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