = 0;
 706         header = &md->cluster->header;
 707         header->magic = cpu_to_le64(HEADER_MAGIC);
 708         header->bytenr = cpu_to_le64(start);
 709         header->nritems = cpu_to_le32(0);
 710         header->compress = md->compress_level > 0 ?
 711                            COMPRESS_ZLIB : COMPRESS_NONE;
 712 }
 713
 714 static void metadump_destroy(struct metadump_struct *md, int num_threads)
 715 {
 716         int i;
 717         struct rb_node *n;
 718
 719         pthread_mutex_lock(&md->mutex);
 720         md->done = 1;
 721         pthread_cond_broadcast(&md->cond);
 722         pthread_mutex_unlock(&md->mutex);
 723
 724         for (i = 0; i < num_threads; i++)
 725                 pthread_join(md->threads[i], NULL);
 726
 727         pthread_cond_destroy(&md->cond);
 728         pthread_mutex_destroy(&md->mutex);
 729
 730         while ((n = rb_first(&md->name_tree))) {
 731                 struct name *name;
 732
 733                 name = rb_entry(n, struct name, n);
 734                 rb_erase(n, &md->name_tree);
 735                 free(name->val);
 736                 free(name->sub);
 737                 free(name);
 738         }
 739         free(md->threads);
 740         free(md->cluster);
 741 }
 742
 743 static int metadump_init(struct metadump_struct *md, struct btrfs_root *root,
 744                          FILE *out, int num_threads, int compress_level,
 745                          int sanitize_names)
 746 {
 747         int i, ret = 0;
 748
 749         memset(md, 0, sizeof(*md));
 750         md->cluster = calloc(1, BLOCK_SIZE);
 751         if (!md->cluster)
 752                 return -ENOMEM;
 753         md->threads = calloc(num_threads, sizeof(pthread_t));
 754         if (!md->threads) {
 755                 free(md->cluster);
 756                 return -ENOMEM;
 757         }
 758         INIT_LIST_HEAD(&md->list);
 759         INIT_LIST_HEAD(&md->ordered);
 760         md->root = root;
 761         md->out = out;
 762         md->pending_start = (u64)-1;
 763         md->compress_level = compress_level;
 764         md->sanitize_names = sanitize_names;
 765         if (sanitize_names > 1)
 766                 crc32c_optimization_init();
 767
 768         md->name_tree.rb_node = NULL;
 769         md->num_threads = num_threads;
 770         pthread_cond_init(&md->cond, NULL);
 771         pthread_mutex_init(&md->mutex, NULL);
 772         meta_cluster_init(md, 0);
 773
 774         if (!num_threads)
 775                 return 0;
 776
 777         for (i = 0; i < num_threads; i++) {
 778                 ret = pthread_create(md->threads + i, NULL, dump_worker, md);
 779                 if (ret)
 780                         break;
 781         }
 782
 783         if (ret)
 784                 metadump_destroy(md, i + 1);
 785
 786         return ret;
 787 }
 788
 789 static int write_zero(FILE *out, size_t size)
 790 {
 791         static char zero[BLOCK_SIZE];
 792         return fwrite(zero, size, 1, out);
 793 }
 794
 795 static int write_buffers(struct metadump_struct *md, u64 *next)
 796 {
 797         struct meta_cluster_header *header = &md->cluster->header;
 798         struct meta_cluster_item *item;
 799         struct async_work *async;
 800         u64 bytenr = 0;
 801         u32 nritems = 0;
 802         int ret;
 803         int err = 0;
 804
 805         if (list_empty(&md->ordered))
 806                 goto out;
 807
 808         /* wait until all buffers are compressed */
 809         while (!err && md->num_items > md->num_ready) {
 810                 struct timespec ts = {
 811                         .tv_sec = 0,
 812                         .tv_nsec = 10000000,
 813                 };
 814                 pthread_mutex_unlock(&md->mutex);
 815                 nanosleep(&ts, NULL);
 816                 pthread_mutex_lock(&md->mutex);
 817                 err = md->error;
 818         }
 819
 820         if (err) {
 821                 error("one of the threads failed: %s", strerror(-err));
 822                 goto out;
 823         }
 824
 825         /* setup and write index block */
 826         list_for_each_entry(async, &md->ordered, ordered) {
 827                 item = md->cluster->items + nritems;
 828                 item->bytenr = cpu_to_le64(async->start);
 829                 item->size = cpu_to_le32(async->bufsize);
 830                 nritems++;
 831         }
 832         header->nritems = cpu_to_le32(nritems);
 833
 834         ret = fwrite(md->cluster, BLOCK_SIZE, 1, md->out);
 835         if (ret != 1) {
 836                 error("unable to write out cluster: %s", strerror(errno));
 837                 return -errno;
 838         }
 839
 840         /* write buffers */
 841         bytenr += le64_to_cpu(header->bytenr) + BLOCK_SIZE;
 842         while (!list_empty(&md->ordered)) {
 843                 async = list_entry(md->ordered.next, struct async_work,
 844                                    ordered);
 845                 list_del_init(&async->ordered);
 846
 847                 bytenr += async->bufsize;
 848                 if (!err)
 849                         ret = fwrite(async->buffer, async->bufsize, 1,
 850                                      md->out);
 851                 if (ret != 1) {
 852                         error("unable to write out cluster: %s",
 853                                 strerror(errno));
 854                         err = -errno;
 855                         ret = 0;
 856                 }
 857
 858                 free(async->buffer);
 859                 free(async);
 860         }
 861
 862         /* zero unused space in the last block */
 863         if (!err && bytenr & BLOCK_MASK) {
 864                 size_t size = BLOCK_SIZE - (bytenr & BLOCK_MASK);
 865
 866                 bytenr += size;
 867                 ret = write_zero(md->out, size);
 868                 if (ret != 1) {
 869                         error("unable to zero out buffer: %s",
 870                                 strerror(errno));
 871                         err = -errno;
 872                 }
 873         }
 874 out:
 875         *next = bytenr;
 876         return err;
 877 }
 878
 879 static int read_data_extent(struct metadump_struct *md,
 880                             struct async_work *async)
 881 {
 882         struct btrfs_root *root = md->root;
 883         u64 bytes_left = async->size;
 884         u64 logical = async->start;
 885         u64 offset = 0;
 886         u64 read_len;
 887         int num_copies;
 888         int cur_mirror;
 889         int ret;
 890
 891         num_copies = btrfs_num_copies(&root->fs_info->mapping_tree, logical,
 892                                       bytes_left);
 893
 894         /* Try our best to read data, just like read_tree_block() */
 895         for (cur_mirror = 0; cur_mirror < num_copies; cur_mirror++) {
 896                 while (bytes_left) {
 897                         read_len = bytes_left;
 898                         ret = read_extent_data(root,
 899                                         (char *)(async->buffer + offset),
 900                                         logical, &read_len, cur_mirror);
 901                         if (ret < 0)
 902                                 break;
 903                         offset += read_len;
 904                         logical += read_len;
 905                         bytes_left -= read_len;
 906                 }
 907         }
 908         if (bytes_left)
 909                 return -EIO;
 910         return 0;
 911 }
 912
 913 static int get_dev_fd(struct btrfs_root *root)
 914 {
 915         struct btrfs_device *dev;
 916
 917         dev = list_first_entry(&root->fs_info->fs_devices->devices,
 918                                struct btrfs_device, dev_list);
 919         return dev->fd;
 920 }
 921
 922 static int flush_pending(struct metadump_struct *md, int done)
 923 {
 924         struct async_work *async = NULL;
 925         struct extent_buffer *eb;
 926         u64 blocksize = md->root->nodesize;
 927         u64 start;
 928         u64 size;
 929         size_t offset;
 930         int ret = 0;
 931
 932         if (md->pending_size) {
 933                 async = calloc(1, sizeof(*async));
 934                 if (!async)
 935                         return -ENOMEM;
 936
 937                 async->start = md->pending_start;
 938                 async->size = md->pending_size;
 939                 async->bufsize = async->size;
 940                 async->buffer = malloc(async->bufsize);
 941                 if (!async->buffer) {
 942                         free(async);
 943                         return -ENOMEM;
 944                 }
 945                 offset = 0;
 946                 start = async->start;
 947                 size = async->size;
 948
 949                 if (md->data) {
 950                         ret = read_data_extent(md, async);
 951                         if (ret) {
 952                                 free(async->buffer);
 953                                 free(async);
 954                                 return ret;
 955                         }
 956                 }
 957
 958                 /*
 959                  * Balance can make the mapping not cover the super block, so
 960                  * just copy directly from one of the devices.
 961                  */
 962                 if (start == BTRFS_SUPER_INFO_OFFSET) {
 963                         int fd = get_dev_fd(md->root);
 964
 965                         ret = pread64(fd, async->buffer, size, start);
 966                         if (ret < size) {
 967                                 free(async->buffer);
 968                                 free(async);
 969                                 error("unable to read superblock at %llu: %s",
 970                                                 (unsigned long long)start,
 971                                                 strerror(errno));
 972                                 return -errno;
 973                         }
 974                         size = 0;
 975                         ret = 0;
 976                 }
 977
 978                 while (!md->data && size > 0) {
 979                         u64 this_read = min(blocksize, size);
 980                         eb = read_tree_block(md->root, start, this_read, 0);
 981                         if (!extent_buffer_uptodate(eb)) {
 982                                 free(async->buffer);
 983                                 free(async);
 984                                 error("unable to read metadata block %llu",
 985                                         (unsigned long long)start);
 986                                 return -EIO;
 987                         }
 988                         copy_buffer(md, async->buffer + offset, eb);
 989                         free_extent_buffer(eb);
 990                         start += this_read;
 991                         offset += this_read;
 992                         size -= this_read;
 993                 }
 994
 995                 md->pending_start = (u64)-1;
 996                 md->pending_size = 0;
 997         } else if (!done) {
 998                 return 0;
 999         }
1000
1001         pthread_mutex_lock(&md->mutex);
1002         if (async) {
1003                 list_add_tail(&async->ordered, &md->ordered);
1004                 md->num_items++;
1005                 if (md->compress_level > 0) {
1006                         list_add_tail(&async->list, &md->list);
1007                         pthread_cond_signal(&md->cond);
1008                 } else {
1009                         md->num_ready++;
1010                 }
1011         }
1012         if (md->num_items >= ITEMS_PER_CLUSTER || done) {
1013                 ret = write_buffers(md, &start);
1014                 if (ret)
1015                         error("unable to write buffers: %s", strerror(-ret));
1016                 else
1017                         meta_cluster_init(md, start);
1018         }
1019         pthread_mutex_unlock(&md->mutex);
1020         return ret;
1021 }
1022
1023 static int add_extent(u64 start, u64 size, struct metadump_struct *md,
1024                       int data)
1025 {
1026         int ret;
1027         if (md->data != data ||
1028             md->pending_size + size > MAX_PENDING_SIZE ||
1029             md->pending_start + md->pending_size != start) {
1030                 ret = flush_pending(md, 0);
1031                 if (ret)
1032                         return ret;
1033                 md->pending_start = start;
1034         }
1035         readahead_tree_block(md->root, start, size, 0);
1036         md->pending_size += size;
1037         md->data = data;
1038         return 0;
1039 }
1040
1041 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1042 static int is_tree_block(struct btrfs_root *extent_root,
1043                          struct btrfs_path *path, u64 bytenr)
1044 {
1045         struct extent_buffer *leaf;
1046         struct btrfs_key key;
1047         u64 ref_objectid;
1048         int ret;
1049
1050         leaf = path->nodes[0];
1051         while (1) {
1052                 struct btrfs_extent_ref_v0 *ref_item;
1053                 path->slots[0]++;
1054                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1055                         ret = btrfs_next_leaf(extent_root, path);
1056                         if (ret < 0)
1057                                 return ret;
1058                         if (ret > 0)
1059                                 break;
1060                         leaf = path->nodes[0];
1061                 }
1062                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1063                 if (key.objectid != bytenr)
1064                         break;
1065                 if (key.type != BTRFS_EXTENT_REF_V0_KEY)
1066                         continue;
1067                 ref_item = btrfs_item_ptr(leaf, path->slots[0],
1068                                           struct btrfs_extent_ref_v0);
1069                 ref_objectid = btrfs_ref_objectid_v0(leaf, ref_item);
1070                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID)
1071                         return 1;
1072                 break;
1073         }
1074         return 0;
1075 }
1076 #endif
1077
1078 static int copy_tree_blocks(struct btrfs_root *root, struct extent_buffer *eb,
1079                             struct metadump_struct *metadump, int root_tree)
1080 {
1081         struct extent_buffer *tmp;
1082         struct btrfs_root_item *ri;
1083         struct btrfs_key key;
1084         u64 bytenr;
1085         int level;
1086         int nritems = 0;
1087         int i = 0;
1088         int ret;
1089
1090         ret = add_extent(btrfs_header_bytenr(eb), root->nodesize, metadump, 0);
1091         if (ret) {
1092                 error("unable to add metadata block %llu: %d",
1093                                 btrfs_header_bytenr(eb), ret);
1094                 return ret;
1095         }
1096
1097         if (btrfs_header_level(eb) == 0 && !root_tree)
1098                 return 0;
1099
1100         level = btrfs_header_level(eb);
1101         nritems = btrfs_header_nritems(eb);
1102         for (i = 0; i < nritems; i++) {
1103                 if (level == 0) {
1104                         btrfs_item_key_to_cpu(eb, &key, i);
1105                         if (key.type != BTRFS_ROOT_ITEM_KEY)
1106                                 continue;
1107                         ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
1108                         bytenr = btrfs_disk_root_bytenr(eb, ri);
1109                         tmp = read_tree_block(root, bytenr, root->nodesize, 0);
1110                         if (!extent_buffer_uptodate(tmp)) {
1111                                 error("unable to read log root block");
1112                                 return -EIO;
1113                         }
1114                         ret = copy_tree_blocks(root, tmp, metadump, 0);
1115                         free_extent_buffer(tmp);
1116                         if (ret)
1117                                 return ret;
1118                 } else {
1119                         bytenr = btrfs_node_blockptr(eb, i);
1120                         tmp = read_tree_block(root, bytenr, root->nodesize, 0);
1121                         if (!extent_buffer_uptodate(tmp)) {
1122                                 error("unable to read log root block");
1123                                 return -EIO;
1124                         }
1125                         ret = copy_tree_blocks(root, tmp, metadump, root_tree);
1126                         free_extent_buffer(tmp);
1127                         if (ret)
1128                                 return ret;
1129                 }
1130         }
1131
1132         return 0;
1133 }
1134
1135 static int copy_log_trees(struct btrfs_root *root,
1136                           struct metadump_struct *metadump,
1137                           struct btrfs_path *path)
1138 {
1139         u64 blocknr = btrfs_super_log_root(root->fs_info->super_copy);
1140
1141         if (blocknr == 0)
1142                 return 0;
1143
1144         if (!root->fs_info->log_root_tree ||
1145             !root->fs_info->log_root_tree->node) {
1146                 error("unable to copy tree log, it has not been setup");
1147                 return -EIO;
1148         }
1149
1150         return copy_tree_blocks(root, root->fs_info->log_root_tree->node,
1151                                 metadump, 1);
1152 }
1153
1154 static int copy_space_cache(struct btrfs_root *root,
1155                             struct metadump_struct *metadump,
1156                             struct btrfs_path *path)
1157 {
1158         struct extent_buffer *leaf;
1159         struct btrfs_file_extent_item *fi;
1160         struct btrfs_key key;
1161         u64 bytenr, num_bytes;
1162         int ret;
1163
1164         root = root->fs_info->tree_root;
1165
1166         key.objectid = 0;
1167         key.type = BTRFS_EXTENT_DATA_KEY;
1168         key.offset = 0;
1169
1170         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1171         if (ret < 0) {
1172                 error("free space inode not found: %d", ret);
1173                 return ret;
1174         }
1175
1176         leaf = path->nodes[0];
1177
1178         while (1) {
1179                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1180                         ret = btrfs_next_leaf(root, path);
1181                         if (ret < 0) {
1182                                 error("cannot go to next leaf %d", ret);
1183                                 return ret;
1184                         }
1185                         if (ret > 0)
1186                                 break;
1187                         leaf = path->nodes[0];
1188                 }
1189
1190                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1191                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
1192                         path->slots[0]++;
1193                         continue;
1194                 }
1195
1196                 fi = btrfs_item_ptr(leaf, path->slots[0],
1197                                     struct btrfs_file_extent_item);
1198                 if (btrfs_file_extent_type(leaf, fi) !=
1199                     BTRFS_FILE_EXTENT_REG) {
1200                         path->slots[0]++;
1201                         continue;
1202                 }
1203
1204                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1205                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1206                 ret = add_extent(bytenr, num_bytes, metadump, 1);
1207                 if (ret) {
1208                         error("unable to add space cache blocks %d", ret);
1209                         btrfs_release_path(path);
1210                         return ret;
1211                 }
1212                 path->slots[0]++;
1213         }
1214
1215         return 0;
1216 }
1217
1218 static int copy_from_extent_tree(struct metadump_struct *metadump,
1219                                  struct btrfs_path *path)
1220 {
1221         struct btrfs_root *extent_root;
1222         struct extent_buffer *leaf;
1223         struct btrfs_extent_item *ei;
1224         struct btrfs_key key;
1225         u64 bytenr;
1226         u64 num_bytes;
1227         int ret;
1228
1229         extent_root = metadump->root->fs_info->extent_root;
1230         bytenr = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
1231         key.objectid = bytenr;
1232         key.type = BTRFS_EXTENT_ITEM_KEY;
1233         key.offset = 0;
1234
1235         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1236         if (ret < 0) {
1237                 error("extent root not found: %d", ret);
1238                 return ret;
1239         }
1240         ret = 0;
1241
1242         leaf = path->nodes[0];
1243
1244         while (1) {
1245                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1246                         ret = btrfs_next_leaf(extent_root, path);
1247                         if (ret < 0) {
1248                                 error("cannot go to next leaf %d", ret);
1249                                 break;
1250                         }
1251                         if (ret > 0) {
1252                                 ret = 0;
1253                                 break;
1254                         }
1255                         leaf = path->nodes[0];
1256                 }
1257
1258                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1259                 if (key.objectid < bytenr ||
1260                     (key.type != BTRFS_EXTENT_ITEM_KEY &&
1261                      key.type != BTRFS_METADATA_ITEM_KEY)) {
1262                         path->slots[0]++;
1263                         continue;
1264                 }
1265
1266                 bytenr = key.objectid;
1267                 if (key.type == BTRFS_METADATA_ITEM_KEY)
1268                         num_bytes = extent_root->nodesize;
1269                 else
1270                         num_bytes = key.offset;
1271
1272                 if (btrfs_item_size_nr(leaf, path->slots[0]) > sizeof(*ei)) {
1273                         ei = btrfs_item_ptr(leaf, path->slots[0],
1274                                             struct btrfs_extent_item);
1275                         if (btrfs_extent_flags(leaf, ei) &
1276                             BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1277                                 ret = add_extent(bytenr, num_bytes, metadump,
1278                                                  0);
1279                                 if (ret) {
1280                                         error("unable to add block %llu: %d",
1281                                                 (unsigned long long)bytenr, ret);
1282                                         break;
1283                                 }
1284                         }
1285                 } else {
1286 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1287                         ret = is_tree_block(extent_root, path, bytenr);
1288                         if (ret < 0) {
1289                                 error("failed to check tree block %llu: %d",
1290                                         (unsigned long long)bytenr, ret);
1291                                 break;
1292                         }
1293
1294                         if (ret) {
1295                                 ret = add_extent(bytenr, num_bytes, metadump,
1296                                                  0);
1297                                 if (ret) {
1298                                         error("unable to add block %llu: %d",
1299                                                 (unsigned long long)bytenr, ret);
1300                                         break;
1301                                 }
1302                         }
1303                         ret = 0;
1304 #else
1305                         error(
1306         "either extent tree is corrupted or you haven't built with V0 support");
1307                         ret = -EIO;
1308                         break;
1309 #endif
1310                 }
1311                 bytenr += num_bytes;
1312         }
1313
1314         btrfs_release_path(path);
1315
1316         return ret;
1317 }
1318
1319 static int create_metadump(const char *input, FILE *out, int num_threads,
1320                            int compress_level, int sanitize, int walk_trees)
1321 {
1322         struct btrfs_root *root;
1323         struct btrfs_path *path = NULL;
1324         struct metadump_struct metadump;
1325         int ret;
1326         int err = 0;
1327
1328         root = open_ctree(input, 0, 0);
1329         if (!root) {
1330                 error("open ctree failed");
1331                 return -EIO;
1332         }
1333
1334         ret = metadump_init(&metadump, root, out, num_threads,
1335                             compress_level, sanitize);
1336         if (ret) {
1337                 error("failed to initialize metadump: %d", ret);
1338                 close_ctree(root);
1339                 return ret;
1340         }
1341
1342         ret = add_extent(BTRFS_SUPER_INFO_OFFSET, BTRFS_SUPER_INFO_SIZE,
1343                         &metadump, 0);
1344         if (ret) {
1345                 error("unable to add metadata: %d", ret);
1346                 err = ret;
1347                 goto out;
1348         }
1349
1350         path = btrfs_alloc_path();
1351         if (!path) {
1352                 error("not enough memory to allocate path");
1353                 err = -ENOMEM;
1354                 goto out;
1355         }
1356
1357         if (walk_trees) {
1358                 ret = copy_tree_blocks(root, root->fs_info->chunk_root->node,
1359                                        &metadump, 1);
1360                 if (ret) {
1361                         err = ret;
1362                         goto out;
1363                 }
1364
1365                 ret = copy_tree_blocks(root, root->fs_info->tree_root->node,
1366                                        &metadump, 1);
1367                 if (ret) {
1368                         err = ret;
1369                         goto out;
1370                 }
1371         } else {
1372                 ret = copy_from_extent_tree(&metadump, path);
1373                 if (ret) {
1374                         err = ret;
1375                         goto out;
1376                 }
1377         }
1378
1379         ret = copy_log_trees(root, &metadump, path);
1380         if (ret) {
1381                 err = ret;
1382                 goto out;
1383         }
1384
1385         ret = copy_space_cache(root, &metadump, path);
1386 out:
1387         ret = flush_pending(&metadump, 1);
1388         if (ret) {
1389                 if (!err)
1390                         err = ret;
1391                 error("failed to flush pending data: %d", ret);
1392         }
1393
1394         metadump_destroy(&metadump, num_threads);
1395
1396         btrfs_free_path(path);
1397         ret = close_ctree(root);
1398         return err ? err : ret;
1399 }
1400
1401 static void update_super_old(u8 *buffer)
1402 {
1403         struct btrfs_super_block *super = (struct btrfs_super_block *)buffer;
1404         struct btrfs_chunk *chunk;
1405         struct btrfs_disk_key *key;
1406         u32 sectorsize = btrfs_super_sectorsize(super);
1407         u64 flags = btrfs_super_flags(super);
1408
1409         flags |= BTRFS_SUPER_FLAG_METADUMP;
1410         btrfs_set_super_flags(super, flags);
1411
1412         key = (struct btrfs_disk_key *)(super->sys_chunk_array);
1413         chunk = (struct btrfs_chunk *)(super->sys_chunk_array +
1414                                        sizeof(struct btrfs_disk_key));
1415
1416         btrfs_set_disk_key_objectid(key, BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1417         btrfs_set_disk_key_type(key, BTRFS_CHUNK_ITEM_KEY);
1418         btrfs_set_disk_key_offset(key, 0);
1419
1420         btrfs_set_stack_chunk_length(chunk, (u64)-1);
1421         btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
1422         btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
1423         btrfs_set_stack_chunk_type(chunk, BTRFS_BLOCK_GROUP_SYSTEM);
1424         btrfs_set_stack_chunk_io_align(chunk, sectorsize);
1425         btrfs_set_stack_chunk_io_width(chunk, sectorsize);
1426         btrfs_set_stack_chunk_sector_size(chunk, sectorsize);
1427         btrfs_set_stack_chunk_num_stripes(chunk, 1);
1428         btrfs_set_stack_chunk_sub_stripes(chunk, 0);
1429         chunk->stripe.devid = super->dev_item.devid;
1430         btrfs_set_stack_stripe_offset(&chunk->stripe, 0);
1431         memcpy(chunk->stripe.dev_uuid, super->dev_item.uuid, BTRFS_UUID_SIZE);
1432         btrfs_set_super_sys_array_size(super, sizeof(*key) + sizeof(*chunk));
1433         csum_block(buffer, BTRFS_SUPER_INFO_SIZE);
1434 }
1435
1436 static int update_super(struct mdrestore_struct *mdres, u8 *buffer)
1437 {
1438         struct btrfs_super_block *super = (struct btrfs_super_block *)buffer;
1439         struct btrfs_chunk *chunk;
1440         struct btrfs_disk_key *disk_key;
1441         struct btrfs_key key;
1442         u64 flags = btrfs_super_flags(super);
1443         u32 new_array_size = 0;
1444         u32 array_size;
1445         u32 cur = 0;
1446         u8 *ptr, *write_ptr;
1447         int old_num_stripes;
1448
1449         write_ptr = ptr = super->sys_chunk_array;
1450         array_size = btrfs_super_sys_array_size(super);
1451
1452         while (cur < array_size) {
1453                 disk_key = (struct btrfs_disk_key *)ptr;
1454                 btrfs_disk_key_to_cpu(&key, disk_key);
1455
1456                 new_array_size += sizeof(*disk_key);
1457                 memmove(write_ptr, ptr, sizeof(*disk_key));
1458
1459                 write_ptr += sizeof(*disk_key);
1460                 ptr += sizeof(*disk_key);
1461                 cur += sizeof(*disk_key);
1462
1463                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1464                         u64 type, physical, physical_dup, size = 0;
1465
1466                         chunk = (struct btrfs_chunk *)ptr;
1467                         old_num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1468                         chunk = (struct btrfs_chunk *)write_ptr;
1469
1470                         memmove(write_ptr, ptr, sizeof(*chunk));
1471                         btrfs_set_stack_chunk_sub_stripes(chunk, 0);
1472                         type = btrfs_stack_chunk_type(chunk);
1473                         if (type & BTRFS_BLOCK_GROUP_DUP) {
1474                                 new_array_size += sizeof(struct btrfs_stripe);
1475                                 write_ptr += sizeof(struct btrfs_stripe);
1476                         } else {
1477                                 btrfs_set_stack_chunk_num_stripes(chunk, 1);
1478                                 btrfs_set_stack_chunk_type(chunk,
1479                                                 BTRFS_BLOCK_GROUP_SYSTEM);
1480                         }
1481                         chunk->stripe.devid = super->dev_item.devid;
1482                         physical = logical_to_physical(mdres, key.offset,
1483                                                        &size, &physical_dup);
1484                         if (size != (u64)-1)
1485                                 btrfs_set_stack_stripe_offset(&chunk->stripe,
1486                                                               physical);
1487                         memcpy(chunk->stripe.dev_uuid, super->dev_item.uuid,
1488                                BTRFS_UUID_SIZE);
1489                         new_array_size += sizeof(*chunk);
1490                 } else {
1491                         error("bogus key in the sys array %d", key.type);
1492                         return -EIO;
1493                 }
1494                 write_ptr += sizeof(*chunk);
1495                 ptr += btrfs_chunk_item_size(old_num_stripes);
1496                 cur += btrfs_chunk_item_size(old_num_stripes);
1497         }
1498
1499         if (mdres->clear_space_cache)
1500                 btrfs_set_super_cache_generation(super, 0);
1501
1502         flags |= BTRFS_SUPER_FLAG_METADUMP_V2;
1503         btrfs_set_super_flags(super, flags);
1504         btrfs_set_super_sys_array_size(super, new_array_size);
1505         csum_block(buffer, BTRFS_SUPER_INFO_SIZE);
1506
1507         return 0;
1508 }
1509
1510 static struct extent_buffer *alloc_dummy_eb(u64 bytenr, u32 size)
1511 {
1512         struct extent_buffer *eb;
1513
1514         eb = calloc(1, sizeof(struct extent_buffer) + size);
1515         if (!eb)
1516                 return NULL;
1517
1518         eb->start = bytenr;
1519         eb->len = size;
1520         return eb;
1521 }
1522
1523 static void truncate_item(struct extent_buffer *eb, int slot, u32 new_size)
1524 {
1525         struct btrfs_item *item;
1526         u32 nritems;
1527         u32 old_size;
1528         u32 old_data_start;
1529         u32 size_diff;
1530         u32 data_end;
1531         int i;
1532
1533         old_size = btrfs_item_size_nr(eb, slot);
1534         if (old_size == new_size)
1535                 return;
1536
1537         nritems = btrfs_header_nritems(eb);
1538         data_end = btrfs_item_offset_nr(eb, nritems - 1);
1539
1540         old_data_start = btrfs_item_offset_nr(eb, slot);
1541         size_diff = old_size - new_size;
1542
1543         for (i = slot; i < nritems; i++) {
1544                 u32 ioff;
1545                 item = btrfs_item_nr(i);
1546                 ioff = btrfs_item_offset(eb, item);
1547                 btrfs_set_item_offset(eb, item, ioff + size_diff);
1548         }
1549
1550         memmove_extent_buffer(eb, btrfs_leaf_data(eb) + data_end + size_diff,
1551                               btrfs_leaf_data(eb) + data_end,
1552                               old_data_start + new_size - data_end);
1553         item = btrfs_item_nr(slot);
1554         btrfs_set_item_size(eb, item, new_size);
1555 }
1556
1557 static int fixup_chunk_tree_block(struct mdrestore_struct *mdres,
1558                                   struct async_work *async, u8 *buffer,
1559                                   size_t size)
1560 {
1561         struct extent_buffer *eb;
1562         size_t size_left = size;
1563         u64 bytenr = async->start;
1564         int i;
1565
1566         if (size_left % mdres->nodesize)
1567                 return 0;
1568
1569         eb = alloc_dummy_eb(bytenr, mdres->nodesize);
1570         if (!eb)
1571                 return -ENOMEM;
1572
1573         while (size_left) {
1574                 eb->start = bytenr;
1575                 memcpy(eb->data, buffer, mdres->nodesize);
1576
1577                 if (btrfs_header_bytenr(eb) != bytenr)
1578                         break;
1579                 if (memcmp(mdres->fsid,
1580                            eb->data + offsetof(struct btrfs_header, fsid),
1581                            BTRFS_FSID_SIZE))
1582                         break;
1583
1584                 if (btrfs_header_owner(eb) != BTRFS_CHUNK_TREE_OBJECTID)
1585                         goto next;
1586
1587                 if (btrfs_header_level(eb) != 0)
1588                         goto next;
1589
1590                 for (i = 0; i < btrfs_header_nritems(eb); i++) {
1591                         struct btrfs_chunk *chunk;
1592                         struct btrfs_key key;
1593                         u64 type, physical, physical_dup, size = (u64)-1;
1594
1595                         btrfs_item_key_to_cpu(eb, &key, i);
1596                         if (key.type != BTRFS_CHUNK_ITEM_KEY)
1597                                 continue;
1598
1599                         size = 0;
1600                         physical = logical_to_physical(mdres, key.offset,
1601                                                        &size, &physical_dup);
1602
1603                         if (!physical_dup)
1604                                 truncate_item(eb, i, sizeof(*chunk));
1605                         chunk = btrfs_item_ptr(eb, i, struct btrfs_chunk);
1606
1607
1608                         /* Zero out the RAID profile */
1609                         type = btrfs_chunk_type(eb, chunk);
1610                         type &= (BTRFS_BLOCK_GROUP_DATA |
1611                                  BTRFS_BLOCK_GROUP_SYSTEM |
1612                                  BTRFS_BLOCK_GROUP_METADATA |
1613                                  BTRFS_BLOCK_GROUP_DUP);
1614                         btrfs_set_chunk_type(eb, chunk, type);
1615
1616                         if (!physical_dup)
1617                                 btrfs_set_chunk_num_stripes(eb, chunk, 1);
1618                         btrfs_set_chunk_sub_stripes(eb, chunk, 0);
1619                         btrfs_set_stripe_devid_nr(eb, chunk, 0, mdres->devid);
1620                         if (size != (u64)-1)
1621                                 btrfs_set_stripe_offset_nr(eb, chunk, 0,
1622                                                            physical);
1623                         /* update stripe 2 offset */
1624                         if (physical_dup)
1625                                 btrfs_set_stripe_offset_nr(eb, chunk, 1,
1626                                                            physical_dup);
1627
1628                         write_extent_buffer(eb, mdres->uuid,
1629                                         (unsigned long)btrfs_stripe_dev_uuid_nr(
1630                                                 chunk, 0),
1631                                         BTRFS_UUID_SIZE);
1632                 }
1633                 memcpy(buffer, eb->data, eb->len);
1634                 csum_block(buffer, eb->len);
1635 next:
1636                 size_left -= mdres->nodesize;
1637                 buffer += mdres->nodesize;
1638                 bytenr += mdres->nodesize;
1639         }
1640
1641         free(eb);
1642         return 0;
1643 }
1644
1645 static void write_backup_supers(int fd, u8 *buf)
1646 {
1647         struct btrfs_super_block *super = (struct btrfs_super_block *)buf;
1648         struct stat st;
1649         u64 size;
1650         u64 bytenr;
1651         int i;
1652         int ret;
1653
1654         if (fstat(fd, &st)) {
1655                 error(
1656         "cannot stat restore point, won't be able to write backup supers: %s",
1657                         strerror(errno));
1658                 return;
1659         }
1660
1661         size = btrfs_device_size(fd, &st);
1662
1663         for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1664                 bytenr = btrfs_sb_offset(i);
1665                 if (bytenr + BTRFS_SUPER_INFO_SIZE > size)
1666                         break;
1667                 btrfs_set_super_bytenr(super, bytenr);
1668                 csum_block(buf, BTRFS_SUPER_INFO_SIZE);
1669                 ret = pwrite64(fd, buf, BTRFS_SUPER_INFO_SIZE, bytenr);
1670                 if (ret < BTRFS_SUPER_INFO_SIZE) {
1671                         if (ret < 0)
1672                                 error(
1673                                 "problem writing out backup super block %d: %s",
1674                                                 i, strerror(errno));
1675                         else
1676                                 error("short write writing out backup super block");
1677                         break;
1678                 }
1679         }
1680 }
1681
1682 static void *restore_worker(void *data)
1683 {
1684         struct mdrestore_struct *mdres = (struct mdrestore_struct *)data;
1685         struct async_work *async;
1686         size_t size;
1687         u8 *buffer;
1688         u8 *outbuf;
1689         int outfd;
1690         int ret;
1691         int compress_size = MAX_PENDING_SIZE * 4;
1692
1693         outfd = fileno(mdres->out);
1694         buffer = malloc(compress_size);
1695         if (!buffer) {
1696                 error("not enough memory for restore worker buffer");
1697                 pthread_mutex_lock(&mdres->mutex);
1698                 if (!mdres->error)
1699                         mdres->error = -ENOMEM;
1700                 pthread_mutex_unlock(&mdres->mutex);
1701                 pthread_exit(NULL);
1702         }
1703
1704         while (1) {
1705                 u64 bytenr, physical_dup;
1706                 off_t offset = 0;
1707                 int err = 0;
1708
1709                 pthread_mutex_lock(&mdres->mutex);
1710                 while (!mdres->nodesize || list_empty(&mdres->list)) {
1711                         if (mdres->done) {
1712                                 pthread_mutex_unlock(&mdres->mutex);
1713                                 goto out;
1714                         }
1715                         pthread_cond_wait(&mdres->cond, &mdres->mutex);
1716                 }
1717                 async = list_entry(mdres->list.next, struct async_work, list);
1718                 list_del_init(&async->list);
1719                 pthread_mutex_unlock(&mdres->mutex);
1720
1721                 if (mdres->compress_method == COMPRESS_ZLIB) {
1722                         size = compress_size;
1723                         ret = uncompress(buffer, (unsigned long *)&size,
1724                                          async->buffer, async->bufsize);
1725                         if (ret != Z_OK) {
1726                                 error("decompressiion failed with %d", ret);
1727                                 err = -EIO;
1728                         }
1729                         outbuf = buffer;
1730                 } else {
1731                         outbuf = async->buffer;
1732                         size = async->bufsize;
1733                 }
1734
1735                 if (!mdres->multi_devices) {
1736                         if (async->start == BTRFS_SUPER_INFO_OFFSET) {
1737                                 if (mdres->old_restore) {
1738                                         update_super_old(outbuf);
1739                                 } else {
1740                                         ret = update_super(mdres, outbuf);
1741                                         if (ret)
1742                                                 err = ret;
1743                                 }
1744                         } else if (!mdres->old_restore) {
1745                                 ret = fixup_chunk_tree_block(mdres, async, outbuf, size);
1746                                 if (ret)
1747                                         err = ret;
1748                         }
1749                 }
1750
1751                 if (!mdres->fixup_offset) {
1752                         while (size) {
1753                                 u64 chunk_size = size;
1754                                 physical_dup = 0;
1755                                 if (!mdres->multi_devices && !mdres->old_restore)
1756                                         bytenr = logical_to_physical(mdres,
1757                                                      async->start + offset,
1758                                                      &chunk_size,
1759                                                      &physical_dup);
1760                                 else
1761                                         bytenr = async->start + offset;
1762
1763                                 ret = pwrite64(outfd, outbuf+offset, chunk_size,
1764                                                bytenr);
1765                                 if (ret != chunk_size)
1766                                         goto error;
1767
1768                                 if (physical_dup)
1769                                         ret = pwrite64(outfd, outbuf+offset,
1770                                                        chunk_size,
1771                                                        physical_dup);
1772                                 if (ret != chunk_size)
1773                                         goto error;
1774
1775                                 size -= chunk_size;
1776                                 offset += chunk_size;
1777                                 continue;
1778
1779 error:
1780                                 if (ret < 0) {
1781                                         error("unable to write to device: %s",
1782                                                         strerror(errno));
1783                                         err = errno;
1784                                 } else {
1785                                         error("short write");
1786                                         err = -EIO;
1787                                 }
1788                         }
1789                 } else if (async->start != BTRFS_SUPER_INFO_OFFSET) {
1790                         ret = write_data_to_disk(mdres->info, outbuf, async->start, size, 0);
1791                         if (ret) {
1792                                 error("failed to write data");
1793                                 exit(1);
1794                         }
1795                 }
1796
1797
1798                 /* backup super blocks are already there at fixup_offset stage */
1799                 if (!mdres->multi_devices && async->start == BTRFS_SUPER_INFO_OFFSET)
1800                         write_backup_supers(outfd, outbuf);
1801
1802                 pthread_mutex_lock(&mdres->mutex);
1803                 if (err && !mdres->error)
1804                         mdres->error = err;
1805                 mdres->num_items--;
1806                 pthread_mutex_unlock(&mdres->mutex);
1807
1808                 free(async->buffer);
1809                 free(async);
1810         }
1811 out:
1812         free(buffer);
1813         pthread_exit(NULL);
1814 }
1815
1816 static void mdrestore_destroy(struct mdrestore_struct *mdres, int num_threads)
1817 {
1818         struct rb_node *n;
1819         int i;
1820
1821         while ((n = rb_first(&mdres->chunk_tree))) {
1822                 struct fs_chunk *entry;
1823
1824                 entry = rb_entry(n, struct fs_chunk, l);
1825                 rb_erase(n, &mdres->chunk_tree);
1826                 rb_erase(&entry->p, &mdres->physical_tree);
1827                 free(entry);
1828         }
1829         pthread_mutex_lock(&mdres->mutex);
1830         mdres->done = 1;
1831         pthread_cond_broadcast(&mdres->cond);
1832         pthread_mutex_unlock(&mdres->mutex);
1833
1834         for (i = 0; i < num_threads; i++)
1835                 pthread_join(mdres->threads[i], NULL);
1836
1837         pthread_cond_destroy(&mdres->cond);
1838         pthread_mutex_destroy(&mdres->mutex);
1839         free(mdres->threads);
1840 }
1841
1842 static int mdrestore_init(struct mdrestore_struct *mdres,
1843                           FILE *in, FILE *out, int old_restore,
1844                           int num_threads, int fixup_offset,
1845                           struct btrfs_fs_info *info, int multi_devices)
1846 {
1847         int i, ret = 0;
1848
1849         memset(mdres, 0, sizeof(*mdres));
1850         pthread_cond_init(&mdres->cond, NULL);
1851         pthread_mutex_init(&mdres->mutex, NULL);
1852         INIT_LIST_HEAD(&mdres->list);
1853         INIT_LIST_HEAD(&mdres->overlapping_chunks);
1854         mdres->in = in;
1855         mdres->out = out;
1856         mdres->old_restore = old_restore;
1857         mdres->chunk_tree.rb_node = NULL;
1858         mdres->fixup_offset = fixup_offset;
1859         mdres->info = info;
1860         mdres->multi_devices = multi_devices;
1861         mdres->clear_space_cache = 0;
1862         mdres->last_physical_offset = 0;
1863         mdres->alloced_chunks = 0;
1864
1865         if (!num_threads)
1866                 return 0;
1867
1868         mdres->num_threads = num_threads;
1869         mdres->threads = calloc(num_threads, sizeof(pthread_t));
1870         if (!mdres->threads)
1871                 return -ENOMEM;
1872         for (i = 0; i < num_threads; i++) {
1873                 ret = pthread_create(mdres->threads + i, NULL, restore_worker,
1874                                      mdres);
1875                 if (ret)
1876                         break;
1877         }
1878         if (ret)
1879                 mdrestore_destroy(mdres, i + 1);
1880         return ret;
1881 }
1882
1883 static int fill_mdres_info(struct mdrestore_struct *mdres,
1884                            struct async_work *async)
1885 {
1886         struct btrfs_super_block *super;
1887         u8 *buffer = NULL;
1888         u8 *outbuf;
1889         int ret;
1890
1891         /* We've already been initialized */
1892         if (mdres->nodesize)
1893                 return 0;
1894
1895         if (mdres->compress_method == COMPRESS_ZLIB) {
1896                 size_t size = MAX_PENDING_SIZE * 2;
1897
1898                 buffer = malloc(MAX_PENDING_SIZE * 2);
1899                 if (!buffer)
1900                         return -ENOMEM;
1901                 ret = uncompress(buffer, (unsigned long *)&size,
1902                                  async->buffer, async->bufsize);
1903                 if (ret != Z_OK) {
1904                         error("decompressiion failed with %d", ret);
1905                         free(buffer);
1906                         return -EIO;
1907                 }
1908                 outbuf = buffer;
1909         } else {
1910                 outbuf = async->buffer;
1911         }
1912
1913         super = (struct btrfs_super_block *)outbuf;
1914         mdres->nodesize = btrfs_super_nodesize(super);
1915         memcpy(mdres->fsid, super->fsid, BTRFS_FSID_SIZE);
1916         memcpy(mdres->uuid, super->dev_item.uuid,
1917                        BTRFS_UUID_SIZE);
1918         mdres->devid = le64_to_cpu(super->dev_item.devid);
1919         free(buffer);
1920         return 0;
1921 }
1922
1923 static int add_cluster(struct meta_cluster *cluster,
1924                        struct mdrestore_struct *mdres, u64 *next)
1925 {
1926         struct meta_cluster_item *item;
1927         struct meta_cluster_header *header = &cluster->header;
1928         struct async_work *async;
1929         u64 bytenr;
1930         u32 i, nritems;
1931         int ret;
1932
1933         mdres->compress_method = header->compress;
1934
1935         bytenr = le64_to_cpu(header->bytenr) + BLOCK_SIZE;
1936         nritems = le32_to_cpu(header->nritems);
1937         for (i = 0; i < nritems; i++) {
1938                 item = &cluster->items[i];
1939                 async = calloc(1, sizeof(*async));
1940                 if (!async) {
1941                         error("not enough memory for async data");
1942                         return -ENOMEM;
1943                 }
1944                 async->start = le64_to_cpu(item->bytenr);
1945                 async->bufsize = le32_to_cpu(item->size);
1946                 async->buffer = malloc(async->bufsize);
1947                 if (!async->buffer) {
1948                         error("not enough memory for async buffer");
1949                         free(async);
1950                         return -ENOMEM;
1951                 }
1952                 ret = fread(async->buffer, async->bufsize, 1, mdres->in);
1953                 if (ret != 1) {
1954                         error("unable to read buffer: %s", strerror(errno));
1955                         free(async->buffer);
1956                         free(async);
1957                         return -EIO;
1958                 }
1959                 bytenr += async->bufsize;
1960
1961                 pthread_mutex_lock(&mdres->mutex);
1962                 if (async->start == BTRFS_SUPER_INFO_OFFSET) {
1963                         ret = fill_mdres_info(mdres, async);
1964                         if (ret) {
1965                                 error("unable to set up restore state");
1966                                 pthread_mutex_unlock(&mdres->mutex);
1967                                 free(async->buffer);
1968                                 free(async);
1969                                 return ret;
1970                         }
1971                 }
1972                 list_add_tail(&async->list, &mdres->list);
1973                 mdres->num_items++;
1974                 pthread_cond_signal(&mdres->cond);
1975                 pthread_mutex_unlock(&mdres->mutex);
1976         }
1977         if (bytenr & BLOCK_MASK) {
1978                 char buffer[BLOCK_MASK];
1979                 size_t size = BLOCK_SIZE - (bytenr & BLOCK_MASK);
1980
1981                 bytenr += size;
1982                 ret = fread(buffer, size, 1, mdres->in);
1983                 if (ret != 1) {
1984                         error("failed to read buffer: %s", strerror(errno));
1985                         return -EIO;
1986                 }
1987         }
1988         *next = bytenr;
1989         return 0;
1990 }
1991
1992 static int wait_for_worker(struct mdrestore_struct *mdres)
1993 {
1994         int ret = 0;
1995
1996         pthread_mutex_lock(&mdres->mutex);
1997         ret = mdres->error;
1998         while (!ret && mdres->num_items > 0) {
1999                 struct timespec ts = {
2000                         .tv_sec = 0,
2001                         .tv_nsec = 10000000,
2002                 };
2003                 pthread_mutex_unlock(&mdres->mutex);
2004                 nanosleep(&ts, NULL);
2005                 pthread_mutex_lock(&mdres->mutex);
2006                 ret = mdres->error;
2007         }
2008         pthread_mutex_unlock(&mdres->mutex);
2009         return ret;
2010 }
2011
2012 static int read_chunk_block(struct mdrestore_struct *mdres, u8 *buffer,
2013                             u64 bytenr, u64 item_bytenr, u32 bufsize,
2014                             u64 cluster_bytenr)
2015 {
2016         struct extent_buffer *eb;
2017         int ret = 0;
2018         int i;
2019
2020         eb = alloc_dummy_eb(bytenr, mdres->nodesize);
2021         if (!eb) {
2022                 ret = -ENOMEM;
2023                 goto out;
2024         }
2025
2026         while (item_bytenr != bytenr) {
2027                 buffer += mdres->nodesize;
2028                 item_bytenr += mdres->nodesize;
2029         }
2030
2031         memcpy(eb->data, buffer, mdres->nodesize);
2032         if (btrfs_header_bytenr(eb) != bytenr) {
2033                 error("eb bytenr does not match found bytenr: %llu != %llu",
2034                                 (unsigned long long)btrfs_header_bytenr(eb),
2035                                 (unsigned long long)bytenr);
2036                 ret = -EIO;
2037                 goto out;
2038         }
2039
2040         if (memcmp(mdres->fsid, eb->data + offsetof(struct btrfs_header, fsid),
2041                    BTRFS_FSID_SIZE)) {
2042                 error("filesystem UUID of eb %llu does not match",
2043                                 (unsigned long long)bytenr);
2044                 ret = -EIO;
2045                 goto out;
2046         }
2047
2048         if (btrfs_header_owner(eb) != BTRFS_CHUNK_TREE_OBJECTID) {
2049                 error("wrong eb %llu owner %llu",
2050                                 (unsigned long long)bytenr,
2051                                 (unsigned long long)btrfs_header_owner(eb));
2052                 ret = -EIO;
2053                 goto out;
2054         }
2055
2056         for (i = 0; i < btrfs_header_nritems(eb); i++) {
2057                 struct btrfs_chunk *chunk;
2058                 struct fs_chunk *fs_chunk;
2059                 struct btrfs_key key;
2060                 u64 type;
2061
2062                 if (btrfs_header_level(eb)) {
2063                         u64 blockptr = btrfs_node_blockptr(eb, i);
2064
2065                         ret = search_for_chunk_blocks(mdres, blockptr,
2066                                                       cluster_bytenr);
2067                         if (ret)
2068                                 break;
2069                         continue;
2070                 }
2071
2072                 /* Yay a leaf!  We loves leafs! */
2073                 btrfs_item_key_to_cpu(eb, &key, i);
2074                 if (key.type != BTRFS_CHUNK_ITEM_KEY)
2075                         continue;
2076
2077                 fs_chunk = malloc(sizeof(struct fs_chunk));
2078                 if (!fs_chunk) {
2079                         error("not enough memory to allocate chunk");
2080                         ret = -ENOMEM;
2081                         break;
2082                 }
2083                 memset(fs_chunk, 0, sizeof(*fs_chunk));
2084                 chunk = btrfs_item_ptr(eb, i, struct btrfs_chunk);
2085
2086                 fs_chunk->logical = key.offset;
2087                 fs_chunk->physical = btrfs_stripe_offset_nr(eb, chunk, 0);
2088                 fs_chunk->bytes = btrfs_chunk_length(eb, chunk);
2089                 INIT_LIST_HEAD(&fs_chunk->list);
2090                 if (tree_search(&mdres->physical_tree, &fs_chunk->p,
2091                                 physical_cmp, 1) != NULL)
2092                         list_add(&fs_chunk->list, &mdres->overlapping_chunks);
2093                 else
2094                         tree_insert(&mdres->physical_tree, &fs_chunk->p,
2095                                     physical_cmp);
2096
2097                 type = btrfs_chunk_type(eb, chunk);
2098                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2099                         fs_chunk->physical_dup =
2100                                         btrfs_stripe_offset_nr(eb, chunk, 1);
2101                 }
2102
2103                 if (fs_chunk->physical_dup + fs_chunk->bytes >
2104                     mdres->last_physical_offset)
2105                         mdres->last_physical_offset = fs_chunk->physical_dup +
2106                                 fs_chunk->bytes;
2107                 else if (fs_chunk->physical + fs_chunk->bytes >
2108                     mdres->last_physical_offset)
2109                         mdres->last_physical_offset = fs_chunk->physical +
2110                                 fs_chunk->bytes;
2111                 mdres->alloced_chunks += fs_chunk->bytes;
2112                 /* in dup case, fs_chunk->bytes should add twice */
2113                 if (fs_chunk->physical_dup)
2114                         mdres->alloced_chunks += fs_chunk->bytes;
2115                 tree_insert(&mdres->chunk_tree, &fs_chunk->l, chunk_cmp);
2116         }
2117 out:
2118         free(eb);
2119         return ret;
2120 }
2121
2122 /* If you have to ask you aren't worthy */
2123 static int search_for_chunk_blocks(struct mdrestore_struct *mdres,
2124                                    u64 search, u64 cluster_bytenr)
2125 {
2126         struct meta_cluster *cluster;
2127         struct meta_cluster_header *header;
2128         struct meta_cluster_item *item;
2129         u64 current_cluster = cluster_bytenr, bytenr;
2130         u64 item_bytenr;
2131         u32 bufsize, nritems, i;
2132         u32 max_size = MAX_PENDING_SIZE * 2;
2133         u8 *buffer, *tmp = NULL;
2134         int ret = 0;
2135
2136         cluster = malloc(BLOCK_SIZE);
2137         if (!cluster) {
2138                 error("not enough memory for cluster");
2139                 return -ENOMEM;
2140         }
2141
2142         buffer = malloc(max_size);
2143         if (!buffer) {
2144                 error("not enough memory for buffer");
2145                 free(cluster);
2146                 return -ENOMEM;
2147         }
2148
2149         if (mdres->compress_method == COMPRESS_ZLIB) {
2150                 tmp = malloc(max_size);
2151                 if (!tmp) {
2152                         error("not enough memory for buffer");
2153                         free(cluster);
2154                         free(buffer);
2155                         return -ENOMEM;
2156                 }
2157         }
2158
2159         bytenr = current_cluster;
2160         while (1) {
2161                 if (fseek(mdres->in, current_cluster, SEEK_SET)) {
2162                         error("seek failed: %s\n", strerror(errno));
2163                         ret = -EIO;
2164                         break;
2165                 }
2166
2167                 ret = fread(cluster, BLOCK_SIZE, 1, mdres->in);
2168                 if (ret == 0) {
2169                         if (cluster_bytenr != 0) {
2170                                 cluster_bytenr = 0;
2171                                 current_cluster = 0;
2172                                 bytenr = 0;
2173                                 continue;
2174                         }
2175                         error(
2176         "unknown state after reading cluster at %llu, probably crrupted data",
2177                                         cluster_bytenr);
2178                         ret = -EIO;
2179                         break;
2180                 } else if (ret < 0) {
2181                         error("unable to read image at %llu: %s",
2182                                         (unsigned long long)cluster_bytenr,
2183                                         strerror(errno));
2184                         break;
2185                 }
2186                 ret = 0;
2187
2188                 header = &cluster->header;
2189                 if (le64_to_cpu(header->magic) != HEADER_MAGIC ||
2190                     le64_to_cpu(header->bytenr) != current_cluster) {
2191                         error("bad header in metadump image");
2192                         ret = -EIO;
2193                         break;
2194                 }
2195
2196                 bytenr += BLOCK_SIZE;
2197                 nritems = le32_to_cpu(header->nritems);
2198                 for (i = 0; i < nritems; i++) {
2199                         size_t size;
2200
2201                         item = &cluster->items[i];
2202                         bufsize = le32_to_cpu(item->size);
2203                         item_bytenr = le64_to_cpu(item->bytenr);
2204
2205                         if (bufsize > max_size) {
2206                                 error("item %u too big: %u > %u", i, bufsize,
2207                                                 max_size);
2208                                 ret = -EIO;
2209                                 break;
2210                         }
2211
2212                         if (mdres->compress_method == COMPRESS_ZLIB) {
2213                                 ret = fread(tmp, bufsize, 1, mdres->in);
2214                                 if (ret != 1) {
2215                                         error("read error: %s", strerror(errno));
2216                                         ret = -EIO;
2217                                         break;
2218                                 }
2219
2220                                 size = max_size;
2221                                 ret = uncompress(buffer,
2222                                                  (unsigned long *)&size, tmp,
2223                                                  bufsize);
2224                                 if (ret != Z_OK) {
2225                                         error("decompressiion failed with %d",
2226                                                         ret);
2227                                         ret = -EIO;
2228                                         break;
2229                                 }
2230                         } else {
2231                                 ret = fread(buffer, bufsize, 1, mdres->in);
2232                                 if (ret != 1) {
2233                                         error("read error: %s",
2234                                                         strerror(errno));
2235                                         ret = -EIO;
2236                                         break;
2237                                 }
2238                                 size = bufsize;
2239                         }
2240                         ret = 0;
2241
2242                         if (item_bytenr <= search &&
2243                             item_bytenr + size > search) {
2244                                 ret = read_chunk_block(mdres, buffer, search,
2245                                                        item_bytenr, size,
2246                                                        current_cluster);
2247                                 if (!ret)
2248                                         ret = 1;
2249                                 break;
2250                         }
2251                         bytenr += bufsize;
2252                 }
2253                 if (ret) {
2254                         if (ret > 0)
2255                                 ret = 0;
2256                         break;
2257                 }
2258                 if (bytenr & BLOCK_MASK)
2259                         bytenr += BLOCK_SIZE - (bytenr & BLOCK_MASK);
2260                 current_cluster = bytenr;
2261         }
2262
2263         free(tmp);
2264         free(buffer);
2265         free(cluster);
2266         return ret;
2267 }
2268
2269 static int build_chunk_tree(struct mdrestore_struct *mdres,
2270                             struct meta_cluster *cluster)
2271 {
2272         struct btrfs_super_block *super;
2273         struct meta_cluster_header *header;
2274         struct meta_cluster_item *item = NULL;
2275         u64 chunk_root_bytenr = 0;
2276         u32 i, nritems;
2277         u64 bytenr = 0;
2278         u8 *buffer;
2279         int ret;
2280
2281         /* We can't seek with stdin so don't bother doing this */
2282         if (mdres->in == stdin)
2283                 return 0;
2284
2285         ret = fread(cluster, BLOCK_SIZE, 1, mdres->in);
2286         if (ret <= 0) {
2287                 error("unable to read cluster: %s", strerror(errno));
2288                 return -EIO;
2289         }
2290         ret = 0;
2291
2292         header = &cluster->header;
2293         if (le64_to_cpu(header->magic) != HEADER_MAGIC ||
2294             le64_to_cpu(header->bytenr) != 0) {
2295                 error("bad header in metadump image");
2296                 return -EIO;
2297         }
2298
2299         bytenr += BLOCK_SIZE;
2300         mdres->compress_method = header->compress;
2301         nritems = le32_to_cpu(header->nritems);
2302         for (i = 0; i < nritems; i++) {
2303                 item = &cluster->items[i];
2304
2305                 if (le64_to_cpu(item->bytenr) == BTRFS_SUPER_INFO_OFFSET)
2306                         break;
2307                 bytenr += le32_to_cpu(item->size);
2308                 if (fseek(mdres->in, le32_to_cpu(item->size), SEEK_CUR)) {
2309                         error("seek failed: %s\n", strerror(errno));
2310                         return -EIO;
2311                 }
2312         }
2313
2314         if (!item || le64_to_cpu(item->bytenr) != BTRFS_SUPER_INFO_OFFSET) {
2315                 error("did not find superblock at %llu",
2316                                 le64_to_cpu(item->bytenr));
2317                 return -EINVAL;
2318         }
2319
2320         buffer = malloc(le32_to_cpu(item->size));
2321         if (!buffer) {
2322                 error("not enough memory to allocate buffer");
2323                 return -ENOMEM;
2324         }
2325
2326         ret = fread(buffer, le32_to_cpu(item->size), 1, mdres->in);
2327         if (ret != 1) {
2328                 error("unable to read buffer: %s", strerror(errno));
2329                 free(buffer);
2330                 return -EIO;
2331         }
2332
2333         if (mdres->compress_method == COMPRESS_ZLIB) {
2334                 size_t size = MAX_PENDING_SIZE * 2;
2335                 u8 *tmp;
2336
2337                 tmp = malloc(MAX_PENDING_SIZE * 2);
2338                 if (!tmp) {
2339                         free(buffer);
2340                         return -ENOMEM;
2341                 }
2342                 ret = uncompress(tmp, (unsigned long *)&size,
2343                                  buffer, le32_to_cpu(item->size));
2344                 if (ret != Z_OK) {
2345                         error("decompressiion failed with %d", ret);
2346                         free(buffer);
2347                         free(tmp);
2348                         return -EIO;
2349                 }
2350                 free(buffer);
2351                 buffer = tmp;
2352         }
2353
2354         pthread_mutex_lock(&mdres->mutex);
2355         super = (struct btrfs_super_block *)buffer;
2356         chunk_root_bytenr = btrfs_super_chunk_root(super);
2357         mdres->nodesize = btrfs_super_nodesize(super);
2358         memcpy(mdres->fsid, super->fsid, BTRFS_FSID_SIZE);
2359         memcpy(mdres->uuid, super->dev_item.uuid,
2360                        BTRFS_UUID_SIZE);
2361         mdres->devid = le64_to_cpu(super->dev_item.devid);
2362         free(buffer);
2363         pthread_mutex_unlock(&mdres->mutex);
2364
2365         return search_for_chunk_blocks(mdres, chunk_root_bytenr, 0);
2366 }
2367
2368 static int range_contains_super(u64 physical, u64 bytes)
2369 {
2370         u64 super_bytenr;
2371         int i;
2372
2373         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
2374                 super_bytenr = btrfs_sb_offset(i);
2375                 if (super_bytenr >= physical &&
2376                     super_bytenr < physical + bytes)
2377                         return 1;
2378         }
2379
2380         return 0;
2381 }
2382
2383 static void remap_overlapping_chunks(struct mdrestore_struct *mdres)
2384 {
2385         struct fs_chunk *fs_chunk;
2386
2387         while (!list_empty(&mdres->overlapping_chunks)) {
2388                 fs_chunk = list_first_entry(&mdres->overlapping_chunks,
2389                                             struct fs_chunk, list);
2390                 list_del_init(&fs_chunk->list);
2391                 if (range_contains_super(fs_chunk->physical,
2392                                          fs_chunk->bytes)) {
2393                         warning(
2394 "remapping a chunk that had a super mirror inside of it, clearing space cache so we don't end up with corruption");
2395                         mdres->clear_space_cache = 1;
2396                 }
2397                 fs_chunk->physical = mdres->last_physical_offset;
2398                 tree_insert(&mdres->physical_tree, &fs_chunk->p, physical_cmp);
2399                 mdres->last_physical_offset += fs_chunk->bytes;
2400         }
2401 }
2402
2403 static int fixup_devices(struct btrfs_fs_info *fs_info,
2404                          struct mdrestore_struct *mdres, off_t dev_size)
2405 {
2406         struct btrfs_trans_handle *trans;
2407         struct btrfs_dev_item *dev_item;
2408         struct btrfs_path *path;
2409         struct extent_buffer *leaf;
2410         struct btrfs_root *root = fs_info->chunk_root;
2411         struct btrfs_key key;
2412         u64 devid, cur_devid;
2413         int ret;
2414
2415         path = btrfs_alloc_path();
2416         if (!path) {
2417                 error("not enough memory to allocate path");
2418                 return -ENOMEM;
2419         }
2420
2421         trans = btrfs_start_transaction(fs_info->tree_root, 1);
2422         if (IS_ERR(trans)) {
2423                 error("cannot starting transaction %ld", PTR_ERR(trans));
2424                 btrfs_free_path(path);
2425                 return PTR_ERR(trans);
2426         }
2427
2428         dev_item = &fs_info->super_copy->dev_item;
2429
2430         devid = btrfs_stack_device_id(dev_item);
2431
2432         btrfs_set_stack_device_total_bytes(dev_item, dev_size);
2433         btrfs_set_stack_device_bytes_used(dev_item, mdres->alloced_chunks);
2434
2435         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2436         key.type = BTRFS_DEV_ITEM_KEY;
2437         key.offset = 0;
2438
2439 again:
2440         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2441         if (ret < 0) {
2442                 error("search failed: %d", ret);
2443                 exit(1);
2444         }
2445
2446         while (1) {
2447                 leaf = path->nodes[0];
2448                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2449                         ret = btrfs_next_leaf(root, path);
2450                         if (ret < 0) {
2451                                 error("cannot go to next leaf %d", ret);
2452                                 exit(1);
2453                         }
2454                         if (ret > 0) {
2455                                 ret = 0;
2456                                 break;
2457                         }
2458                         leaf = path->nodes[0];
2459                 }
2460
2461                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2462                 if (key.type > BTRFS_DEV_ITEM_KEY)
2463                         break;
2464                 if (key.type != BTRFS_DEV_ITEM_KEY) {
2465                         path->slots[0]++;
2466                         continue;
2467                 }
2468
2469                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2470                                           struct btrfs_dev_item);
2471                 cur_devid = btrfs_device_id(leaf, dev_item);
2472                 if (devid != cur_devid) {
2473                         ret = btrfs_del_item(trans, root, path);
2474                         if (ret) {
2475                                 error("cannot delete item: %d", ret);
2476                                 exit(1);
2477                         }
2478                         btrfs_release_path(path);
2479                         goto again;
2480                 }
2481
2482                 btrfs_set_device_total_bytes(leaf, dev_item, dev_size);
2483                 btrfs_set_device_bytes_used(leaf, dev_item,
2484                                             mdres->alloced_chunks);
2485                 btrfs_mark_buffer_dirty(leaf);
2486                 path->slots[0]++;
2487         }
2488
2489         btrfs_free_path(path);
2490         ret = btrfs_commit_transaction(trans, fs_info->tree_root);
2491         if (ret) {
2492                 error("unable to commit transaction: %d", ret);
2493                 return ret;
2494         }
2495         return 0;
2496 }
2497
2498 static int restore_metadump(const char *input, FILE *out, int old_restore,
2499                             int num_threads, int fixup_offset,
2500                             const char *target, int multi_devices)
2501 {
2502         struct meta_cluster *cluster = NULL;
2503         struct meta_cluster_header *header;
2504         struct mdrestore_struct mdrestore;
2505         struct btrfs_fs_info *info = NULL;
2506         u64 bytenr = 0;
2507         FILE *in = NULL;
2508         int ret = 0;
2509
2510         if (!strcmp(input, "-")) {
2511                 in = stdin;
2512         } else {
2513                 in = fopen(input, "r");
2514                 if (!in) {
2515                         error("unable to open metadump image: %s",
2516                                         strerror(errno));
2517                         return 1;
2518                 }
2519         }
2520
2521         /* NOTE: open with write mode */
2522         if (fixup_offset) {
2523                 BUG_ON(!target);
2524                 info = open_ctree_fs_info(target, 0, 0, 0,
2525                                           OPEN_CTREE_WRITES |
2526                                           OPEN_CTREE_RESTORE |
2527                                           OPEN_CTREE_PARTIAL);
2528                 if (!info) {
2529                         error("open ctree failed");
2530                         ret = -EIO;
2531                         goto failed_open;
2532                 }
2533         }
2534
2535         cluster = malloc(BLOCK_SIZE);
2536         if (!cluster) {
2537                 error("not enough memory for cluster");
2538                 ret = -ENOMEM;
2539                 goto failed_info;
2540         }
2541
2542         ret = mdrestore_init(&mdrestore, in, out, old_restore, num_threads,
2543                              fixup_offset, info, multi_devices);
2544         if (ret) {
2545                 error("failed to intialize metadata restore state: %d", ret);
2546                 goto failed_cluster;
2547         }
2548
2549         if (!multi_devices && !old_restore) {
2550                 ret = build_chunk_tree(&mdrestore, cluster);
2551                 if (ret)
2552                         goto out;
2553                 if (!list_empty(&mdrestore.overlapping_chunks))
2554                         remap_overlapping_chunks(&mdrestore);
2555         }
2556
2557         if (in != stdin && fseek(in, 0, SEEK_SET)) {
2558                 error("seek failed: %s\n", strerror(errno));
2559                 goto out;
2560         }
2561
2562         while (!mdrestore.error) {
2563                 ret = fread(cluster, BLOCK_SIZE, 1, in);
2564                 if (!ret)
2565                         break;
2566
2567                 header = &cluster->header;
2568                 if (le64_to_cpu(header->magic) != HEADER_MAGIC ||
2569                     le64_to_cpu(header->bytenr) != bytenr) {
2570                         error("bad header in metadump image");
2571                         ret = -EIO;
2572                         break;
2573                 }
2574                 ret = add_cluster(cluster, &mdrestore, &bytenr);
2575                 if (ret) {
2576                         error("failed to add cluster: %d", ret);
2577                         break;
2578                 }
2579         }
2580         ret = wait_for_worker(&mdrestore);
2581
2582         if (!ret && !multi_devices && !old_restore) {
2583                 struct btrfs_root *root;
2584                 struct stat st;
2585
2586                 root = open_ctree_fd(fileno(out), target, 0,
2587                                           OPEN_CTREE_PARTIAL |
2588                                           OPEN_CTREE_WRITES |
2589                                           OPEN_CTREE_NO_DEVICES);
2590                 if (!root) {
2591                         error("open ctree failed in %s", target);
2592                         ret = -EIO;
2593                         goto out;
2594                 }
2595                 info = root->fs_info;
2596
2597                 if (stat(target, &st)) {
2598                         error("stat %s failed: %s", target, strerror(errno));
2599                         close_ctree(info->chunk_root);
2600                         free(cluster);
2601                         return 1;
2602                 }
2603
2604                 ret = fixup_devices(info, &mdrestore, st.st_size);
2605                 close_ctree(info->chunk_root);
2606                 if (ret)
2607                         goto out;
2608         }
2609 out:
2610         mdrestore_destroy(&mdrestore, num_threads);
2611 failed_cluster:
2612         free(cluster);
2613 failed_info:
2614         if (fixup_offset && info)
2615                 close_ctree(info->chunk_root);
2616 failed_open:
2617         if (in != stdin)
2618                 fclose(in);
2619         return ret;
2620 }
2621
2622 static int update_disk_super_on_device(struct btrfs_fs_info *info,
2623                                        const char *other_dev, u64 cur_devid)
2624 {
2625         struct btrfs_key key;
2626         struct extent_buffer *leaf;
2627         struct btrfs_path path;
2628         struct btrfs_dev_item *dev_item;
2629         struct btrfs_super_block *disk_super;
2630         char dev_uuid[BTRFS_UUID_SIZE];
2631         char fs_uuid[BTRFS_UUID_SIZE];
2632         u64 devid, type, io_align, io_width;
2633         u64 sector_size, total_bytes, bytes_used;
2634         char buf[BTRFS_SUPER_INFO_SIZE];
2635         int fp = -1;
2636         int ret;
2637
2638         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2639         key.type = BTRFS_DEV_ITEM_KEY;
2640         key.offset = cur_devid;
2641
2642         btrfs_init_path(&path);
2643         ret = btrfs_search_slot(NULL, info->chunk_root, &key, &path, 0, 0);
2644         if (ret) {
2645                 error("search key failed: %d", ret);
2646                 ret = -EIO;
2647                 goto out;
2648         }
2649
2650         leaf = path.nodes[0];
2651         dev_item = btrfs_item_ptr(leaf, path.slots[0],
2652                                   struct btrfs_dev_item);
2653
2654         devid = btrfs_device_id(leaf, dev_item);
2655         if (devid != cur_devid) {
2656                 error("devid mismatch: %llu != %llu",
2657                                 (unsigned long long)devid,
2658                                 (unsigned long long)cur_devid);
2659                 ret = -EIO;
2660                 goto out;
2661         }
2662
2663         type = btrfs_device_type(leaf, dev_item);
2664         io_align = btrfs_device_io_align(leaf, dev_item);
2665         io_width = btrfs_device_io_width(leaf, dev_item);
2666         sector_size = btrfs_device_sector_size(leaf, dev_item);
2667         total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2668         bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2669         read_extent_buffer(leaf, dev_uuid, (unsigned long)btrfs_device_uuid(dev_item), BTRFS_UUID_SIZE);
2670         read_extent_buffer(leaf, fs_uuid, (unsigned long)btrfs_device_fsid(dev_item), BTRFS_UUID_SIZE);
2671
2672         btrfs_release_path(&path);
2673
2674         printf("update disk super on %s devid=%llu\n", other_dev, devid);
2675
2676         /* update other devices' super block */
2677         fp = open(other_dev, O_CREAT | O_RDWR, 0600);
2678         if (fp < 0) {
2679                 error("could not open %s: %s", other_dev, strerror(errno));
2680                 ret = -EIO;
2681                 goto out;
2682         }
2683
2684         memcpy(buf, info->super_copy, BTRFS_SUPER_INFO_SIZE);
2685
2686         disk_super = (struct btrfs_super_block *)buf;
2687         dev_item = &disk_super->dev_item;
2688
2689         btrfs_set_stack_device_type(dev_item, type);
2690         btrfs_set_stack_device_id(dev_item, devid);
2691         btrfs_set_stack_device_total_bytes(dev_item, total_bytes);
2692         btrfs_set_stack_device_bytes_used(dev_item, bytes_used);
2693         btrfs_set_stack_device_io_align(dev_item, io_align);
2694         btrfs_set_stack_device_io_width(dev_item, io_width);
2695         btrfs_set_stack_device_sector_size(dev_item, sector_size);
2696         memcpy(dev_item->uuid, dev_uuid, BTRFS_UUID_SIZE);
2697         memcpy(dev_item->fsid, fs_uuid, BTRFS_UUID_SIZE);
2698         csum_block((u8 *)buf, BTRFS_SUPER_INFO_SIZE);
2699
2700         ret = pwrite64(fp, buf, BTRFS_SUPER_INFO_SIZE, BTRFS_SUPER_INFO_OFFSET);
2701         if (ret != BTRFS_SUPER_INFO_SIZE) {
2702                 if (ret < 0)
2703                         error("cannot write superblock: %s", strerror(ret));
2704                 else
2705                         error("cannot write superblock");
2706                 ret = -EIO;
2707                 goto out;
2708         }
2709
2710         write_backup_supers(fp, (u8 *)buf);
2711
2712 out:
2713         if (fp != -1)
2714                 close(fp);
2715         return ret;
2716 }
2717
2718 static void print_usage(int ret)
2719 {
2720         printf("usage: btrfs-image [options] source target\n");
2721         printf("\t-r      \trestore metadump image\n");
2722         printf("\t-c value\tcompression level (0 ~ 9)\n");
2723         printf("\t-t value\tnumber of threads (1 ~ 32)\n");
2724         printf("\t-o      \tdon't mess with the chunk tree when restoring\n");
2725         printf("\t-s      \tsanitize file names, use once to just use garbage, use twice if you want crc collisions\n");
2726         printf("\t-w      \twalk all trees instead of using extent tree, do this if your extent tree is broken\n");
2727         printf("\t-m       \trestore for multiple devices\n");
2728         printf("\n");
2729         printf("\tIn the dump mode, source is the btrfs device and target is the output file (use '-' for stdout).\n");
2730         printf("\tIn the restore mode, source is the dumped image and target is the btrfs device/file.\n");
2731         exit(ret);
2732 }
2733
2734 int main(int argc, char *argv[])
2735 {
2736         char *source;
2737         char *target;
2738         u64 num_threads = 0;
2739         u64 compress_level = 0;
2740         int create = 1;
2741         int old_restore = 0;
2742         int walk_trees = 0;
2743         int multi_devices = 0;
2744         int ret;
2745         int sanitize = 0;
2746         int dev_cnt = 0;
2747         int usage_error = 0;
2748         FILE *out;
2749
2750         while (1) {
2751                 static const struct option long_options[] = {
2752                         { "help", no_argument, NULL, GETOPT_VAL_HELP},
2753                         { NULL, 0, NULL, 0 }
2754                 };
2755                 int c = getopt_long(argc, argv, "rc:t:oswm", long_options, NULL);
2756                 if (c < 0)
2757                         break;
2758                 switch (c) {
2759                 case 'r':
2760                         create = 0;
2761                         break;
2762                 case 't':
2763                         num_threads = arg_strtou64(optarg);
2764                         if (num_threads > 32) {
2765                                 error("number of threads out of range: %llu",
2766                                         (unsigned long long)num_threads);
2767                                 return 1;
2768                         }
2769                         break;
2770                 case 'c':
2771                         compress_level = arg_strtou64(optarg);
2772                         if (compress_level > 9) {
2773                                 error("compression level out of range: %llu",
2774                                         (unsigned long long)compress_level);
2775                                 return 1;
2776                         }
2777                         break;
2778                 case 'o':
2779                         old_restore = 1;
2780                         break;
2781                 case 's':
2782                         sanitize++;
2783                         break;
2784                 case 'w':
2785                         walk_trees = 1;
2786                         break;
2787                 case 'm':
2788                         create = 0;
2789                         multi_devices = 1;
2790                         break;
2791                         case GETOPT_VAL_HELP:
2792                 default:
2793                         print_usage(c != GETOPT_VAL_HELP);
2794                 }
2795         }
2796
2797         set_argv0(argv);
2798         if (check_argc_min(argc - optind, 2))
2799                 print_usage(1);
2800
2801         dev_cnt = argc - optind - 1;
2802
2803         if (create) {
2804                 if (old_restore) {
2805                         error(
2806                         "create and restore cannot be used at the same time");
2807                         usage_error++;
2808                 }
2809         } else {
2810                 if (walk_trees || sanitize || compress_level) {
2811                         error(
2812                         "useing -w, -s, -c options for restore makes no sense");
2813                         usage_error++;
2814                 }
2815                 if (multi_devices && dev_cnt < 2) {
2816                         error("not enough devices specified for -m option");
2817                         usage_error++;
2818                 }
2819                 if (!multi_devices && dev_cnt != 1) {
2820                         error("accepts only 1 device without -m option");
2821                         usage_error++;
2822                 }
2823         }
2824
2825         if (usage_error)
2826                 print_usage(1);
2827
2828         source = argv[optind];
2829         target = argv[optind + 1];
2830
2831         if (create && !strcmp(target, "-")) {
2832                 out = stdout;
2833         } else {
2834                 out = fopen(target, "w+");
2835                 if (!out) {
2836                         error("unable to create target file %s", target);
2837                         exit(1);
2838                 }
2839         }
2840
2841         if (compress_level > 0 || create == 0) {
2842                 if (num_threads == 0) {
2843                         long tmp = sysconf(_SC_NPROCESSORS_ONLN);
2844
2845                         if (tmp <= 0)
2846                                 tmp = 1;
2847                         num_threads = tmp;
2848                 }
2849         } else {
2850                 num_threads = 0;
2851         }
2852
2853         if (create) {
2854                 ret = check_mounted(source);
2855                 if (ret < 0) {
2856                         warning("unable to check mount status of: %s",
2857                                         strerror(-ret));
2858                 } else if (ret) {
2859                         warning("%s already mounted, results may be inaccurate",
2860                                         source);
2861                 }
2862
2863                 ret = create_metadump(source, out, num_threads,
2864                                       compress_level, sanitize, walk_trees);
2865         } else {
2866                 ret = restore_metadump(source, out, old_restore, num_threads,
2867                                        0, target, multi_devices);
2868         }
2869         if (ret) {
2870                 error("%s failed: %s", (create) ? "create" : "restore",
2871                        strerror(errno));
2872                 goto out;
2873         }
2874
2875          /* extended support for multiple devices */
2876         if (!create && multi_devices) {
2877                 struct btrfs_fs_info *info;
2878                 u64 total_devs;
2879                 int i;
2880
2881                 info = open_ctree_fs_info(target, 0, 0, 0,
2882                                           OPEN_CTREE_PARTIAL |
2883                                           OPEN_CTREE_RESTORE);
2884                 if (!info) {
2885                         error("open ctree failed at %s", target);
2886                         return 1;
2887                 }
2888
2889                 total_devs = btrfs_super_num_devices(info->super_copy);
2890                 if (total_devs != dev_cnt) {
2891                         error("it needs %llu devices but has only %d",
2892                                 total_devs, dev_cnt);
2893                         close_ctree(info->chunk_root);
2894                         goto out;
2895                 }
2896
2897                 /* update super block on other disks */
2898                 for (i = 2; i <= dev_cnt; i++) {
2899                         ret = update_disk_super_on_device(info,
2900                                         argv[optind + i], (u64)i);
2901                         if (ret) {
2902                                 error("update disk superblock failed devid %d: %d",
2903                                         i, ret);
2904                                 close_ctree(info->chunk_root);
2905                                 exit(1);
2906                         }
2907                 }
2908
2909                 close_ctree(info->chunk_root);
2910
2911                 /* fix metadata block to map correct chunk */
2912                 ret = restore_metadump(source, out, 0, num_threads, 1,
2913                                        target, 1);
2914                 if (ret) {
2915                         error("unable to fixup metadump: %d", ret);
2916                         exit(1);
2917                 }
2918         }
2919 out:
2920         if (out == stdout) {
2921                 fflush(out);
2922         } else {
2923                 fclose(out);
2924                 if (ret && create) {
2925                         int unlink_ret;
2926
2927                         unlink_ret = unlink(target);
2928                         if (unlink_ret)
2929                                 error("unlink output file %s failed: %s",
2930                                                 target, strerror(errno));
2931                 }
2932         }
2933
2934         btrfs_close_all_devices();
2935
2936         return !!ret;
2937 }