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 nodesize;
 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 allocating 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->nodesize, 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->nodesize, 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->nodesize, 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->nodesize;
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         ret = metadump_init(&metadump, root, out, num_threads,
1325                             compress_level, sanitize);
1326         if (ret) {
1327                 fprintf(stderr, "Error initializing metadump %d\n", ret);
1328                 close_ctree(root);
1329                 return ret;
1330         }
1331
1332         ret = add_extent(BTRFS_SUPER_INFO_OFFSET, BTRFS_SUPER_INFO_SIZE,
1333                         &metadump, 0);
1334         if (ret) {
1335                 fprintf(stderr, "Error adding metadata %d\n", ret);
1336                 err = ret;
1337                 goto out;
1338         }
1339
1340         path = btrfs_alloc_path();
1341         if (!path) {
1342                 fprintf(stderr, "Out of memory allocating path\n");
1343                 err = -ENOMEM;
1344                 goto out;
1345         }
1346
1347         if (walk_trees) {
1348                 ret = copy_tree_blocks(root, root->fs_info->chunk_root->node,
1349                                        &metadump, 1);
1350                 if (ret) {
1351                         err = ret;
1352                         goto out;
1353                 }
1354
1355                 ret = copy_tree_blocks(root, root->fs_info->tree_root->node,
1356                                        &metadump, 1);
1357                 if (ret) {
1358                         err = ret;
1359                         goto out;
1360                 }
1361         } else {
1362                 ret = copy_from_extent_tree(&metadump, path);
1363                 if (ret) {
1364                         err = ret;
1365                         goto out;
1366                 }
1367         }
1368
1369         ret = copy_log_trees(root, &metadump, path);
1370         if (ret) {
1371                 err = ret;
1372                 goto out;
1373         }
1374
1375         ret = copy_space_cache(root, &metadump, path);
1376 out:
1377         ret = flush_pending(&metadump, 1);
1378         if (ret) {
1379                 if (!err)
1380                         err = ret;
1381                 fprintf(stderr, "Error flushing pending %d\n", ret);
1382         }
1383
1384         metadump_destroy(&metadump, num_threads);
1385
1386         btrfs_free_path(path);
1387         ret = close_ctree(root);
1388         return err ? err : ret;
1389 }
1390
1391 static void update_super_old(u8 *buffer)
1392 {
1393         struct btrfs_super_block *super = (struct btrfs_super_block *)buffer;
1394         struct btrfs_chunk *chunk;
1395         struct btrfs_disk_key *key;
1396         u32 sectorsize = btrfs_super_sectorsize(super);
1397         u64 flags = btrfs_super_flags(super);
1398
1399         flags |= BTRFS_SUPER_FLAG_METADUMP;
1400         btrfs_set_super_flags(super, flags);
1401
1402         key = (struct btrfs_disk_key *)(super->sys_chunk_array);
1403         chunk = (struct btrfs_chunk *)(super->sys_chunk_array +
1404                                        sizeof(struct btrfs_disk_key));
1405
1406         btrfs_set_disk_key_objectid(key, BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1407         btrfs_set_disk_key_type(key, BTRFS_CHUNK_ITEM_KEY);
1408         btrfs_set_disk_key_offset(key, 0);
1409
1410         btrfs_set_stack_chunk_length(chunk, (u64)-1);
1411         btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
1412         btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
1413         btrfs_set_stack_chunk_type(chunk, BTRFS_BLOCK_GROUP_SYSTEM);
1414         btrfs_set_stack_chunk_io_align(chunk, sectorsize);
1415         btrfs_set_stack_chunk_io_width(chunk, sectorsize);
1416         btrfs_set_stack_chunk_sector_size(chunk, sectorsize);
1417         btrfs_set_stack_chunk_num_stripes(chunk, 1);
1418         btrfs_set_stack_chunk_sub_stripes(chunk, 0);
1419         chunk->stripe.devid = super->dev_item.devid;
1420         btrfs_set_stack_stripe_offset(&chunk->stripe, 0);
1421         memcpy(chunk->stripe.dev_uuid, super->dev_item.uuid, BTRFS_UUID_SIZE);
1422         btrfs_set_super_sys_array_size(super, sizeof(*key) + sizeof(*chunk));
1423         csum_block(buffer, BTRFS_SUPER_INFO_SIZE);
1424 }
1425
1426 static int update_super(struct mdrestore_struct *mdres, u8 *buffer)
1427 {
1428         struct btrfs_super_block *super = (struct btrfs_super_block *)buffer;
1429         struct btrfs_chunk *chunk;
1430         struct btrfs_disk_key *disk_key;
1431         struct btrfs_key key;
1432         u64 flags = btrfs_super_flags(super);
1433         u32 new_array_size = 0;
1434         u32 array_size;
1435         u32 cur = 0;
1436         u8 *ptr, *write_ptr;
1437         int old_num_stripes;
1438
1439         write_ptr = ptr = super->sys_chunk_array;
1440         array_size = btrfs_super_sys_array_size(super);
1441
1442         while (cur < array_size) {
1443                 disk_key = (struct btrfs_disk_key *)ptr;
1444                 btrfs_disk_key_to_cpu(&key, disk_key);
1445
1446                 new_array_size += sizeof(*disk_key);
1447                 memmove(write_ptr, ptr, sizeof(*disk_key));
1448
1449                 write_ptr += sizeof(*disk_key);
1450                 ptr += sizeof(*disk_key);
1451                 cur += sizeof(*disk_key);
1452
1453                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1454                         u64 physical, size = 0;
1455
1456                         chunk = (struct btrfs_chunk *)ptr;
1457                         old_num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1458                         chunk = (struct btrfs_chunk *)write_ptr;
1459
1460                         memmove(write_ptr, ptr, sizeof(*chunk));
1461                         btrfs_set_stack_chunk_num_stripes(chunk, 1);
1462                         btrfs_set_stack_chunk_sub_stripes(chunk, 0);
1463                         btrfs_set_stack_chunk_type(chunk,
1464                                                    BTRFS_BLOCK_GROUP_SYSTEM);
1465                         chunk->stripe.devid = super->dev_item.devid;
1466                         physical = logical_to_physical(mdres, key.offset,
1467                                                        &size);
1468                         if (size != (u64)-1)
1469                                 btrfs_set_stack_stripe_offset(&chunk->stripe,
1470                                                               physical);
1471                         memcpy(chunk->stripe.dev_uuid, super->dev_item.uuid,
1472                                BTRFS_UUID_SIZE);
1473                         new_array_size += sizeof(*chunk);
1474                 } else {
1475                         fprintf(stderr, "Bogus key in the sys chunk array "
1476                                 "%d\n", key.type);
1477                         return -EIO;
1478                 }
1479                 write_ptr += sizeof(*chunk);
1480                 ptr += btrfs_chunk_item_size(old_num_stripes);
1481                 cur += btrfs_chunk_item_size(old_num_stripes);
1482         }
1483
1484         if (mdres->clear_space_cache)
1485                 btrfs_set_super_cache_generation(super, 0);
1486
1487         flags |= BTRFS_SUPER_FLAG_METADUMP_V2;
1488         btrfs_set_super_flags(super, flags);
1489         btrfs_set_super_sys_array_size(super, new_array_size);
1490         csum_block(buffer, BTRFS_SUPER_INFO_SIZE);
1491
1492         return 0;
1493 }
1494
1495 static struct extent_buffer *alloc_dummy_eb(u64 bytenr, u32 size)
1496 {
1497         struct extent_buffer *eb;
1498
1499         eb = calloc(1, sizeof(struct extent_buffer) + size);
1500         if (!eb)
1501                 return NULL;
1502
1503         eb->start = bytenr;
1504         eb->len = size;
1505         return eb;
1506 }
1507
1508 static void truncate_item(struct extent_buffer *eb, int slot, u32 new_size)
1509 {
1510         struct btrfs_item *item;
1511         u32 nritems;
1512         u32 old_size;
1513         u32 old_data_start;
1514         u32 size_diff;
1515         u32 data_end;
1516         int i;
1517
1518         old_size = btrfs_item_size_nr(eb, slot);
1519         if (old_size == new_size)
1520                 return;
1521
1522         nritems = btrfs_header_nritems(eb);
1523         data_end = btrfs_item_offset_nr(eb, nritems - 1);
1524
1525         old_data_start = btrfs_item_offset_nr(eb, slot);
1526         size_diff = old_size - new_size;
1527
1528         for (i = slot; i < nritems; i++) {
1529                 u32 ioff;
1530                 item = btrfs_item_nr(i);
1531                 ioff = btrfs_item_offset(eb, item);
1532                 btrfs_set_item_offset(eb, item, ioff + size_diff);
1533         }
1534
1535         memmove_extent_buffer(eb, btrfs_leaf_data(eb) + data_end + size_diff,
1536                               btrfs_leaf_data(eb) + data_end,
1537                               old_data_start + new_size - data_end);
1538         item = btrfs_item_nr(slot);
1539         btrfs_set_item_size(eb, item, new_size);
1540 }
1541
1542 static int fixup_chunk_tree_block(struct mdrestore_struct *mdres,
1543                                   struct async_work *async, u8 *buffer,
1544                                   size_t size)
1545 {
1546         struct extent_buffer *eb;
1547         size_t size_left = size;
1548         u64 bytenr = async->start;
1549         int i;
1550
1551         if (size_left % mdres->nodesize)
1552                 return 0;
1553
1554         eb = alloc_dummy_eb(bytenr, mdres->nodesize);
1555         if (!eb)
1556                 return -ENOMEM;
1557
1558         while (size_left) {
1559                 eb->start = bytenr;
1560                 memcpy(eb->data, buffer, mdres->nodesize);
1561
1562                 if (btrfs_header_bytenr(eb) != bytenr)
1563                         break;
1564                 if (memcmp(mdres->fsid,
1565                            eb->data + offsetof(struct btrfs_header, fsid),
1566                            BTRFS_FSID_SIZE))
1567                         break;
1568
1569                 if (btrfs_header_owner(eb) != BTRFS_CHUNK_TREE_OBJECTID)
1570                         goto next;
1571
1572                 if (btrfs_header_level(eb) != 0)
1573                         goto next;
1574
1575                 for (i = 0; i < btrfs_header_nritems(eb); i++) {
1576                         struct btrfs_chunk chunk;
1577                         struct btrfs_key key;
1578                         u64 type, physical, size = (u64)-1;
1579
1580                         btrfs_item_key_to_cpu(eb, &key, i);
1581                         if (key.type != BTRFS_CHUNK_ITEM_KEY)
1582                                 continue;
1583                         truncate_item(eb, i, sizeof(chunk));
1584                         read_extent_buffer(eb, &chunk,
1585                                            btrfs_item_ptr_offset(eb, i),
1586                                            sizeof(chunk));
1587
1588                         size = 0;
1589                         physical = logical_to_physical(mdres, key.offset,
1590                                                        &size);
1591
1592                         /* Zero out the RAID profile */
1593                         type = btrfs_stack_chunk_type(&chunk);
1594                         type &= (BTRFS_BLOCK_GROUP_DATA |
1595                                  BTRFS_BLOCK_GROUP_SYSTEM |
1596                                  BTRFS_BLOCK_GROUP_METADATA |
1597                                  BTRFS_BLOCK_GROUP_DUP);
1598                         btrfs_set_stack_chunk_type(&chunk, type);
1599
1600                         btrfs_set_stack_chunk_num_stripes(&chunk, 1);
1601                         btrfs_set_stack_chunk_sub_stripes(&chunk, 0);
1602                         btrfs_set_stack_stripe_devid(&chunk.stripe, mdres->devid);
1603                         if (size != (u64)-1)
1604                                 btrfs_set_stack_stripe_offset(&chunk.stripe,
1605                                                               physical);
1606                         memcpy(chunk.stripe.dev_uuid, mdres->uuid,
1607                                BTRFS_UUID_SIZE);
1608                         write_extent_buffer(eb, &chunk,
1609                                             btrfs_item_ptr_offset(eb, i),
1610                                             sizeof(chunk));
1611                 }
1612                 memcpy(buffer, eb->data, eb->len);
1613                 csum_block(buffer, eb->len);
1614 next:
1615                 size_left -= mdres->nodesize;
1616                 buffer += mdres->nodesize;
1617                 bytenr += mdres->nodesize;
1618         }
1619
1620         free(eb);
1621         return 0;
1622 }
1623
1624 static void write_backup_supers(int fd, u8 *buf)
1625 {
1626         struct btrfs_super_block *super = (struct btrfs_super_block *)buf;
1627         struct stat st;
1628         u64 size;
1629         u64 bytenr;
1630         int i;
1631         int ret;
1632
1633         if (fstat(fd, &st)) {
1634                 fprintf(stderr, "Couldn't stat restore point, won't be able "
1635                         "to write backup supers: %d\n", errno);
1636                 return;
1637         }
1638
1639         size = btrfs_device_size(fd, &st);
1640
1641         for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1642                 bytenr = btrfs_sb_offset(i);
1643                 if (bytenr + BTRFS_SUPER_INFO_SIZE > size)
1644                         break;
1645                 btrfs_set_super_bytenr(super, bytenr);
1646                 csum_block(buf, BTRFS_SUPER_INFO_SIZE);
1647                 ret = pwrite64(fd, buf, BTRFS_SUPER_INFO_SIZE, bytenr);
1648                 if (ret < BTRFS_SUPER_INFO_SIZE) {
1649                         if (ret < 0)
1650                                 fprintf(stderr, "Problem writing out backup "
1651                                         "super block %d, err %d\n", i, errno);
1652                         else
1653                                 fprintf(stderr, "Short write writing out "
1654                                         "backup super block\n");
1655                         break;
1656                 }
1657         }
1658 }
1659
1660 static void *restore_worker(void *data)
1661 {
1662         struct mdrestore_struct *mdres = (struct mdrestore_struct *)data;
1663         struct async_work *async;
1664         size_t size;
1665         u8 *buffer;
1666         u8 *outbuf;
1667         int outfd;
1668         int ret;
1669         int compress_size = MAX_PENDING_SIZE * 4;
1670
1671         outfd = fileno(mdres->out);
1672         buffer = malloc(compress_size);
1673         if (!buffer) {
1674                 fprintf(stderr, "Error allocating buffer\n");
1675                 pthread_mutex_lock(&mdres->mutex);
1676                 if (!mdres->error)
1677                         mdres->error = -ENOMEM;
1678                 pthread_mutex_unlock(&mdres->mutex);
1679                 pthread_exit(NULL);
1680         }
1681
1682         while (1) {
1683                 u64 bytenr;
1684                 off_t offset = 0;
1685                 int err = 0;
1686
1687                 pthread_mutex_lock(&mdres->mutex);
1688                 while (!mdres->nodesize || list_empty(&mdres->list)) {
1689                         if (mdres->done) {
1690                                 pthread_mutex_unlock(&mdres->mutex);
1691                                 goto out;
1692                         }
1693                         pthread_cond_wait(&mdres->cond, &mdres->mutex);
1694                 }
1695                 async = list_entry(mdres->list.next, struct async_work, list);
1696                 list_del_init(&async->list);
1697                 pthread_mutex_unlock(&mdres->mutex);
1698
1699                 if (mdres->compress_method == COMPRESS_ZLIB) {
1700                         size = compress_size;
1701                         ret = uncompress(buffer, (unsigned long *)&size,
1702                                          async->buffer, async->bufsize);
1703                         if (ret != Z_OK) {
1704                                 fprintf(stderr, "Error decompressing %d\n",
1705                                         ret);
1706                                 err = -EIO;
1707                         }
1708                         outbuf = buffer;
1709                 } else {
1710                         outbuf = async->buffer;
1711                         size = async->bufsize;
1712                 }
1713
1714                 if (!mdres->multi_devices) {
1715                         if (async->start == BTRFS_SUPER_INFO_OFFSET) {
1716                                 if (mdres->old_restore) {
1717                                         update_super_old(outbuf);
1718                                 } else {
1719                                         ret = update_super(mdres, outbuf);
1720                                         if (ret)
1721                                                 err = ret;
1722                                 }
1723                         } else if (!mdres->old_restore) {
1724                                 ret = fixup_chunk_tree_block(mdres, async, outbuf, size);
1725                                 if (ret)
1726                                         err = ret;
1727                         }
1728                 }
1729
1730                 if (!mdres->fixup_offset) {
1731                         while (size) {
1732                                 u64 chunk_size = size;
1733                                 if (!mdres->multi_devices && !mdres->old_restore)
1734                                         bytenr = logical_to_physical(mdres,
1735                                                                      async->start + offset,
1736                                                                      &chunk_size);
1737                                 else
1738                                         bytenr = async->start + offset;
1739
1740                                 ret = pwrite64(outfd, outbuf+offset, chunk_size,
1741                                                bytenr);
1742                                 if (ret != chunk_size) {
1743                                         if (ret < 0) {
1744                                                 fprintf(stderr, "Error writing to "
1745                                                         "device %d\n", errno);
1746                                                 err = errno;
1747                                                 break;
1748                                         } else {
1749                                                 fprintf(stderr, "Short write\n");
1750                                                 err = -EIO;
1751                                                 break;
1752                                         }
1753                                 }
1754                                 size -= chunk_size;
1755                                 offset += chunk_size;
1756                         }
1757                 } else if (async->start != BTRFS_SUPER_INFO_OFFSET) {
1758                         ret = write_data_to_disk(mdres->info, outbuf, async->start, size, 0);
1759                         if (ret) {
1760                                 printk("Error write data\n");
1761                                 exit(1);
1762                         }
1763                 }
1764
1765
1766                 /* backup super blocks are already there at fixup_offset stage */
1767                 if (!mdres->multi_devices && async->start == BTRFS_SUPER_INFO_OFFSET)
1768                         write_backup_supers(outfd, outbuf);
1769
1770                 pthread_mutex_lock(&mdres->mutex);
1771                 if (err && !mdres->error)
1772                         mdres->error = err;
1773                 mdres->num_items--;
1774                 pthread_mutex_unlock(&mdres->mutex);
1775
1776                 free(async->buffer);
1777                 free(async);
1778         }
1779 out:
1780         free(buffer);
1781         pthread_exit(NULL);
1782 }
1783
1784 static void mdrestore_destroy(struct mdrestore_struct *mdres, int num_threads)
1785 {
1786         struct rb_node *n;
1787         int i;
1788
1789         while ((n = rb_first(&mdres->chunk_tree))) {
1790                 struct fs_chunk *entry;
1791
1792                 entry = rb_entry(n, struct fs_chunk, l);
1793                 rb_erase(n, &mdres->chunk_tree);
1794                 rb_erase(&entry->p, &mdres->physical_tree);
1795                 free(entry);
1796         }
1797         pthread_mutex_lock(&mdres->mutex);
1798         mdres->done = 1;
1799         pthread_cond_broadcast(&mdres->cond);
1800         pthread_mutex_unlock(&mdres->mutex);
1801
1802         for (i = 0; i < num_threads; i++)
1803                 pthread_join(mdres->threads[i], NULL);
1804
1805         pthread_cond_destroy(&mdres->cond);
1806         pthread_mutex_destroy(&mdres->mutex);
1807         free(mdres->threads);
1808 }
1809
1810 static int mdrestore_init(struct mdrestore_struct *mdres,
1811                           FILE *in, FILE *out, int old_restore,
1812                           int num_threads, int fixup_offset,
1813                           struct btrfs_fs_info *info, int multi_devices)
1814 {
1815         int i, ret = 0;
1816
1817         memset(mdres, 0, sizeof(*mdres));
1818         pthread_cond_init(&mdres->cond, NULL);
1819         pthread_mutex_init(&mdres->mutex, NULL);
1820         INIT_LIST_HEAD(&mdres->list);
1821         INIT_LIST_HEAD(&mdres->overlapping_chunks);
1822         mdres->in = in;
1823         mdres->out = out;
1824         mdres->old_restore = old_restore;
1825         mdres->chunk_tree.rb_node = NULL;
1826         mdres->fixup_offset = fixup_offset;
1827         mdres->info = info;
1828         mdres->multi_devices = multi_devices;
1829         mdres->clear_space_cache = 0;
1830         mdres->last_physical_offset = 0;
1831         mdres->alloced_chunks = 0;
1832
1833         if (!num_threads)
1834                 return 0;
1835
1836         mdres->num_threads = num_threads;
1837         mdres->threads = calloc(num_threads, sizeof(pthread_t));
1838         if (!mdres->threads)
1839                 return -ENOMEM;
1840         for (i = 0; i < num_threads; i++) {
1841                 ret = pthread_create(mdres->threads + i, NULL, restore_worker,
1842                                      mdres);
1843                 if (ret)
1844                         break;
1845         }
1846         if (ret)
1847                 mdrestore_destroy(mdres, i + 1);
1848         return ret;
1849 }
1850
1851 static int fill_mdres_info(struct mdrestore_struct *mdres,
1852                            struct async_work *async)
1853 {
1854         struct btrfs_super_block *super;
1855         u8 *buffer = NULL;
1856         u8 *outbuf;
1857         int ret;
1858
1859         /* We've already been initialized */
1860         if (mdres->nodesize)
1861                 return 0;
1862
1863         if (mdres->compress_method == COMPRESS_ZLIB) {
1864                 size_t size = MAX_PENDING_SIZE * 2;
1865
1866                 buffer = malloc(MAX_PENDING_SIZE * 2);
1867                 if (!buffer)
1868                         return -ENOMEM;
1869                 ret = uncompress(buffer, (unsigned long *)&size,
1870                                  async->buffer, async->bufsize);
1871                 if (ret != Z_OK) {
1872                         fprintf(stderr, "Error decompressing %d\n", ret);
1873                         free(buffer);
1874                         return -EIO;
1875                 }
1876                 outbuf = buffer;
1877         } else {
1878                 outbuf = async->buffer;
1879         }
1880
1881         super = (struct btrfs_super_block *)outbuf;
1882         mdres->nodesize = btrfs_super_nodesize(super);
1883         memcpy(mdres->fsid, super->fsid, BTRFS_FSID_SIZE);
1884         memcpy(mdres->uuid, super->dev_item.uuid,
1885                        BTRFS_UUID_SIZE);
1886         mdres->devid = le64_to_cpu(super->dev_item.devid);
1887         free(buffer);
1888         return 0;
1889 }
1890
1891 static int add_cluster(struct meta_cluster *cluster,
1892                        struct mdrestore_struct *mdres, u64 *next)
1893 {
1894         struct meta_cluster_item *item;
1895         struct meta_cluster_header *header = &cluster->header;
1896         struct async_work *async;
1897         u64 bytenr;
1898         u32 i, nritems;
1899         int ret;
1900
1901         mdres->compress_method = header->compress;
1902
1903         bytenr = le64_to_cpu(header->bytenr) + BLOCK_SIZE;
1904         nritems = le32_to_cpu(header->nritems);
1905         for (i = 0; i < nritems; i++) {
1906                 item = &cluster->items[i];
1907                 async = calloc(1, sizeof(*async));
1908                 if (!async) {
1909                         fprintf(stderr, "Error allocating async\n");
1910                         return -ENOMEM;
1911                 }
1912                 async->start = le64_to_cpu(item->bytenr);
1913                 async->bufsize = le32_to_cpu(item->size);
1914                 async->buffer = malloc(async->bufsize);
1915                 if (!async->buffer) {
1916                         fprintf(stderr, "Error allocating async buffer\n");
1917                         free(async);
1918                         return -ENOMEM;
1919                 }
1920                 ret = fread(async->buffer, async->bufsize, 1, mdres->in);
1921                 if (ret != 1) {
1922                         fprintf(stderr, "Error reading buffer %d\n", errno);
1923                         free(async->buffer);
1924                         free(async);
1925                         return -EIO;
1926                 }
1927                 bytenr += async->bufsize;
1928
1929                 pthread_mutex_lock(&mdres->mutex);
1930                 if (async->start == BTRFS_SUPER_INFO_OFFSET) {
1931                         ret = fill_mdres_info(mdres, async);
1932                         if (ret) {
1933                                 fprintf(stderr, "Error setting up restore\n");
1934                                 pthread_mutex_unlock(&mdres->mutex);
1935                                 free(async->buffer);
1936                                 free(async);
1937                                 return ret;
1938                         }
1939                 }
1940                 list_add_tail(&async->list, &mdres->list);
1941                 mdres->num_items++;
1942                 pthread_cond_signal(&mdres->cond);
1943                 pthread_mutex_unlock(&mdres->mutex);
1944         }
1945         if (bytenr & BLOCK_MASK) {
1946                 char buffer[BLOCK_MASK];
1947                 size_t size = BLOCK_SIZE - (bytenr & BLOCK_MASK);
1948
1949                 bytenr += size;
1950                 ret = fread(buffer, size, 1, mdres->in);
1951                 if (ret != 1) {
1952                         fprintf(stderr, "Error reading in buffer %d\n", errno);
1953                         return -EIO;
1954                 }
1955         }
1956         *next = bytenr;
1957         return 0;
1958 }
1959
1960 static int wait_for_worker(struct mdrestore_struct *mdres)
1961 {
1962         int ret = 0;
1963
1964         pthread_mutex_lock(&mdres->mutex);
1965         ret = mdres->error;
1966         while (!ret && mdres->num_items > 0) {
1967                 struct timespec ts = {
1968                         .tv_sec = 0,
1969                         .tv_nsec = 10000000,
1970                 };
1971                 pthread_mutex_unlock(&mdres->mutex);
1972                 nanosleep(&ts, NULL);
1973                 pthread_mutex_lock(&mdres->mutex);
1974                 ret = mdres->error;
1975         }
1976         pthread_mutex_unlock(&mdres->mutex);
1977         return ret;
1978 }
1979
1980 static int read_chunk_block(struct mdrestore_struct *mdres, u8 *buffer,
1981                             u64 bytenr, u64 item_bytenr, u32 bufsize,
1982                             u64 cluster_bytenr)
1983 {
1984         struct extent_buffer *eb;
1985         int ret = 0;
1986         int i;
1987
1988         eb = alloc_dummy_eb(bytenr, mdres->nodesize);
1989         if (!eb) {
1990                 ret = -ENOMEM;
1991                 goto out;
1992         }
1993
1994         while (item_bytenr != bytenr) {
1995                 buffer += mdres->nodesize;
1996                 item_bytenr += mdres->nodesize;
1997         }
1998
1999         memcpy(eb->data, buffer, mdres->nodesize);
2000         if (btrfs_header_bytenr(eb) != bytenr) {
2001                 fprintf(stderr, "Eb bytenr doesn't match found bytenr\n");
2002                 ret = -EIO;
2003                 goto out;
2004         }
2005
2006         if (memcmp(mdres->fsid, eb->data + offsetof(struct btrfs_header, fsid),
2007                    BTRFS_FSID_SIZE)) {
2008                 fprintf(stderr, "Fsid doesn't match\n");
2009                 ret = -EIO;
2010                 goto out;
2011         }
2012
2013         if (btrfs_header_owner(eb) != BTRFS_CHUNK_TREE_OBJECTID) {
2014                 fprintf(stderr, "Does not belong to the chunk tree\n");
2015                 ret = -EIO;
2016                 goto out;
2017         }
2018
2019         for (i = 0; i < btrfs_header_nritems(eb); i++) {
2020                 struct btrfs_chunk chunk;
2021                 struct fs_chunk *fs_chunk;
2022                 struct btrfs_key key;
2023
2024                 if (btrfs_header_level(eb)) {
2025                         u64 blockptr = btrfs_node_blockptr(eb, i);
2026
2027                         ret = search_for_chunk_blocks(mdres, blockptr,
2028                                                       cluster_bytenr);
2029                         if (ret)
2030                                 break;
2031                         continue;
2032                 }
2033
2034                 /* Yay a leaf!  We loves leafs! */
2035                 btrfs_item_key_to_cpu(eb, &key, i);
2036                 if (key.type != BTRFS_CHUNK_ITEM_KEY)
2037                         continue;
2038
2039                 fs_chunk = malloc(sizeof(struct fs_chunk));
2040                 if (!fs_chunk) {
2041                         fprintf(stderr, "Error allocating chunk\n");
2042                         ret = -ENOMEM;
2043                         break;
2044                 }
2045                 memset(fs_chunk, 0, sizeof(*fs_chunk));
2046                 read_extent_buffer(eb, &chunk, btrfs_item_ptr_offset(eb, i),
2047                                    sizeof(chunk));
2048
2049                 fs_chunk->logical = key.offset;
2050                 fs_chunk->physical = btrfs_stack_stripe_offset(&chunk.stripe);
2051                 fs_chunk->bytes = btrfs_stack_chunk_length(&chunk);
2052                 INIT_LIST_HEAD(&fs_chunk->list);
2053                 if (tree_search(&mdres->physical_tree, &fs_chunk->p,
2054                                 physical_cmp, 1) != NULL)
2055                         list_add(&fs_chunk->list, &mdres->overlapping_chunks);
2056                 else
2057                         tree_insert(&mdres->physical_tree, &fs_chunk->p,
2058                                     physical_cmp);
2059                 if (fs_chunk->physical + fs_chunk->bytes >
2060                     mdres->last_physical_offset)
2061                         mdres->last_physical_offset = fs_chunk->physical +
2062                                 fs_chunk->bytes;
2063                 mdres->alloced_chunks += fs_chunk->bytes;
2064                 tree_insert(&mdres->chunk_tree, &fs_chunk->l, chunk_cmp);
2065         }
2066 out:
2067         free(eb);
2068         return ret;
2069 }
2070
2071 /* If you have to ask you aren't worthy */
2072 static int search_for_chunk_blocks(struct mdrestore_struct *mdres,
2073                                    u64 search, u64 cluster_bytenr)
2074 {
2075         struct meta_cluster *cluster;
2076         struct meta_cluster_header *header;
2077         struct meta_cluster_item *item;
2078         u64 current_cluster = cluster_bytenr, bytenr;
2079         u64 item_bytenr;
2080         u32 bufsize, nritems, i;
2081         u32 max_size = MAX_PENDING_SIZE * 2;
2082         u8 *buffer, *tmp = NULL;
2083         int ret = 0;
2084
2085         cluster = malloc(BLOCK_SIZE);
2086         if (!cluster) {
2087                 fprintf(stderr, "Error allocating cluster\n");
2088                 return -ENOMEM;
2089         }
2090
2091         buffer = malloc(max_size);
2092         if (!buffer) {
2093                 fprintf(stderr, "Error allocating buffer\n");
2094                 free(cluster);
2095                 return -ENOMEM;
2096         }
2097
2098         if (mdres->compress_method == COMPRESS_ZLIB) {
2099                 tmp = malloc(max_size);
2100                 if (!tmp) {
2101                         fprintf(stderr, "Error allocating tmp buffer\n");
2102                         free(cluster);
2103                         free(buffer);
2104                         return -ENOMEM;
2105                 }
2106         }
2107
2108         bytenr = current_cluster;
2109         while (1) {
2110                 if (fseek(mdres->in, current_cluster, SEEK_SET)) {
2111                         fprintf(stderr, "Error seeking: %d\n", errno);
2112                         ret = -EIO;
2113                         break;
2114                 }
2115
2116                 ret = fread(cluster, BLOCK_SIZE, 1, mdres->in);
2117                 if (ret == 0) {
2118                         if (cluster_bytenr != 0) {
2119                                 cluster_bytenr = 0;
2120                                 current_cluster = 0;
2121                                 bytenr = 0;
2122                                 continue;
2123                         }
2124                         printf("ok this is where we screwed up?\n");
2125                         ret = -EIO;
2126                         break;
2127                 } else if (ret < 0) {
2128                         fprintf(stderr, "Error reading image\n");
2129                         break;
2130                 }
2131                 ret = 0;
2132
2133                 header = &cluster->header;
2134                 if (le64_to_cpu(header->magic) != HEADER_MAGIC ||
2135                     le64_to_cpu(header->bytenr) != current_cluster) {
2136                         fprintf(stderr, "bad header in metadump image\n");
2137                         ret = -EIO;
2138                         break;
2139                 }
2140
2141                 bytenr += BLOCK_SIZE;
2142                 nritems = le32_to_cpu(header->nritems);
2143                 for (i = 0; i < nritems; i++) {
2144                         size_t size;
2145
2146                         item = &cluster->items[i];
2147                         bufsize = le32_to_cpu(item->size);
2148                         item_bytenr = le64_to_cpu(item->bytenr);
2149
2150                         if (bufsize > max_size) {
2151                                 fprintf(stderr, "item %u size %u too big\n",
2152                                         i, bufsize);
2153                                 ret = -EIO;
2154                                 break;
2155                         }
2156
2157                         if (mdres->compress_method == COMPRESS_ZLIB) {
2158                                 ret = fread(tmp, bufsize, 1, mdres->in);
2159                                 if (ret != 1) {
2160                                         fprintf(stderr, "Error reading: %d\n",
2161                                                 errno);
2162                                         ret = -EIO;
2163                                         break;
2164                                 }
2165
2166                                 size = max_size;
2167                                 ret = uncompress(buffer,
2168                                                  (unsigned long *)&size, tmp,
2169                                                  bufsize);
2170                                 if (ret != Z_OK) {
2171                                         fprintf(stderr, "Error decompressing "
2172                                                 "%d\n", ret);
2173                                         ret = -EIO;
2174                                         break;
2175                                 }
2176                         } else {
2177                                 ret = fread(buffer, bufsize, 1, mdres->in);
2178                                 if (ret != 1) {
2179                                         fprintf(stderr, "Error reading: %d\n",
2180                                                 errno);
2181                                         ret = -EIO;
2182                                         break;
2183                                 }
2184                                 size = bufsize;
2185                         }
2186                         ret = 0;
2187
2188                         if (item_bytenr <= search &&
2189                             item_bytenr + size > search) {
2190                                 ret = read_chunk_block(mdres, buffer, search,
2191                                                        item_bytenr, size,
2192                                                        current_cluster);
2193                                 if (!ret)
2194                                         ret = 1;
2195                                 break;
2196                         }
2197                         bytenr += bufsize;
2198                 }
2199                 if (ret) {
2200                         if (ret > 0)
2201                                 ret = 0;
2202                         break;
2203                 }
2204                 if (bytenr & BLOCK_MASK)
2205                         bytenr += BLOCK_SIZE - (bytenr & BLOCK_MASK);
2206                 current_cluster = bytenr;
2207         }
2208
2209         free(tmp);
2210         free(buffer);
2211         free(cluster);
2212         return ret;
2213 }
2214
2215 static int build_chunk_tree(struct mdrestore_struct *mdres,
2216                             struct meta_cluster *cluster)
2217 {
2218         struct btrfs_super_block *super;
2219         struct meta_cluster_header *header;
2220         struct meta_cluster_item *item = NULL;
2221         u64 chunk_root_bytenr = 0;
2222         u32 i, nritems;
2223         u64 bytenr = 0;
2224         u8 *buffer;
2225         int ret;
2226
2227         /* We can't seek with stdin so don't bother doing this */
2228         if (mdres->in == stdin)
2229                 return 0;
2230
2231         ret = fread(cluster, BLOCK_SIZE, 1, mdres->in);
2232         if (ret <= 0) {
2233                 fprintf(stderr, "Error reading in cluster: %d\n", errno);
2234                 return -EIO;
2235         }
2236         ret = 0;
2237
2238         header = &cluster->header;
2239         if (le64_to_cpu(header->magic) != HEADER_MAGIC ||
2240             le64_to_cpu(header->bytenr) != 0) {
2241                 fprintf(stderr, "bad header in metadump image\n");
2242                 return -EIO;
2243         }
2244
2245         bytenr += BLOCK_SIZE;
2246         mdres->compress_method = header->compress;
2247         nritems = le32_to_cpu(header->nritems);
2248         for (i = 0; i < nritems; i++) {
2249                 item = &cluster->items[i];
2250
2251                 if (le64_to_cpu(item->bytenr) == BTRFS_SUPER_INFO_OFFSET)
2252                         break;
2253                 bytenr += le32_to_cpu(item->size);
2254                 if (fseek(mdres->in, le32_to_cpu(item->size), SEEK_CUR)) {
2255                         fprintf(stderr, "Error seeking: %d\n", errno);
2256                         return -EIO;
2257                 }
2258         }
2259
2260         if (!item || le64_to_cpu(item->bytenr) != BTRFS_SUPER_INFO_OFFSET) {
2261                 fprintf(stderr, "Huh, didn't find the super?\n");
2262                 return -EINVAL;
2263         }
2264
2265         buffer = malloc(le32_to_cpu(item->size));
2266         if (!buffer) {
2267                 fprintf(stderr, "Error allocating buffer\n");
2268                 return -ENOMEM;
2269         }
2270
2271         ret = fread(buffer, le32_to_cpu(item->size), 1, mdres->in);
2272         if (ret != 1) {
2273                 fprintf(stderr, "Error reading buffer: %d\n", errno);
2274                 free(buffer);
2275                 return -EIO;
2276         }
2277
2278         if (mdres->compress_method == COMPRESS_ZLIB) {
2279                 size_t size = MAX_PENDING_SIZE * 2;
2280                 u8 *tmp;
2281
2282                 tmp = malloc(MAX_PENDING_SIZE * 2);
2283                 if (!tmp) {
2284                         free(buffer);
2285                         return -ENOMEM;
2286                 }
2287                 ret = uncompress(tmp, (unsigned long *)&size,
2288                                  buffer, le32_to_cpu(item->size));
2289                 if (ret != Z_OK) {
2290                         fprintf(stderr, "Error decompressing %d\n", ret);
2291                         free(buffer);
2292                         free(tmp);
2293                         return -EIO;
2294                 }
2295                 free(buffer);
2296                 buffer = tmp;
2297         }
2298
2299         pthread_mutex_lock(&mdres->mutex);
2300         super = (struct btrfs_super_block *)buffer;
2301         chunk_root_bytenr = btrfs_super_chunk_root(super);
2302         mdres->nodesize = btrfs_super_nodesize(super);
2303         memcpy(mdres->fsid, super->fsid, BTRFS_FSID_SIZE);
2304         memcpy(mdres->uuid, super->dev_item.uuid,
2305                        BTRFS_UUID_SIZE);
2306         mdres->devid = le64_to_cpu(super->dev_item.devid);
2307         free(buffer);
2308         pthread_mutex_unlock(&mdres->mutex);
2309
2310         return search_for_chunk_blocks(mdres, chunk_root_bytenr, 0);
2311 }
2312
2313 static int range_contains_super(u64 physical, u64 bytes)
2314 {
2315         u64 super_bytenr;
2316         int i;
2317
2318         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
2319                 super_bytenr = btrfs_sb_offset(i);
2320                 if (super_bytenr >= physical &&
2321                     super_bytenr < physical + bytes)
2322                         return 1;
2323         }
2324
2325         return 0;
2326 }
2327
2328 static void remap_overlapping_chunks(struct mdrestore_struct *mdres)
2329 {
2330         struct fs_chunk *fs_chunk;
2331
2332         while (!list_empty(&mdres->overlapping_chunks)) {
2333                 fs_chunk = list_first_entry(&mdres->overlapping_chunks,
2334                                             struct fs_chunk, list);
2335                 list_del_init(&fs_chunk->list);
2336                 if (range_contains_super(fs_chunk->physical,
2337                                          fs_chunk->bytes)) {
2338                         fprintf(stderr, "Remapping a chunk that had a super "
2339                                 "mirror inside of it, clearing space cache "
2340                                 "so we don't end up with corruption\n");
2341                         mdres->clear_space_cache = 1;
2342                 }
2343                 fs_chunk->physical = mdres->last_physical_offset;
2344                 tree_insert(&mdres->physical_tree, &fs_chunk->p, physical_cmp);
2345                 mdres->last_physical_offset += fs_chunk->bytes;
2346         }
2347 }
2348
2349 static int fixup_devices(struct btrfs_fs_info *fs_info,
2350                          struct mdrestore_struct *mdres, off_t dev_size)
2351 {
2352         struct btrfs_trans_handle *trans;
2353         struct btrfs_dev_item *dev_item;
2354         struct btrfs_path *path;
2355         struct extent_buffer *leaf;
2356         struct btrfs_root *root = fs_info->chunk_root;
2357         struct btrfs_key key;
2358         u64 devid, cur_devid;
2359         int ret;
2360
2361         path = btrfs_alloc_path();
2362         if (!path) {
2363                 fprintf(stderr, "Error allocating path\n");
2364                 return -ENOMEM;
2365         }
2366
2367         trans = btrfs_start_transaction(fs_info->tree_root, 1);
2368         if (IS_ERR(trans)) {
2369                 fprintf(stderr, "Error starting transaction %ld\n",
2370                         PTR_ERR(trans));
2371                 btrfs_free_path(path);
2372                 return PTR_ERR(trans);
2373         }
2374
2375         dev_item = &fs_info->super_copy->dev_item;
2376
2377         devid = btrfs_stack_device_id(dev_item);
2378
2379         btrfs_set_stack_device_total_bytes(dev_item, dev_size);
2380         btrfs_set_stack_device_bytes_used(dev_item, mdres->alloced_chunks);
2381
2382         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2383         key.type = BTRFS_DEV_ITEM_KEY;
2384         key.offset = 0;
2385
2386 again:
2387         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2388         if (ret < 0) {
2389                 fprintf(stderr, "search failed %d\n", ret);
2390                 exit(1);
2391         }
2392
2393         while (1) {
2394                 leaf = path->nodes[0];
2395                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2396                         ret = btrfs_next_leaf(root, path);
2397                         if (ret < 0) {
2398                                 fprintf(stderr, "Error going to next leaf "
2399                                         "%d\n", ret);
2400                                 exit(1);
2401                         }
2402                         if (ret > 0) {
2403                                 ret = 0;
2404                                 break;
2405                         }
2406                         leaf = path->nodes[0];
2407                 }
2408
2409                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2410                 if (key.type > BTRFS_DEV_ITEM_KEY)
2411                         break;
2412                 if (key.type != BTRFS_DEV_ITEM_KEY) {
2413                         path->slots[0]++;
2414                         continue;
2415                 }
2416
2417                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2418                                           struct btrfs_dev_item);
2419                 cur_devid = btrfs_device_id(leaf, dev_item);
2420                 if (devid != cur_devid) {
2421                         ret = btrfs_del_item(trans, root, path);
2422                         if (ret) {
2423                                 fprintf(stderr, "Error deleting item %d\n",
2424                                         ret);
2425                                 exit(1);
2426                         }
2427                         btrfs_release_path(path);
2428                         goto again;
2429                 }
2430
2431                 btrfs_set_device_total_bytes(leaf, dev_item, dev_size);
2432                 btrfs_set_device_bytes_used(leaf, dev_item,
2433                                             mdres->alloced_chunks);
2434                 btrfs_mark_buffer_dirty(leaf);
2435                 path->slots[0]++;
2436         }
2437
2438         btrfs_free_path(path);
2439         ret = btrfs_commit_transaction(trans, fs_info->tree_root);
2440         if (ret) {
2441                 fprintf(stderr, "Commit failed %d\n", ret);
2442                 return ret;
2443         }
2444         return 0;
2445 }
2446
2447 static int restore_metadump(const char *input, FILE *out, int old_restore,
2448                             int num_threads, int fixup_offset,
2449                             const char *target, int multi_devices)
2450 {
2451         struct meta_cluster *cluster = NULL;
2452         struct meta_cluster_header *header;
2453         struct mdrestore_struct mdrestore;
2454         struct btrfs_fs_info *info = NULL;
2455         u64 bytenr = 0;
2456         FILE *in = NULL;
2457         int ret = 0;
2458
2459         if (!strcmp(input, "-")) {
2460                 in = stdin;
2461         } else {
2462                 in = fopen(input, "r");
2463                 if (!in) {
2464                         perror("unable to open metadump image");
2465                         return 1;
2466                 }
2467         }
2468
2469         /* NOTE: open with write mode */
2470         if (fixup_offset) {
2471                 BUG_ON(!target);
2472                 info = open_ctree_fs_info(target, 0, 0, 0,
2473                                           OPEN_CTREE_WRITES |
2474                                           OPEN_CTREE_RESTORE |
2475                                           OPEN_CTREE_PARTIAL);
2476                 if (!info) {
2477                         fprintf(stderr, "%s: open ctree failed\n", __func__);
2478                         ret = -EIO;
2479                         goto failed_open;
2480                 }
2481         }
2482
2483         cluster = malloc(BLOCK_SIZE);
2484         if (!cluster) {
2485                 fprintf(stderr, "Error allocating cluster\n");
2486                 ret = -ENOMEM;
2487                 goto failed_info;
2488         }
2489
2490         ret = mdrestore_init(&mdrestore, in, out, old_restore, num_threads,
2491                              fixup_offset, info, multi_devices);
2492         if (ret) {
2493                 fprintf(stderr, "Error initializing mdrestore %d\n", ret);
2494                 goto failed_cluster;
2495         }
2496
2497         if (!multi_devices && !old_restore) {
2498                 ret = build_chunk_tree(&mdrestore, cluster);
2499                 if (ret)
2500                         goto out;
2501                 if (!list_empty(&mdrestore.overlapping_chunks))
2502                         remap_overlapping_chunks(&mdrestore);
2503         }
2504
2505         if (in != stdin && fseek(in, 0, SEEK_SET)) {
2506                 fprintf(stderr, "Error seeking %d\n", errno);
2507                 goto out;
2508         }
2509
2510         while (!mdrestore.error) {
2511                 ret = fread(cluster, BLOCK_SIZE, 1, in);
2512                 if (!ret)
2513                         break;
2514
2515                 header = &cluster->header;
2516                 if (le64_to_cpu(header->magic) != HEADER_MAGIC ||
2517                     le64_to_cpu(header->bytenr) != bytenr) {
2518                         fprintf(stderr, "bad header in metadump image\n");
2519                         ret = -EIO;
2520                         break;
2521                 }
2522                 ret = add_cluster(cluster, &mdrestore, &bytenr);
2523                 if (ret) {
2524                         fprintf(stderr, "Error adding cluster\n");
2525                         break;
2526                 }
2527         }
2528         ret = wait_for_worker(&mdrestore);
2529
2530         if (!ret && !multi_devices && !old_restore) {
2531                 struct btrfs_root *root;
2532                 struct stat st;
2533
2534                 root = open_ctree_fd(fileno(out), target, 0,
2535                                           OPEN_CTREE_PARTIAL |
2536                                           OPEN_CTREE_WRITES |
2537                                           OPEN_CTREE_NO_DEVICES);
2538                 if (!root) {
2539                         fprintf(stderr, "unable to open %s\n", target);
2540                         ret = -EIO;
2541                         goto out;
2542                 }
2543                 info = root->fs_info;
2544
2545                 if (stat(target, &st)) {
2546                         fprintf(stderr, "statting %s failed\n", target);
2547                         close_ctree(info->chunk_root);
2548                         return 1;
2549                 }
2550
2551                 ret = fixup_devices(info, &mdrestore, st.st_size);
2552                 close_ctree(info->chunk_root);
2553                 if (ret)
2554                         goto out;
2555         }
2556 out:
2557         mdrestore_destroy(&mdrestore, num_threads);
2558 failed_cluster:
2559         free(cluster);
2560 failed_info:
2561         if (fixup_offset && info)
2562                 close_ctree(info->chunk_root);
2563 failed_open:
2564         if (in != stdin)
2565                 fclose(in);
2566         return ret;
2567 }
2568
2569 static int update_disk_super_on_device(struct btrfs_fs_info *info,
2570                                        const char *other_dev, u64 cur_devid)
2571 {
2572         struct btrfs_key key;
2573         struct extent_buffer *leaf;
2574         struct btrfs_path path;
2575         struct btrfs_dev_item *dev_item;
2576         struct btrfs_super_block *disk_super;
2577         char dev_uuid[BTRFS_UUID_SIZE];
2578         char fs_uuid[BTRFS_UUID_SIZE];
2579         u64 devid, type, io_align, io_width;
2580         u64 sector_size, total_bytes, bytes_used;
2581         char buf[BTRFS_SUPER_INFO_SIZE];
2582         int fp = -1;
2583         int ret;
2584
2585         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2586         key.type = BTRFS_DEV_ITEM_KEY;
2587         key.offset = cur_devid;
2588
2589         btrfs_init_path(&path);
2590         ret = btrfs_search_slot(NULL, info->chunk_root, &key, &path, 0, 0);
2591         if (ret) {
2592                 fprintf(stderr, "ERROR: search key failed\n");
2593                 ret = -EIO;
2594                 goto out;
2595         }
2596
2597         leaf = path.nodes[0];
2598         dev_item = btrfs_item_ptr(leaf, path.slots[0],
2599                                   struct btrfs_dev_item);
2600
2601         devid = btrfs_device_id(leaf, dev_item);
2602         if (devid != cur_devid) {
2603                 printk("ERROR: devid %llu mismatch with %llu\n", devid, cur_devid);
2604                 ret = -EIO;
2605                 goto out;
2606         }
2607
2608         type = btrfs_device_type(leaf, dev_item);
2609         io_align = btrfs_device_io_align(leaf, dev_item);
2610         io_width = btrfs_device_io_width(leaf, dev_item);
2611         sector_size = btrfs_device_sector_size(leaf, dev_item);
2612         total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2613         bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2614         read_extent_buffer(leaf, dev_uuid, (unsigned long)btrfs_device_uuid(dev_item), BTRFS_UUID_SIZE);
2615         read_extent_buffer(leaf, fs_uuid, (unsigned long)btrfs_device_fsid(dev_item), BTRFS_UUID_SIZE);
2616
2617         btrfs_release_path(&path);
2618
2619         printk("update disk super on %s devid=%llu\n", other_dev, devid);
2620
2621         /* update other devices' super block */
2622         fp = open(other_dev, O_CREAT | O_RDWR, 0600);
2623         if (fp < 0) {
2624                 fprintf(stderr, "ERROR: could not open %s\n", other_dev);
2625                 ret = -EIO;
2626                 goto out;
2627         }
2628
2629         memcpy(buf, info->super_copy, BTRFS_SUPER_INFO_SIZE);
2630
2631         disk_super = (struct btrfs_super_block *)buf;
2632         dev_item = &disk_super->dev_item;
2633
2634         btrfs_set_stack_device_type(dev_item, type);
2635         btrfs_set_stack_device_id(dev_item, devid);
2636         btrfs_set_stack_device_total_bytes(dev_item, total_bytes);
2637         btrfs_set_stack_device_bytes_used(dev_item, bytes_used);
2638         btrfs_set_stack_device_io_align(dev_item, io_align);
2639         btrfs_set_stack_device_io_width(dev_item, io_width);
2640         btrfs_set_stack_device_sector_size(dev_item, sector_size);
2641         memcpy(dev_item->uuid, dev_uuid, BTRFS_UUID_SIZE);
2642         memcpy(dev_item->fsid, fs_uuid, BTRFS_UUID_SIZE);
2643         csum_block((u8 *)buf, BTRFS_SUPER_INFO_SIZE);
2644
2645         ret = pwrite64(fp, buf, BTRFS_SUPER_INFO_SIZE, BTRFS_SUPER_INFO_OFFSET);
2646         if (ret != BTRFS_SUPER_INFO_SIZE) {
2647                 if (ret < 0)
2648                         fprintf(stderr, "ERROR: cannot write superblock: %s\n", strerror(ret));
2649                 else
2650                         fprintf(stderr, "ERROR: cannot write superblock\n");
2651                 ret = -EIO;
2652                 goto out;
2653         }
2654
2655         write_backup_supers(fp, (u8 *)buf);
2656
2657 out:
2658         if (fp != -1)
2659                 close(fp);
2660         return ret;
2661 }
2662
2663 static void print_usage(int ret)
2664 {
2665         fprintf(stderr, "usage: btrfs-image [options] source target\n");
2666         fprintf(stderr, "\t-r      \trestore metadump image\n");
2667         fprintf(stderr, "\t-c value\tcompression level (0 ~ 9)\n");
2668         fprintf(stderr, "\t-t value\tnumber of threads (1 ~ 32)\n");
2669         fprintf(stderr, "\t-o      \tdon't mess with the chunk tree when restoring\n");
2670         fprintf(stderr, "\t-s      \tsanitize file names, use once to just use garbage, use twice if you want crc collisions\n");
2671         fprintf(stderr, "\t-w      \twalk all trees instead of using extent tree, do this if your extent tree is broken\n");
2672         fprintf(stderr, "\t-m      \trestore for multiple devices\n");
2673         fprintf(stderr, "\n");
2674         fprintf(stderr, "\tIn the dump mode, source is the btrfs device and target is the output file (use '-' for stdout).\n");
2675         fprintf(stderr, "\tIn the restore mode, source is the dumped image and target is the btrfs device/file.\n");
2676         exit(ret);
2677 }
2678
2679 int main(int argc, char *argv[])
2680 {
2681         char *source;
2682         char *target;
2683         u64 num_threads = 0;
2684         u64 compress_level = 0;
2685         int create = 1;
2686         int old_restore = 0;
2687         int walk_trees = 0;
2688         int multi_devices = 0;
2689         int ret;
2690         int sanitize = 0;
2691         int dev_cnt = 0;
2692         int usage_error = 0;
2693         FILE *out;
2694
2695         while (1) {
2696                 static const struct option long_options[] = {
2697                         { "help", no_argument, NULL, GETOPT_VAL_HELP},
2698                         { NULL, 0, NULL, 0 }
2699                 };
2700                 int c = getopt_long(argc, argv, "rc:t:oswm", long_options, NULL);
2701                 if (c < 0)
2702                         break;
2703                 switch (c) {
2704                 case 'r':
2705                         create = 0;
2706                         break;
2707                 case 't':
2708                         num_threads = arg_strtou64(optarg);
2709                         if (num_threads > 32)
2710                                 print_usage(1);
2711                         break;
2712                 case 'c':
2713                         compress_level = arg_strtou64(optarg);
2714                         if (compress_level > 9)
2715                                 print_usage(1);
2716                         break;
2717                 case 'o':
2718                         old_restore = 1;
2719                         break;
2720                 case 's':
2721                         sanitize++;
2722                         break;
2723                 case 'w':
2724                         walk_trees = 1;
2725                         break;
2726                 case 'm':
2727                         create = 0;
2728                         multi_devices = 1;
2729                         break;
2730                         case GETOPT_VAL_HELP:
2731                 default:
2732                         print_usage(c != GETOPT_VAL_HELP);
2733                 }
2734         }
2735
2736         set_argv0(argv);
2737         if (check_argc_min(argc - optind, 2))
2738                 print_usage(1);
2739
2740         dev_cnt = argc - optind - 1;
2741
2742         if (create) {
2743                 if (old_restore) {
2744                         fprintf(stderr, "Usage error: create and restore cannot be used at the same time\n");
2745                         usage_error++;
2746                 }
2747         } else {
2748                 if (walk_trees || sanitize || compress_level) {
2749                         fprintf(stderr, "Usage error: use -w, -s, -c options for restore makes no sense\n");
2750                         usage_error++;
2751                 }
2752                 if (multi_devices && dev_cnt < 2) {
2753                         fprintf(stderr, "Usage error: not enough devices specified for -m option\n");
2754                         usage_error++;
2755                 }
2756                 if (!multi_devices && dev_cnt != 1) {
2757                         fprintf(stderr, "Usage error: accepts only 1 device without -m option\n");
2758                         usage_error++;
2759                 }
2760         }
2761
2762         if (usage_error)
2763                 print_usage(1);
2764
2765         source = argv[optind];
2766         target = argv[optind + 1];
2767
2768         if (create && !strcmp(target, "-")) {
2769                 out = stdout;
2770         } else {
2771                 out = fopen(target, "w+");
2772                 if (!out) {
2773                         perror("unable to create target file");
2774                         exit(1);
2775                 }
2776         }
2777
2778         if (compress_level > 0 || create == 0) {
2779                 if (num_threads == 0) {
2780                         long tmp = sysconf(_SC_NPROCESSORS_ONLN);
2781
2782                         if (tmp <= 0)
2783                                 tmp = 1;
2784                         num_threads = tmp;
2785                 }
2786         } else {
2787                 num_threads = 0;
2788         }
2789
2790         if (create) {
2791                 ret = check_mounted(source);
2792                 if (ret < 0) {
2793                         fprintf(stderr, "Could not check mount status: %s\n",
2794                                 strerror(-ret));
2795                         exit(1);
2796                 } else if (ret)
2797                         fprintf(stderr,
2798                 "WARNING: The device is mounted. Make sure the filesystem is quiescent.\n");
2799
2800                 ret = create_metadump(source, out, num_threads,
2801                                       compress_level, sanitize, walk_trees);
2802         } else {
2803                 ret = restore_metadump(source, out, old_restore, num_threads,
2804                                        0, target, multi_devices);
2805         }
2806         if (ret) {
2807                 printk("%s failed (%s)\n", (create) ? "create" : "restore",
2808                        strerror(errno));
2809                 goto out;
2810         }
2811
2812          /* extended support for multiple devices */
2813         if (!create && multi_devices) {
2814                 struct btrfs_fs_info *info;
2815                 u64 total_devs;
2816                 int i;
2817
2818                 info = open_ctree_fs_info(target, 0, 0, 0,
2819                                           OPEN_CTREE_PARTIAL |
2820                                           OPEN_CTREE_RESTORE);
2821                 if (!info) {
2822                         fprintf(stderr, "unable to open %s error = %s\n",
2823                                 target, strerror(errno));
2824                         return 1;
2825                 }
2826
2827                 total_devs = btrfs_super_num_devices(info->super_copy);
2828                 if (total_devs != dev_cnt) {
2829                         printk("it needs %llu devices but has only %d\n",
2830                                 total_devs, dev_cnt);
2831                         close_ctree(info->chunk_root);
2832                         goto out;
2833                 }
2834
2835                 /* update super block on other disks */
2836                 for (i = 2; i <= dev_cnt; i++) {
2837                         ret = update_disk_super_on_device(info,
2838                                         argv[optind + i], (u64)i);
2839                         if (ret) {
2840                                 printk("update disk super failed devid=%d (error=%d)\n",
2841                                         i, ret);
2842                                 close_ctree(info->chunk_root);
2843                                 exit(1);
2844                         }
2845                 }
2846
2847                 close_ctree(info->chunk_root);
2848
2849                 /* fix metadata block to map correct chunk */
2850                 ret = restore_metadump(source, out, 0, num_threads, 1,
2851                                        target, 1);
2852                 if (ret) {
2853                         fprintf(stderr, "fix metadump failed (error=%d)\n",
2854                                 ret);
2855                         exit(1);
2856                 }
2857         }
2858 out:
2859         if (out == stdout) {
2860                 fflush(out);
2861         } else {
2862                 fclose(out);
2863                 if (ret && create) {
2864                         int unlink_ret;
2865
2866                         unlink_ret = unlink(target);
2867                         if (unlink_ret)
2868                                 fprintf(stderr,
2869                                         "unlink output file failed : %s\n",
2870                                         strerror(errno));
2871                 }
2872         }
2873
2874         btrfs_close_all_devices();
2875
2876         return !!ret;
2877 }