btrfs-image.c

   1 /*
   2  * Copyright (C) 2008 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <pthread.h>
  20 #include <stdio.h>
  21 #include <stdlib.h>
  22 #include <sys/types.h>
  23 #include <sys/stat.h>
  24 #include <fcntl.h>
  25 #include <unistd.h>
  26 #include <dirent.h>
  27 #include <zlib.h>
  28 #include "kerncompat.h"
  29 #include "crc32c.h"
  30 #include "ctree.h"
  31 #include "disk-io.h"
  32 #include "transaction.h"
  33 #include "utils.h"
  34 #include "volumes.h"
  35 #include "extent_io.h"
  36
  37 #define HEADER_MAGIC            0xbd5c25e27295668bULL
  38 #define MAX_PENDING_SIZE        (256 * 1024)
  39 #define BLOCK_SIZE              1024
  40 #define BLOCK_MASK              (BLOCK_SIZE - 1)
  41
  42 #define COMPRESS_NONE           0
  43 #define COMPRESS_ZLIB           1
  44
  45 struct meta_cluster_item {
  46         __le64 bytenr;
  47         __le32 size;
  48 } __attribute__ ((__packed__));
  49
  50 struct meta_cluster_header {
  51         __le64 magic;
  52         __le64 bytenr;
  53         __le32 nritems;
  54         u8 compress;
  55 } __attribute__ ((__packed__));
  56
  57 /* cluster header + index items + buffers */
  58 struct meta_cluster {
  59         struct meta_cluster_header header;
  60         struct meta_cluster_item items[];
  61 } __attribute__ ((__packed__));
  62
  63 #define ITEMS_PER_CLUSTER ((BLOCK_SIZE - sizeof(struct meta_cluster)) / \
  64                            sizeof(struct meta_cluster_item))
  65
  66 struct fs_chunk {
  67         u64 logical;
  68         u64 physical;
  69         u64 bytes;
  70         struct rb_node l;
  71         struct rb_node p;
  72         struct list_head list;
  73 };
  74
  75 struct async_work {
  76         struct list_head list;
  77         struct list_head ordered;
  78         u64 start;
  79         u64 size;
  80         u8 *buffer;
  81         size_t bufsize;
  82         int error;
  83 };
  84
  85 struct metadump_struct {
  86         struct btrfs_root *root;
  87         FILE *out;
  88
  89         struct meta_cluster *cluster;
  90
  91         pthread_t *threads;
  92         size_t num_threads;
  93         pthread_mutex_t mutex;
  94         pthread_cond_t cond;
  95         struct rb_root name_tree;
  96
  97         struct list_head list;
  98         struct list_head ordered;
  99         size_t num_items;
 100         size_t num_ready;
 101
 102         u64 pending_start;
 103         u64 pending_size;
 104
 105         int compress_level;
 106         int done;
 107         int data;
 108         int sanitize_names;
 109
 110         int error;
 111 };
 112
 113 struct name {
 114         struct rb_node n;
 115         char *val;
 116         char *sub;
 117         u32 len;
 118 };
 119
 120 struct mdrestore_struct {
 121         FILE *in;
 122         FILE *out;
 123
 124         pthread_t *threads;
 125         size_t num_threads;
 126         pthread_mutex_t mutex;
 127         pthread_cond_t cond;
 128
 129         struct rb_root chunk_tree;
 130         struct rb_root physical_tree;
 131         struct list_head list;
 132         struct list_head overlapping_chunks;
 133         size_t num_items;
 134         u32 leafsize;
 135         u64 devid;
 136         u64 alloced_chunks;
 137         u64 last_physical_offset;
 138         u8 uuid[BTRFS_UUID_SIZE];
 139         u8 fsid[BTRFS_FSID_SIZE];
 140
 141         int compress_method;
 142         int done;
 143         int error;
 144         int old_restore;
 145         int fixup_offset;
 146         int multi_devices;
 147         int clear_space_cache;
 148         struct btrfs_fs_info *info;
 149 };
 150
 151 static void print_usage(void) __attribute__((noreturn));
 152 static int search_for_chunk_blocks(struct mdrestore_struct *mdres,
 153                                    u64 search, u64 cluster_bytenr);
 154 static struct extent_buffer *alloc_dummy_eb(u64 bytenr, u32 size);
 155
 156 static void csum_block(u8 *buf, size_t len)
 157 {
 158         char result[BTRFS_CRC32_SIZE];
 159         u32 crc = ~(u32)0;
 160         crc = crc32c(crc, buf + BTRFS_CSUM_SIZE, len - BTRFS_CSUM_SIZE);
 161         btrfs_csum_final(crc, result);
 162         memcpy(buf, result, BTRFS_CRC32_SIZE);
 163 }
 164
 165 static int has_name(struct btrfs_key *key)
 166 {
 167         switch (key->type) {
 168         case BTRFS_DIR_ITEM_KEY:
 169         case BTRFS_DIR_INDEX_KEY:
 170         case BTRFS_INODE_REF_KEY:
 171         case BTRFS_INODE_EXTREF_KEY:
 172         case BTRFS_XATTR_ITEM_KEY:
 173                 return 1;
 174         default:
 175                 break;
 176         }
 177
 178         return 0;
 179 }
 180
 181 static char *generate_garbage(u32 name_len)
 182 {
 183         char *buf = malloc(name_len);
 184         int i;
 185
 186         if (!buf)
 187                 return NULL;
 188
 189         for (i = 0; i < name_len; i++) {
 190                 char c = rand() % 94 + 33;
 191
 192                 if (c == '/')
 193                         c++;
 194                 buf[i] = c;
 195         }
 196
 197         return buf;
 198 }
 199
 200 static int name_cmp(struct rb_node *a, struct rb_node *b, int fuzz)
 201 {
 202         struct name *entry = rb_entry(a, struct name, n);
 203         struct name *ins = rb_entry(b, struct name, n);
 204         u32 len;
 205
 206         len = min(ins->len, entry->len);
 207         return memcmp(ins->val, entry->val, len);
 208 }
 209
 210 static int chunk_cmp(struct rb_node *a, struct rb_node *b, int fuzz)
 211 {
 212         struct fs_chunk *entry = rb_entry(a, struct fs_chunk, l);
 213         struct fs_chunk *ins = rb_entry(b, struct fs_chunk, l);
 214
 215         if (fuzz && ins->logical >= entry->logical &&
 216             ins->logical < entry->logical + entry->bytes)
 217                 return 0;
 218
 219         if (ins->logical < entry->logical)
 220                 return -1;
 221         else if (ins->logical > entry->logical)
 222                 return 1;
 223         return 0;
 224 }
 225
 226 static int physical_cmp(struct rb_node *a, struct rb_node *b, int fuzz)
 227 {
 228         struct fs_chunk *entry = rb_entry(a, struct fs_chunk, p);
 229         struct fs_chunk *ins = rb_entry(b, struct fs_chunk, p);
 230
 231         if (fuzz && ins->physical >= entry->physical &&
 232             ins->physical < entry->physical + entry->bytes)
 233                 return 0;
 234
 235         if (fuzz && entry->physical >= ins->physical &&
 236             entry->physical < ins->physical + ins->bytes)
 237                 return 0;
 238
 239         if (ins->physical < entry->physical)
 240                 return -1;
 241         else if (ins->physical > entry->physical)
 242                 return 1;
 243         return 0;
 244 }
 245
 246 static void tree_insert(struct rb_root *root, struct rb_node *ins,
 247                         int (*cmp)(struct rb_node *a, struct rb_node *b,
 248                                    int fuzz))
 249 {
 250         struct rb_node ** p = &root->rb_node;
 251         struct rb_node * parent = NULL;
 252         int dir;
 253
 254         while(*p) {
 255                 parent = *p;
 256
 257                 dir = cmp(*p, ins, 1);
 258                 if (dir < 0)
 259                         p = &(*p)->rb_left;
 260                 else if (dir > 0)
 261                         p = &(*p)->rb_right;
 262                 else
 263                         BUG();
 264         }
 265
 266         rb_link_node(ins, parent, p);
 267         rb_insert_color(ins, root);
 268 }
 269
 270 static struct rb_node *tree_search(struct rb_root *root,
 271                                    struct rb_node *search,
 272                                    int (*cmp)(struct rb_node *a,
 273                                               struct rb_node *b, int fuzz),
 274                                    int fuzz)
 275 {
 276         struct rb_node *n = root->rb_node;
 277         int dir;
 278
 279         while (n) {
 280                 dir = cmp(n, search, fuzz);
 281                 if (dir < 0)
 282                         n = n->rb_left;
 283                 else if (dir > 0)
 284                         n = n->rb_right;
 285                 else
 286                         return n;
 287         }
 288
 289         return NULL;
 290 }
 291
 292 static u64 logical_to_physical(struct mdrestore_struct *mdres, u64 logical, u64 *size)
 293 {
 294         struct fs_chunk *fs_chunk;
 295         struct rb_node *entry;
 296         struct fs_chunk search;
 297         u64 offset;
 298
 299         if (logical == BTRFS_SUPER_INFO_OFFSET)
 300                 return logical;
 301
 302         search.logical = logical;
 303         entry = tree_search(&mdres->chunk_tree, &search.l, chunk_cmp, 1);
 304         if (!entry) {
 305                 if (mdres->in != stdin)
 306                         printf("Couldn't find a chunk, using logical\n");
 307                 return logical;
 308         }
 309         fs_chunk = rb_entry(entry, struct fs_chunk, l);
 310         if (fs_chunk->logical > logical || fs_chunk->logical + fs_chunk->bytes < logical)
 311                 BUG();
 312         offset = search.logical - fs_chunk->logical;
 313
 314         *size = min(*size, fs_chunk->bytes + fs_chunk->logical - logical);
 315         return fs_chunk->physical + offset;
 316 }
 317
 318
 319 static char *find_collision(struct metadump_struct *md, char *name,
 320                             u32 name_len)
 321 {
 322         struct name *val;
 323         struct rb_node *entry;
 324         struct name tmp;
 325         unsigned long checksum;
 326         int found = 0;
 327         int i;
 328
 329         tmp.val = name;
 330         tmp.len = name_len;
 331         entry = tree_search(&md->name_tree, &tmp.n, name_cmp, 0);
 332         if (entry) {
 333                 val = rb_entry(entry, struct name, n);
 334                 free(name);
 335                 return val->sub;
 336         }
 337
 338         val = malloc(sizeof(struct name));
 339         if (!val) {
 340                 fprintf(stderr, "Couldn't sanitize name, enomem\n");
 341                 free(name);
 342                 return NULL;
 343         }
 344
 345         memset(val, 0, sizeof(*val));
 346
 347         val->val = name;
 348         val->len = name_len;
 349         val->sub = malloc(name_len);
 350         if (!val->sub) {
 351                 fprintf(stderr, "Couldn't sanitize name, enomem\n");
 352                 free(val);
 353                 free(name);
 354                 return NULL;
 355         }
 356
 357         checksum = crc32c(~1, val->val, name_len);
 358         memset(val->sub, ' ', name_len);
 359         i = 0;
 360         while (1) {
 361                 if (crc32c(~1, val->sub, name_len) == checksum &&
 362                     memcmp(val->sub, val->val, val->len)) {
 363                         found = 1;
 364                         break;
 365                 }
 366
 367                 if (val->sub[i] == 127) {
 368                         do {
 369                                 i++;
 370                                 if (i >= name_len)
 371                                         break;
 372                         } while (val->sub[i] == 127);
 373
 374                         if (i >= name_len)
 375                                 break;
 376                         val->sub[i]++;
 377                         if (val->sub[i] == '/')
 378                                 val->sub[i]++;
 379                         memset(val->sub, ' ', i);
 380                         i = 0;
 381                         continue;
 382                 } else {
 383                         val->sub[i]++;
 384                         if (val->sub[i] == '/')
 385                                 val->sub[i]++;
 386                 }
 387         }
 388
 389         if (!found) {
 390                 fprintf(stderr, "Couldn't find a collision for '%.*s', "
 391                         "generating normal garbage, it won't match indexes\n",
 392                         val->len, val->val);
 393                 for (i = 0; i < name_len; i++) {
 394                         char c = rand() % 94 + 33;
 395
 396                         if (c == '/')
 397                                 c++;
 398                         val->sub[i] = c;
 399                 }
 400         }
 401
 402         tree_insert(&md->name_tree, &val->n, name_cmp);
 403         return val->sub;
 404 }
 405
 406 static void sanitize_dir_item(struct metadump_struct *md, struct extent_buffer *eb,
 407                               int slot)
 408 {
 409         struct btrfs_dir_item *dir_item;
 410         char *buf;
 411         char *garbage;
 412         unsigned long name_ptr;
 413         u32 total_len;
 414         u32 cur = 0;
 415         u32 this_len;
 416         u32 name_len;
 417         int free_garbage = (md->sanitize_names == 1);
 418
 419         dir_item = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
 420         total_len = btrfs_item_size_nr(eb, slot);
 421         while (cur < total_len) {
 422                 this_len = sizeof(*dir_item) +
 423                         btrfs_dir_name_len(eb, dir_item) +
 424                         btrfs_dir_data_len(eb, dir_item);
 425                 name_ptr = (unsigned long)(dir_item + 1);
 426                 name_len = btrfs_dir_name_len(eb, dir_item);
 427
 428                 if (md->sanitize_names > 1) {
 429                         buf = malloc(name_len);
 430                         if (!buf) {
 431                                 fprintf(stderr, "Couldn't sanitize name, "
 432                                         "enomem\n");
 433                                 return;
 434                         }
 435                         read_extent_buffer(eb, buf, name_ptr, name_len);
 436                         garbage = find_collision(md, buf, name_len);
 437                 } else {
 438                         garbage = generate_garbage(name_len);
 439                 }
 440                 if (!garbage) {
 441                         fprintf(stderr, "Couldn't sanitize name, enomem\n");
 442                         return;
 443                 }
 444                 write_extent_buffer(eb, garbage, name_ptr, name_len);
 445                 cur += this_len;
 446                 dir_item = (struct btrfs_dir_item *)((char *)dir_item +
 447                                                      this_len);
 448                 if (free_garbage)
 449                         free(garbage);
 450         }
 451 }
 452
 453 static void sanitize_inode_ref(struct metadump_struct *md,
 454                                struct extent_buffer *eb, int slot, int ext)
 455 {
 456         struct btrfs_inode_extref *extref;
 457         struct btrfs_inode_ref *ref;
 458         char *garbage, *buf;
 459         unsigned long ptr;
 460         unsigned long name_ptr;
 461         u32 item_size;
 462         u32 cur_offset = 0;
 463         int len;
 464         int free_garbage = (md->sanitize_names == 1);
 465
 466         item_size = btrfs_item_size_nr(eb, slot);
 467         ptr = btrfs_item_ptr_offset(eb, slot);
 468         while (cur_offset < item_size) {
 469                 if (ext) {
 470                         extref = (struct btrfs_inode_extref *)(ptr +
 471                                                                cur_offset);
 472                         name_ptr = (unsigned long)(&extref->name);
 473                         len = btrfs_inode_extref_name_len(eb, extref);
 474                         cur_offset += sizeof(*extref);
 475                 } else {
 476                         ref = (struct btrfs_inode_ref *)(ptr + cur_offset);
 477                         len = btrfs_inode_ref_name_len(eb, ref);
 478                         name_ptr = (unsigned long)(ref + 1);
 479                         cur_offset += sizeof(*ref);
 480                 }
 481                 cur_offset += len;
 482
 483                 if (md->sanitize_names > 1) {
 484                         buf = malloc(len);
 485                         if (!buf) {
 486                                 fprintf(stderr, "Couldn't sanitize name, "
 487                                         "enomem\n");
 488                                 return;
 489                         }
 490                         read_extent_buffer(eb, buf, name_ptr, len);
 491                         garbage = find_collision(md, buf, len);
 492                 } else {
 493                         garbage = generate_garbage(len);
 494                 }
 495
 496                 if (!garbage) {
 497                         fprintf(stderr, "Couldn't sanitize name, enomem\n");
 498                         return;
 499                 }
 500                 write_extent_buffer(eb, garbage, name_ptr, len);
 501                 if (free_garbage)
 502                         free(garbage);
 503         }
 504 }
 505
 506 static void sanitize_xattr(struct metadump_struct *md,
 507                            struct extent_buffer *eb, int slot)
 508 {
 509         struct btrfs_dir_item *dir_item;
 510         unsigned long data_ptr;
 511         u32 data_len;
 512
 513         dir_item = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
 514         data_len = btrfs_dir_data_len(eb, dir_item);
 515
 516         data_ptr = (unsigned long)((char *)(dir_item + 1) +
 517                                    btrfs_dir_name_len(eb, dir_item));
 518         memset_extent_buffer(eb, 0, data_ptr, data_len);
 519 }
 520
 521 static void sanitize_name(struct metadump_struct *md, u8 *dst,
 522                           struct extent_buffer *src, struct btrfs_key *key,
 523                           int slot)
 524 {
 525         struct extent_buffer *eb;
 526
 527         eb = alloc_dummy_eb(src->start, src->len);
 528         if (!eb) {
 529                 fprintf(stderr, "Couldn't sanitize name, no memory\n");
 530                 return;
 531         }
 532
 533         memcpy(eb->data, dst, eb->len);
 534
 535         switch (key->type) {
 536         case BTRFS_DIR_ITEM_KEY:
 537         case BTRFS_DIR_INDEX_KEY:
 538                 sanitize_dir_item(md, eb, slot);
 539                 break;
 540         case BTRFS_INODE_REF_KEY:
 541                 sanitize_inode_ref(md, eb, slot, 0);
 542                 break;
 543         case BTRFS_INODE_EXTREF_KEY:
 544                 sanitize_inode_ref(md, eb, slot, 1);
 545                 break;
 546         case BTRFS_XATTR_ITEM_KEY:
 547                 sanitize_xattr(md, eb, slot);
 548                 break;
 549         default:
 550                 break;
 551         }
 552
 553         memcpy(dst, eb->data, eb->len);
 554         free(eb);
 555 }
 556
 557 /*
 558  * zero inline extents and csum items
 559  */
 560 static void zero_items(struct metadump_struct *md, u8 *dst,
 561                        struct extent_buffer *src)
 562 {
 563         struct btrfs_file_extent_item *fi;
 564         struct btrfs_item *item;
 565         struct btrfs_key key;
 566         u32 nritems = btrfs_header_nritems(src);
 567         size_t size;
 568         unsigned long ptr;
 569         int i, extent_type;
 570
 571         for (i = 0; i < nritems; i++) {
 572                 item = btrfs_item_nr(i);
 573                 btrfs_item_key_to_cpu(src, &key, i);
 574                 if (key.type == BTRFS_CSUM_ITEM_KEY) {
 575                         size = btrfs_item_size_nr(src, i);
 576                         memset(dst + btrfs_leaf_data(src) +
 577                                btrfs_item_offset_nr(src, i), 0, size);
 578                         continue;
 579                 }
 580
 581                 if (md->sanitize_names && has_name(&key)) {
 582                         sanitize_name(md, dst, src, &key, i);
 583                         continue;
 584                 }
 585
 586                 if (key.type != BTRFS_EXTENT_DATA_KEY)
 587                         continue;
 588
 589                 fi = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
 590                 extent_type = btrfs_file_extent_type(src, fi);
 591                 if (extent_type != BTRFS_FILE_EXTENT_INLINE)
 592                         continue;
 593
 594                 ptr = btrfs_file_extent_inline_start(fi);
 595                 size = btrfs_file_extent_inline_item_len(src, item);
 596                 memset(dst + ptr, 0, size);
 597         }
 598 }
 599
 600 /*
 601  * copy buffer and zero useless data in the buffer
 602  */
 603 static void copy_buffer(struct metadump_struct *md, u8 *dst,
 604                         struct extent_buffer *src)
 605 {
 606         int level;
 607         size_t size;
 608         u32 nritems;
 609
 610         memcpy(dst, src->data, src->len);
 611         if (src->start == BTRFS_SUPER_INFO_OFFSET)
 612                 return;
 613
 614         level = btrfs_header_level(src);
 615         nritems = btrfs_header_nritems(src);
 616
 617         if (nritems == 0) {
 618                 size = sizeof(struct btrfs_header);
 619                 memset(dst + size, 0, src->len - size);
 620         } else if (level == 0) {
 621                 size = btrfs_leaf_data(src) +
 622                         btrfs_item_offset_nr(src, nritems - 1) -
 623                         btrfs_item_nr_offset(nritems);
 624                 memset(dst + btrfs_item_nr_offset(nritems), 0, size);
 625                 zero_items(md, dst, src);
 626         } else {
 627                 size = offsetof(struct btrfs_node, ptrs) +
 628                         sizeof(struct btrfs_key_ptr) * nritems;
 629                 memset(dst + size, 0, src->len - size);
 630         }
 631         csum_block(dst, src->len);
 632 }
 633
 634 static void *dump_worker(void *data)
 635 {
 636         struct metadump_struct *md = (struct metadump_struct *)data;
 637         struct async_work *async;
 638         int ret;
 639
 640         while (1) {
 641                 pthread_mutex_lock(&md->mutex);
 642                 while (list_empty(&md->list)) {
 643                         if (md->done) {
 644                                 pthread_mutex_unlock(&md->mutex);
 645                                 goto out;
 646                         }
 647                         pthread_cond_wait(&md->cond, &md->mutex);
 648                 }
 649                 async = list_entry(md->list.next, struct async_work, list);
 650                 list_del_init(&async->list);
 651                 pthread_mutex_unlock(&md->mutex);
 652
 653                 if (md->compress_level > 0) {
 654                         u8 *orig = async->buffer;
 655
 656                         async->bufsize = compressBound(async->size);
 657                         async->buffer = malloc(async->bufsize);
 658                         if (!async->buffer) {
 659                                 fprintf(stderr, "Error allocing buffer\n");
 660                                 pthread_mutex_lock(&md->mutex);
 661                                 if (!md->error)
 662                                         md->error = -ENOMEM;
 663                                 pthread_mutex_unlock(&md->mutex);
 664                                 pthread_exit(NULL);
 665                         }
 666
 667                         ret = compress2(async->buffer,
 668                                          (unsigned long *)&async->bufsize,
 669                                          orig, async->size, md->compress_level);
 670
 671                         if (ret != Z_OK)
 672                                 async->error = 1;
 673
 674                         free(orig);
 675                 }
 676
 677                 pthread_mutex_lock(&md->mutex);
 678                 md->num_ready++;
 679                 pthread_mutex_unlock(&md->mutex);
 680         }
 681 out:
 682         pthread_exit(NULL);
 683 }
 684
 685 static void meta_cluster_init(struct metadump_struct *md, u64 start)
 686 {
 687         struct meta_cluster_header *header;
 688
 689         md->num_items = 0;
 690         md->num_ready = 0;
 691         header = &md->cluster->header;
 692         header->magic = cpu_to_le64(HEADER_MAGIC);
 693         header->bytenr = cpu_to_le64(start);
 694         header->nritems = cpu_to_le32(0);
 695         header->compress = md->compress_level > 0 ?
 696                            COMPRESS_ZLIB : COMPRESS_NONE;
 697 }
 698
 699 static void metadump_destroy(struct metadump_struct *md, int num_threads)
 700 {
 701         int i;
 702         struct rb_node *n;
 703
 704         pthread_mutex_lock(&md->mutex);
 705         md->done = 1;
 706         pthread_cond_broadcast(&md->cond);
 707         pthread_mutex_unlock(&md->mutex);
 708
 709         for (i = 0; i < num_threads; i++)
 710                 pthread_join(md->threads[i], NULL);
 711
 712         pthread_cond_destroy(&md->cond);
 713         pthread_mutex_destroy(&md->mutex);
 714
 715         while ((n = rb_first(&md->name_tree))) {
 716                 struct name *name;
 717
 718                 name = rb_entry(n, struct name, n);
 719                 rb_erase(n, &md->name_tree);
 720                 free(name->val);
 721                 free(name->sub);
 722                 free(name);
 723         }
 724         free(md->threads);
 725         free(md->cluster);
 726 }
 727
 728 static int metadump_init(struct metadump_struct *md, struct btrfs_root *root,
 729                          FILE *out, int num_threads, int compress_level,
 730                          int sanitize_names)
 731 {
 732         int i, ret = 0;
 733
 734         memset(md, 0, sizeof(*md));
 735         pthread_cond_init(&md->cond, NULL);
 736         pthread_mutex_init(&md->mutex, NULL);
 737         INIT_LIST_HEAD(&md->list);
 738         INIT_LIST_HEAD(&md->ordered);
 739         md->root = root;
 740         md->out = out;
 741         md->pending_start = (u64)-1;
 742         md->compress_level = compress_level;
 743         md->cluster = calloc(1, BLOCK_SIZE);
 744         md->sanitize_names = sanitize_names;
 745         if (sanitize_names > 1)
 746                 crc32c_optimization_init();
 747
 748         if (!md->cluster) {
 749                 pthread_cond_destroy(&md->cond);
 750                 pthread_mutex_destroy(&md->mutex);
 751                 return -ENOMEM;
 752         }
 753
 754         meta_cluster_init(md, 0);
 755         if (!num_threads)
 756                 return 0;
 757
 758         md->name_tree.rb_node = NULL;
 759         md->num_threads = num_threads;
 760         md->threads = calloc(num_threads, sizeof(pthread_t));
 761         if (!md->threads) {
 762                 free(md->cluster);
 763                 pthread_cond_destroy(&md->cond);
 764                 pthread_mutex_destroy(&md->mutex);
 765                 return -ENOMEM;
 766         }
 767
 768         for (i = 0; i < num_threads; i++) {
 769                 ret = pthread_create(md->threads + i, NULL, dump_worker, md);
 770                 if (ret)
 771                         break;
 772         }
 773
 774         if (ret)
 775                 metadump_destroy(md, i + 1);
 776
 777         return ret;
 778 }
 779
 780 static int write_zero(FILE *out, size_t size)
 781 {
 782         static char zero[BLOCK_SIZE];
 783         return fwrite(zero, size, 1, out);
 784 }
 785
 786 static int write_buffers(struct metadump_struct *md, u64 *next)
 787 {
 788         struct meta_cluster_header *header = &md->cluster->header;
 789         struct meta_cluster_item *item;
 790         struct async_work *async;
 791         u64 bytenr = 0;
 792         u32 nritems = 0;
 793         int ret;
 794         int err = 0;
 795
 796         if (list_empty(&md->ordered))
 797                 goto out;
 798
 799         /* wait until all buffers are compressed */
 800         while (!err && md->num_items > md->num_ready) {
 801                 struct timespec ts = {
 802                         .tv_sec = 0,
 803                         .tv_nsec = 10000000,
 804                 };
 805                 pthread_mutex_unlock(&md->mutex);
 806                 nanosleep(&ts, NULL);
 807                 pthread_mutex_lock(&md->mutex);
 808                 err = md->error;
 809         }
 810
 811         if (err) {
 812                 fprintf(stderr, "One of the threads errored out %s\n",
 813                                 strerror(err));
 814                 goto out;
 815         }
 816
 817         /* setup and write index block */
 818         list_for_each_entry(async, &md->ordered, ordered) {
 819                 item = md->cluster->items + nritems;
 820                 item->bytenr = cpu_to_le64(async->start);
 821                 item->size = cpu_to_le32(async->bufsize);
 822                 nritems++;
 823         }
 824         header->nritems = cpu_to_le32(nritems);
 825
 826         ret = fwrite(md->cluster, BLOCK_SIZE, 1, md->out);
 827         if (ret != 1) {
 828                 fprintf(stderr, "Error writing out cluster: %d\n", errno);
 829                 return -EIO;
 830         }
 831
 832         /* write buffers */
 833         bytenr += le64_to_cpu(header->bytenr) + BLOCK_SIZE;
 834         while (!list_empty(&md->ordered)) {
 835                 async = list_entry(md->ordered.next, struct async_work,
 836                                    ordered);
 837                 list_del_init(&async->ordered);
 838
 839                 bytenr += async->bufsize;
 840                 if (!err)
 841                         ret = fwrite(async->buffer, async->bufsize, 1,
 842                                      md->out);
 843                 if (ret != 1) {
 844                         err = -EIO;
 845                         ret = 0;
 846                         fprintf(stderr, "Error writing out cluster: %d\n",
 847                                 errno);
 848                 }
 849
 850                 free(async->buffer);
 851                 free(async);
 852         }
 853
 854         /* zero unused space in the last block */
 855         if (!err && bytenr & BLOCK_MASK) {
 856                 size_t size = BLOCK_SIZE - (bytenr & BLOCK_MASK);
 857
 858                 bytenr += size;
 859                 ret = write_zero(md->out, size);
 860                 if (ret != 1) {
 861                         fprintf(stderr, "Error zeroing out buffer: %d\n",
 862                                 errno);
 863                         err = -EIO;
 864                 }
 865         }
 866 out:
 867         *next = bytenr;
 868         return err;
 869 }
 870
 871 static int read_data_extent(struct metadump_struct *md,
 872                             struct async_work *async)
 873 {
 874         struct btrfs_multi_bio *multi = NULL;
 875         struct btrfs_device *device;
 876         u64 bytes_left = async->size;
 877         u64 logical = async->start;
 878         u64 offset = 0;
 879         u64 bytenr;
 880         u64 read_len;
 881         ssize_t done;
 882         int fd;
 883         int ret;
 884
 885         while (bytes_left) {
 886                 read_len = bytes_left;
 887                 ret = btrfs_map_block(&md->root->fs_info->mapping_tree, READ,
 888                                       logical, &read_len, &multi, 0, NULL);
 889                 if (ret) {
 890                         fprintf(stderr, "Couldn't map data block %d\n", ret);
 891                         return ret;
 892                 }
 893
 894                 device = multi->stripes[0].dev;
 895
 896                 if (device->fd == 0) {
 897                         fprintf(stderr,
 898                                 "Device we need to read from is not open\n");
 899                         free(multi);
 900                         return -EIO;
 901                 }
 902                 fd = device->fd;
 903                 bytenr = multi->stripes[0].physical;
 904                 free(multi);
 905
 906                 read_len = min(read_len, bytes_left);
 907                 done = pread64(fd, async->buffer+offset, read_len, bytenr);
 908                 if (done < read_len) {
 909                         if (done < 0)
 910                                 fprintf(stderr, "Error reading extent %d\n",
 911                                         errno);
 912                         else
 913                                 fprintf(stderr, "Short read\n");
 914                         return -EIO;
 915                 }
 916
 917                 bytes_left -= done;
 918                 offset += done;
 919                 logical += done;
 920         }
 921
 922         return 0;
 923 }
 924
 925 static int get_dev_fd(struct btrfs_root *root)
 926 {
 927         struct btrfs_device *dev;
 928
 929         dev = list_first_entry(&root->fs_info->fs_devices->devices,
 930                                struct btrfs_device, dev_list);
 931         return dev->fd;
 932 }
 933
 934 static int flush_pending(struct metadump_struct *md, int done)
 935 {
 936         struct async_work *async = NULL;
 937         struct extent_buffer *eb;
 938         u64 blocksize = md->root->nodesize;
 939         u64 start;
 940         u64 size;
 941         size_t offset;
 942         int ret = 0;
 943
 944         if (md->pending_size) {
 945                 async = calloc(1, sizeof(*async));
 946                 if (!async)
 947                         return -ENOMEM;
 948
 949                 async->start = md->pending_start;
 950                 async->size = md->pending_size;
 951                 async->bufsize = async->size;
 952                 async->buffer = malloc(async->bufsize);
 953                 if (!async->buffer) {
 954                         free(async);
 955                         return -ENOMEM;
 956                 }
 957                 offset = 0;
 958                 start = async->start;
 959                 size = async->size;
 960
 961                 if (md->data) {
 962                         ret = read_data_extent(md, async);
 963                         if (ret) {
 964                                 free(async->buffer);
 965                                 free(async);
 966                                 return ret;
 967                         }
 968                 }
 969
 970                 /*
 971                  * Balance can make the mapping not cover the super block, so
 972                  * just copy directly from one of the devices.
 973                  */
 974                 if (start == BTRFS_SUPER_INFO_OFFSET) {
 975                         int fd = get_dev_fd(md->root);
 976
 977                         ret = pread64(fd, async->buffer, size, start);
 978                         if (ret < size) {
 979                                 free(async->buffer);
 980                                 free(async);
 981                                 fprintf(stderr, "Error reading superblock\n");
 982                                 return -EIO;
 983                         }
 984                         size = 0;
 985                         ret = 0;
 986                 }
 987
 988                 while (!md->data && size > 0) {
 989                         u64 this_read = min(blocksize, size);
 990                         eb = read_tree_block(md->root, start, this_read, 0);
 991                         if (!extent_buffer_uptodate(eb)) {
 992                                 free(async->buffer);
 993                                 free(async);
 994                                 fprintf(stderr,
 995                                         "Error reading metadata block\n");
 996                                 return -EIO;
 997                         }
 998                         copy_buffer(md, async->buffer + offset, eb);
 999                         free_extent_buffer(eb);
1000                         start += this_read;
1001                         offset += this_read;
1002                         size -= this_read;
1003                 }
1004
1005                 md->pending_start = (u64)-1;
1006                 md->pending_size = 0;
1007         } else if (!done) {
1008                 return 0;
1009         }
1010
1011         pthread_mutex_lock(&md->mutex);
1012         if (async) {
1013                 list_add_tail(&async->ordered, &md->ordered);
1014                 md->num_items++;
1015                 if (md->compress_level > 0) {
1016                         list_add_tail(&async->list, &md->list);
1017                         pthread_cond_signal(&md->cond);
1018                 } else {
1019                         md->num_ready++;
1020                 }
1021         }
1022         if (md->num_items >= ITEMS_PER_CLUSTER || done) {
1023                 ret = write_buffers(md, &start);
1024                 if (ret)
1025                         fprintf(stderr, "Error writing buffers %d\n",
1026                                 errno);
1027                 else
1028                         meta_cluster_init(md, start);
1029         }
1030         pthread_mutex_unlock(&md->mutex);
1031         return ret;
1032 }
1033
1034 static int add_extent(u64 start, u64 size, struct metadump_struct *md,
1035                       int data)
1036 {
1037         int ret;
1038         if (md->data != data ||
1039             md->pending_size + size > MAX_PENDING_SIZE ||
1040             md->pending_start + md->pending_size != start) {
1041                 ret = flush_pending(md, 0);
1042                 if (ret)
1043                         return ret;
1044                 md->pending_start = start;
1045         }
1046         readahead_tree_block(md->root, start, size, 0);
1047         md->pending_size += size;
1048         md->data = data;
1049         return 0;
1050 }
1051
1052 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1053 static int is_tree_block(struct btrfs_root *extent_root,
1054                          struct btrfs_path *path, u64 bytenr)
1055 {
1056         struct extent_buffer *leaf;
1057         struct btrfs_key key;
1058         u64 ref_objectid;
1059         int ret;
1060
1061         leaf = path->nodes[0];
1062         while (1) {
1063                 struct btrfs_extent_ref_v0 *ref_item;
1064                 path->slots[0]++;
1065                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1066                         ret = btrfs_next_leaf(extent_root, path);
1067                         if (ret < 0)
1068                                 return ret;
1069                         if (ret > 0)
1070                                 break;
1071                         leaf = path->nodes[0];
1072                 }
1073                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1074                 if (key.objectid != bytenr)
1075                         break;
1076                 if (key.type != BTRFS_EXTENT_REF_V0_KEY)
1077                         continue;
1078                 ref_item = btrfs_item_ptr(leaf, path->slots[0],
1079                                           struct btrfs_extent_ref_v0);
1080                 ref_objectid = btrfs_ref_objectid_v0(leaf, ref_item);
1081                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID)
1082                         return 1;
1083                 break;
1084         }
1085         return 0;
1086 }
1087 #endif
1088
1089 static int copy_tree_blocks(struct btrfs_root *root, struct extent_buffer *eb,
1090                             struct metadump_struct *metadump, int root_tree)
1091 {
1092         struct extent_buffer *tmp;
1093         struct btrfs_root_item *ri;
1094         struct btrfs_key key;
1095         u64 bytenr;
1096         int level;
1097         int nritems = 0;
1098         int i = 0;
1099         int ret;
1100
1101         ret = add_extent(btrfs_header_bytenr(eb), root->leafsize, metadump, 0);
1102         if (ret) {
1103                 fprintf(stderr, "Error adding metadata block\n");
1104                 return ret;
1105         }
1106
1107         if (btrfs_header_level(eb) == 0 && !root_tree)
1108                 return 0;
1109
1110         level = btrfs_header_level(eb);
1111         nritems = btrfs_header_nritems(eb);
1112         for (i = 0; i < nritems; i++) {
1113                 if (level == 0) {
1114                         btrfs_item_key_to_cpu(eb, &key, i);
1115                         if (key.type != BTRFS_ROOT_ITEM_KEY)
1116                                 continue;
1117                         ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
1118                         bytenr = btrfs_disk_root_bytenr(eb, ri);
1119                         tmp = read_tree_block(root, bytenr, root->leafsize, 0);
1120                         if (!extent_buffer_uptodate(tmp)) {
1121                                 fprintf(stderr,
1122                                         "Error reading log root block\n");
1123                                 return -EIO;
1124                         }
1125                         ret = copy_tree_blocks(root, tmp, metadump, 0);
1126                         free_extent_buffer(tmp);
1127                         if (ret)
1128                                 return ret;
1129                 } else {
1130                         bytenr = btrfs_node_blockptr(eb, i);
1131                         tmp = read_tree_block(root, bytenr, root->leafsize, 0);
1132                         if (!extent_buffer_uptodate(tmp)) {
1133                                 fprintf(stderr, "Error reading log block\n");
1134                                 return -EIO;
1135                         }
1136                         ret = copy_tree_blocks(root, tmp, metadump, root_tree);
1137                         free_extent_buffer(tmp);
1138                         if (ret)
1139                                 return ret;
1140                 }
1141         }
1142
1143         return 0;
1144 }
1145
1146 static int copy_log_trees(struct btrfs_root *root,
1147                           struct metadump_struct *metadump,
1148                           struct btrfs_path *path)
1149 {
1150         u64 blocknr = btrfs_super_log_root(root->fs_info->super_copy);
1151
1152         if (blocknr == 0)
1153                 return 0;
1154
1155         if (!root->fs_info->log_root_tree ||
1156             !root->fs_info->log_root_tree->node) {
1157                 fprintf(stderr, "Error copying tree log, it wasn't setup\n");
1158                 return -EIO;
1159         }
1160
1161         return copy_tree_blocks(root, root->fs_info->log_root_tree->node,
1162                                 metadump, 1);
1163 }
1164
1165 static int copy_space_cache(struct btrfs_root *root,
1166                             struct metadump_struct *metadump,
1167                             struct btrfs_path *path)
1168 {
1169         struct extent_buffer *leaf;
1170         struct btrfs_file_extent_item *fi;
1171         struct btrfs_key key;
1172         u64 bytenr, num_bytes;
1173         int ret;
1174
1175         root = root->fs_info->tree_root;
1176
1177         key.objectid = 0;
1178         key.type = BTRFS_EXTENT_DATA_KEY;
1179         key.offset = 0;
1180
1181         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1182         if (ret < 0) {
1183                 fprintf(stderr, "Error searching for free space inode %d\n",
1184                         ret);
1185                 return ret;
1186         }
1187
1188         leaf = path->nodes[0];
1189
1190         while (1) {
1191                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1192                         ret = btrfs_next_leaf(root, path);
1193                         if (ret < 0) {
1194                                 fprintf(stderr, "Error going to next leaf "
1195                                         "%d\n", ret);
1196                                 return ret;
1197                         }
1198                         if (ret > 0)
1199                                 break;
1200                         leaf = path->nodes[0];
1201                 }
1202
1203                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1204                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
1205                         path->slots[0]++;
1206                         continue;
1207                 }
1208
1209                 fi = btrfs_item_ptr(leaf, path->slots[0],
1210                                     struct btrfs_file_extent_item);
1211                 if (btrfs_file_extent_type(leaf, fi) !=
1212                     BTRFS_FILE_EXTENT_REG) {
1213                         path->slots[0]++;
1214                         continue;
1215                 }
1216
1217                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1218                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1219                 ret = add_extent(bytenr, num_bytes, metadump, 1);
1220                 if (ret) {
1221                         fprintf(stderr, "Error adding space cache blocks %d\n",
1222                                 ret);
1223                         btrfs_release_path(path);
1224                         return ret;
1225                 }
1226                 path->slots[0]++;
1227         }
1228
1229         return 0;
1230 }
1231
1232 static int copy_from_extent_tree(struct metadump_struct *metadump,
1233                                  struct btrfs_path *path)
1234 {
1235         struct btrfs_root *extent_root;
1236         struct extent_buffer *leaf;
1237         struct btrfs_extent_item *ei;
1238         struct btrfs_key key;
1239         u64 bytenr;
1240         u64 num_bytes;
1241         int ret;
1242
1243         extent_root = metadump->root->fs_info->extent_root;
1244         bytenr = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
1245         key.objectid = bytenr;
1246         key.type = BTRFS_EXTENT_ITEM_KEY;
1247         key.offset = 0;
1248
1249         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1250         if (ret < 0) {
1251                 fprintf(stderr, "Error searching extent root %d\n", ret);
1252                 return ret;
1253         }
1254         ret = 0;
1255
1256         leaf = path->nodes[0];
1257
1258         while (1) {
1259                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1260                         ret = btrfs_next_leaf(extent_root, path);
1261                         if (ret < 0) {
1262                                 fprintf(stderr, "Error going to next leaf %d"
1263                                         "\n", ret);
1264                                 break;
1265                         }
1266                         if (ret > 0) {
1267                                 ret = 0;
1268                                 break;
1269                         }
1270                         leaf = path->nodes[0];
1271                 }
1272
1273                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1274                 if (key.objectid < bytenr ||
1275                     (key.type != BTRFS_EXTENT_ITEM_KEY &&
1276                      key.type != BTRFS_METADATA_ITEM_KEY)) {
1277                         path->slots[0]++;
1278                         continue;
1279                 }
1280
1281                 bytenr = key.objectid;
1282                 if (key.type == BTRFS_METADATA_ITEM_KEY)
1283                         num_bytes = extent_root->leafsize;
1284                 else
1285                         num_bytes = key.offset;
1286
1287                 if (btrfs_item_size_nr(leaf, path->slots[0]) > sizeof(*ei)) {
1288                         ei = btrfs_item_ptr(leaf, path->slots[0],
1289                                             struct btrfs_extent_item);
1290                         if (btrfs_extent_flags(leaf, ei) &
1291                             BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1292                                 ret = add_extent(bytenr, num_bytes, metadump,
1293                                                  0);
1294                                 if (ret) {
1295                                         fprintf(stderr, "Error adding block "
1296                                                 "%d\n", ret);
1297                                         break;
1298                                 }
1299                         }
1300                 } else {
1301 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1302                         ret = is_tree_block(extent_root, path, bytenr);
1303                         if (ret < 0) {
1304                                 fprintf(stderr, "Error checking tree block "
1305                                         "%d\n", ret);
1306                                 break;
1307                         }
1308
1309                         if (ret) {
1310                                 ret = add_extent(bytenr, num_bytes, metadump,
1311                                                  0);
1312                                 if (ret) {
1313                                         fprintf(stderr, "Error adding block "
1314                                                 "%d\n", ret);
1315                                         break;
1316                                 }
1317                         }
1318                         ret = 0;
1319 #else
1320                         fprintf(stderr, "Either extent tree corruption or "
1321                                 "you haven't built with V0 support\n");
1322                         ret = -EIO;
1323                         break;
1324 #endif
1325                 }
1326                 bytenr += num_bytes;
1327         }
1328
1329         btrfs_release_path(path);
1330
1331         return ret;
1332 }
1333
1334 static int create_metadump(const char *input, FILE *out, int num_threads,
1335                            int compress_level, int sanitize, int walk_trees)
1336 {
1337         struct btrfs_root *root;
1338         struct btrfs_path *path = NULL;
1339         struct metadump_struct metadump;
1340         int ret;
1341         int err = 0;
1342
1343         root = open_ctree(input, 0, 0);
1344         if (!root) {
1345                 fprintf(stderr, "Open ctree failed\n");
1346                 return -EIO;
1347         }
1348
1349         BUG_ON(root->nodesize != root->leafsize);
1350
1351         ret = metadump_init(&metadump, root, out, num_threads,
1352                             compress_level, sanitize);
1353         if (ret) {
1354                 fprintf(stderr, "Error initing metadump %d\n", ret);
1355                 close_ctree(root);
1356                 return ret;
1357         }
1358
1359         ret = add_extent(BTRFS_SUPER_INFO_OFFSET, BTRFS_SUPER_INFO_SIZE,
1360                         &metadump, 0);
1361         if (ret) {
1362                 fprintf(stderr, "Error adding metadata %d\n", ret);
1363                 err = ret;
1364                 goto out;
1365         }
1366
1367         path = btrfs_alloc_path();
1368         if (!path) {
1369                 fprintf(stderr, "Out of memory allocing path\n");
1370                 err = -ENOMEM;
1371                 goto out;
1372         }
1373
1374         if (walk_trees) {
1375                 ret = copy_tree_blocks(root, root->fs_info->chunk_root->node,
1376                                        &metadump, 1);
1377                 if (ret) {
1378                         err = ret;
1379                         goto out;
1380                 }
1381
1382                 ret = copy_tree_blocks(root, root->fs_info->tree_root->node,
1383                                        &metadump, 1);
1384                 if (ret) {
1385                         err = ret;
1386                         goto out;
1387                 }
1388         } else {
1389                 ret = copy_from_extent_tree(&metadump, path);
1390                 if (ret) {
1391                         err = ret;
1392                         goto out;
1393                 }
1394         }
1395
1396         ret = copy_log_trees(root, &metadump, path);
1397         if (ret) {
1398                 err = ret;
1399                 goto out;
1400         }
1401
1402         ret = copy_space_cache(root, &metadump, path);
1403 out:
1404         ret = flush_pending(&metadump, 1);
1405         if (ret) {
1406                 if (!err)
1407                         err = ret;
1408                 fprintf(stderr, "Error flushing pending %d\n", ret);
1409         }
1410
1411         metadump_destroy(&metadump, num_threads);
1412
1413         btrfs_free_path(path);
1414         ret = close_ctree(root);
1415         return err ? err : ret;
1416 }
1417
1418 static void update_super_old(u8 *buffer)
1419 {
1420         struct btrfs_super_block *super = (struct btrfs_super_block *)buffer;
1421         struct btrfs_chunk *chunk;
1422         struct btrfs_disk_key *key;
1423         u32 sectorsize = btrfs_super_sectorsize(super);
1424         u64 flags = btrfs_super_flags(super);
1425
1426         flags |= BTRFS_SUPER_FLAG_METADUMP;
1427         btrfs_set_super_flags(super, flags);
1428
1429         key = (struct btrfs_disk_key *)(super->sys_chunk_array);
1430         chunk = (struct btrfs_chunk *)(super->sys_chunk_array +
1431                                        sizeof(struct btrfs_disk_key));
1432
1433         btrfs_set_disk_key_objectid(key, BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1434         btrfs_set_disk_key_type(key, BTRFS_CHUNK_ITEM_KEY);
1435         btrfs_set_disk_key_offset(key, 0);
1436
1437         btrfs_set_stack_chunk_length(chunk, (u64)-1);
1438         btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
1439         btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
1440         btrfs_set_stack_chunk_type(chunk, BTRFS_BLOCK_GROUP_SYSTEM);
1441         btrfs_set_stack_chunk_io_align(chunk, sectorsize);
1442         btrfs_set_stack_chunk_io_width(chunk, sectorsize);
1443         btrfs_set_stack_chunk_sector_size(chunk, sectorsize);
1444         btrfs_set_stack_chunk_num_stripes(chunk, 1);
1445         btrfs_set_stack_chunk_sub_stripes(chunk, 0);
1446         chunk->stripe.devid = super->dev_item.devid;
1447         btrfs_set_stack_stripe_offset(&chunk->stripe, 0);
1448         memcpy(chunk->stripe.dev_uuid, super->dev_item.uuid, BTRFS_UUID_SIZE);
1449         btrfs_set_super_sys_array_size(super, sizeof(*key) + sizeof(*chunk));
1450         csum_block(buffer, BTRFS_SUPER_INFO_SIZE);
1451 }
1452
1453 static int update_super(struct mdrestore_struct *mdres, u8 *buffer)
1454 {
1455         struct btrfs_super_block *super = (struct btrfs_super_block *)buffer;
1456         struct btrfs_chunk *chunk;
1457         struct btrfs_disk_key *disk_key;
1458         struct btrfs_key key;
1459         u64 flags = btrfs_super_flags(super);
1460         u32 new_array_size = 0;
1461         u32 array_size;
1462         u32 cur = 0;
1463         u8 *ptr, *write_ptr;
1464         int old_num_stripes;
1465
1466         write_ptr = ptr = super->sys_chunk_array;
1467         array_size = btrfs_super_sys_array_size(super);
1468
1469         while (cur < array_size) {
1470                 disk_key = (struct btrfs_disk_key *)ptr;
1471                 btrfs_disk_key_to_cpu(&key, disk_key);
1472
1473                 new_array_size += sizeof(*disk_key);
1474                 memmove(write_ptr, ptr, sizeof(*disk_key));
1475
1476                 write_ptr += sizeof(*disk_key);
1477                 ptr += sizeof(*disk_key);
1478                 cur += sizeof(*disk_key);
1479
1480                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1481                         u64 physical, size = 0;
1482
1483                         chunk = (struct btrfs_chunk *)ptr;
1484                         old_num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1485                         chunk = (struct btrfs_chunk *)write_ptr;
1486
1487                         memmove(write_ptr, ptr, sizeof(*chunk));
1488                         btrfs_set_stack_chunk_num_stripes(chunk, 1);
1489                         btrfs_set_stack_chunk_sub_stripes(chunk, 0);
1490                         btrfs_set_stack_chunk_type(chunk,
1491                                                    BTRFS_BLOCK_GROUP_SYSTEM);
1492                         btrfs_set_stack_stripe_devid(&chunk->stripe,
1493                                                      super->dev_item.devid);
1494                         physical = logical_to_physical(mdres, key.offset,
1495                                                        &size);
1496                         if (size != (u64)-1)
1497                                 btrfs_set_stack_stripe_offset(&chunk->stripe,
1498                                                               physical);
1499                         memcpy(chunk->stripe.dev_uuid, super->dev_item.uuid,
1500                                BTRFS_UUID_SIZE);
1501                         new_array_size += sizeof(*chunk);
1502                 } else {
1503                         fprintf(stderr, "Bogus key in the sys chunk array "
1504                                 "%d\n", key.type);
1505                         return -EIO;
1506                 }
1507                 write_ptr += sizeof(*chunk);
1508                 ptr += btrfs_chunk_item_size(old_num_stripes);
1509                 cur += btrfs_chunk_item_size(old_num_stripes);
1510         }
1511
1512         if (mdres->clear_space_cache)
1513                 btrfs_set_super_cache_generation(super, 0);
1514
1515         flags |= BTRFS_SUPER_FLAG_METADUMP_V2;
1516         btrfs_set_super_flags(super, flags);
1517         btrfs_set_super_sys_array_size(super, new_array_size);
1518         csum_block(buffer, BTRFS_SUPER_INFO_SIZE);
1519
1520         return 0;
1521 }
1522
1523 static struct extent_buffer *alloc_dummy_eb(u64 bytenr, u32 size)
1524 {
1525         struct extent_buffer *eb;
1526
1527         eb = malloc(sizeof(struct extent_buffer) + size);
1528         if (!eb)
1529                 return NULL;
1530         memset(eb, 0, sizeof(struct extent_buffer) + size);
1531
1532         eb->start = bytenr;
1533         eb->len = size;
1534         return eb;
1535 }
1536
1537 static void truncate_item(struct extent_buffer *eb, int slot, u32 new_size)
1538 {
1539         struct btrfs_item *item;
1540         u32 nritems;
1541         u32 old_size;
1542         u32 old_data_start;
1543         u32 size_diff;
1544         u32 data_end;
1545         int i;
1546
1547         old_size = btrfs_item_size_nr(eb, slot);
1548         if (old_size == new_size)
1549                 return;
1550
1551         nritems = btrfs_header_nritems(eb);
1552         data_end = btrfs_item_offset_nr(eb, nritems - 1);
1553
1554         old_data_start = btrfs_item_offset_nr(eb, slot);
1555         size_diff = old_size - new_size;
1556
1557         for (i = slot; i < nritems; i++) {
1558                 u32 ioff;
1559                 item = btrfs_item_nr(i);
1560                 ioff = btrfs_item_offset(eb, item);
1561                 btrfs_set_item_offset(eb, item, ioff + size_diff);
1562         }
1563
1564         memmove_extent_buffer(eb, btrfs_leaf_data(eb) + data_end + size_diff,
1565                               btrfs_leaf_data(eb) + data_end,
1566                               old_data_start + new_size - data_end);
1567         item = btrfs_item_nr(slot);
1568         btrfs_set_item_size(eb, item, new_size);
1569 }
1570
1571 static int fixup_chunk_tree_block(struct mdrestore_struct *mdres,
1572                                   struct async_work *async, u8 *buffer,
1573                                   size_t size)
1574 {
1575         struct extent_buffer *eb;
1576         size_t size_left = size;
1577         u64 bytenr = async->start;
1578         int i;
1579
1580         if (size_left % mdres->leafsize)
1581                 return 0;
1582
1583         eb = alloc_dummy_eb(bytenr, mdres->leafsize);
1584         if (!eb)
1585                 return -ENOMEM;
1586
1587         while (size_left) {
1588                 eb->start = bytenr;
1589                 memcpy(eb->data, buffer, mdres->leafsize);
1590
1591                 if (btrfs_header_bytenr(eb) != bytenr)
1592                         break;
1593                 if (memcmp(mdres->fsid,
1594                            eb->data + offsetof(struct btrfs_header, fsid),
1595                            BTRFS_FSID_SIZE))
1596                         break;
1597
1598                 if (btrfs_header_owner(eb) != BTRFS_CHUNK_TREE_OBJECTID)
1599                         goto next;
1600
1601                 if (btrfs_header_level(eb) != 0)
1602                         goto next;
1603
1604                 for (i = 0; i < btrfs_header_nritems(eb); i++) {
1605                         struct btrfs_chunk chunk;
1606                         struct btrfs_key key;
1607                         u64 type, physical, size = (u64)-1;
1608
1609                         btrfs_item_key_to_cpu(eb, &key, i);
1610                         if (key.type != BTRFS_CHUNK_ITEM_KEY)
1611                                 continue;
1612                         truncate_item(eb, i, sizeof(chunk));
1613                         read_extent_buffer(eb, &chunk,
1614                                            btrfs_item_ptr_offset(eb, i),
1615                                            sizeof(chunk));
1616
1617                         size = 0;
1618                         physical = logical_to_physical(mdres, key.offset,
1619                                                        &size);
1620
1621                         /* Zero out the RAID profile */
1622                         type = btrfs_stack_chunk_type(&chunk);
1623                         type &= (BTRFS_BLOCK_GROUP_DATA |
1624                                  BTRFS_BLOCK_GROUP_SYSTEM |
1625                                  BTRFS_BLOCK_GROUP_METADATA |
1626                                  BTRFS_BLOCK_GROUP_DUP);
1627                         btrfs_set_stack_chunk_type(&chunk, type);
1628
1629                         btrfs_set_stack_chunk_num_stripes(&chunk, 1);
1630                         btrfs_set_stack_chunk_sub_stripes(&chunk, 0);
1631                         btrfs_set_stack_stripe_devid(&chunk.stripe, mdres->devid);
1632                         if (size != (u64)-1)
1633                                 btrfs_set_stack_stripe_offset(&chunk.stripe,
1634                                                               physical);
1635                         memcpy(chunk.stripe.dev_uuid, mdres->uuid,
1636                                BTRFS_UUID_SIZE);
1637                         write_extent_buffer(eb, &chunk,
1638                                             btrfs_item_ptr_offset(eb, i),
1639                                             sizeof(chunk));
1640                 }
1641                 memcpy(buffer, eb->data, eb->len);
1642                 csum_block(buffer, eb->len);
1643 next:
1644                 size_left -= mdres->leafsize;
1645                 buffer += mdres->leafsize;
1646                 bytenr += mdres->leafsize;
1647         }
1648
1649         free(eb);
1650         return 0;
1651 }
1652
1653 static void write_backup_supers(int fd, u8 *buf)
1654 {
1655         struct btrfs_super_block *super = (struct btrfs_super_block *)buf;
1656         struct stat st;
1657         u64 size;
1658         u64 bytenr;
1659         int i;
1660         int ret;
1661
1662         if (fstat(fd, &st)) {
1663                 fprintf(stderr, "Couldn't stat restore point, won't be able "
1664                         "to write backup supers: %d\n", errno);
1665                 return;
1666         }
1667
1668         size = btrfs_device_size(fd, &st);
1669
1670         for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1671                 bytenr = btrfs_sb_offset(i);
1672                 if (bytenr + BTRFS_SUPER_INFO_SIZE > size)
1673                         break;
1674                 btrfs_set_super_bytenr(super, bytenr);
1675                 csum_block(buf, BTRFS_SUPER_INFO_SIZE);
1676                 ret = pwrite64(fd, buf, BTRFS_SUPER_INFO_SIZE, bytenr);
1677                 if (ret < BTRFS_SUPER_INFO_SIZE) {
1678                         if (ret < 0)
1679                                 fprintf(stderr, "Problem writing out backup "
1680                                         "super block %d, err %d\n", i, errno);
1681                         else
1682                                 fprintf(stderr, "Short write writing out "
1683                                         "backup super block\n");
1684                         break;
1685                 }
1686         }
1687 }
1688
1689 static void *restore_worker(void *data)
1690 {
1691         struct mdrestore_struct *mdres = (struct mdrestore_struct *)data;
1692         struct async_work *async;
1693         size_t size;
1694         u8 *buffer;
1695         u8 *outbuf;
1696         int outfd;
1697         int ret;
1698         int compress_size = MAX_PENDING_SIZE * 4;
1699
1700         outfd = fileno(mdres->out);
1701         buffer = malloc(compress_size);
1702         if (!buffer) {
1703                 fprintf(stderr, "Error allocing buffer\n");
1704                 pthread_mutex_lock(&mdres->mutex);
1705                 if (!mdres->error)
1706                         mdres->error = -ENOMEM;
1707                 pthread_mutex_unlock(&mdres->mutex);
1708                 pthread_exit(NULL);
1709         }
1710
1711         while (1) {
1712                 u64 bytenr;
1713                 off_t offset = 0;
1714                 int err = 0;
1715
1716                 pthread_mutex_lock(&mdres->mutex);
1717                 while (!mdres->leafsize || list_empty(&mdres->list)) {
1718                         if (mdres->done) {
1719                                 pthread_mutex_unlock(&mdres->mutex);
1720                                 goto out;
1721                         }
1722                         pthread_cond_wait(&mdres->cond, &mdres->mutex);
1723                 }
1724                 async = list_entry(mdres->list.next, struct async_work, list);
1725                 list_del_init(&async->list);
1726                 pthread_mutex_unlock(&mdres->mutex);
1727
1728                 if (mdres->compress_method == COMPRESS_ZLIB) {
1729                         size = compress_size;
1730                         ret = uncompress(buffer, (unsigned long *)&size,
1731                                          async->buffer, async->bufsize);
1732                         if (ret != Z_OK) {
1733                                 fprintf(stderr, "Error decompressing %d\n",
1734                                         ret);
1735                                 err = -EIO;
1736                         }
1737                         outbuf = buffer;
1738                 } else {
1739                         outbuf = async->buffer;
1740                         size = async->bufsize;
1741                 }
1742
1743                 if (!mdres->multi_devices) {
1744                         if (async->start == BTRFS_SUPER_INFO_OFFSET) {
1745                                 if (mdres->old_restore) {
1746                                         update_super_old(outbuf);
1747                                 } else {
1748                                         ret = update_super(mdres, outbuf);
1749                                         if (ret)
1750                                                 err = ret;
1751                                 }
1752                         } else if (!mdres->old_restore) {
1753                                 ret = fixup_chunk_tree_block(mdres, async, outbuf, size);
1754                                 if (ret)
1755                                         err = ret;
1756                         }
1757                 }
1758
1759                 if (!mdres->fixup_offset) {
1760                         while (size) {
1761                                 u64 chunk_size = size;
1762                                 if (!mdres->multi_devices && !mdres->old_restore)
1763                                         bytenr = logical_to_physical(mdres,
1764                                                                      async->start + offset,
1765                                                                      &chunk_size);
1766                                 else
1767                                         bytenr = async->start + offset;
1768
1769                                 ret = pwrite64(outfd, outbuf+offset, chunk_size,
1770                                                bytenr);
1771                                 if (ret != chunk_size) {
1772                                         if (ret < 0) {
1773                                                 fprintf(stderr, "Error writing to "
1774                                                         "device %d\n", errno);
1775                                                 err = errno;
1776                                                 break;
1777                                         } else {
1778                                                 fprintf(stderr, "Short write\n");
1779                                                 err = -EIO;
1780                                                 break;
1781                                         }
1782                                 }
1783                                 size -= chunk_size;
1784                                 offset += chunk_size;
1785                         }
1786                 } else if (async->start != BTRFS_SUPER_INFO_OFFSET) {
1787                         ret = write_data_to_disk(mdres->info, outbuf, async->start, size, 0);
1788                         if (ret) {
1789                                 printk("Error write data\n");
1790                                 exit(1);
1791                         }
1792                 }
1793
1794
1795                 /* backup super blocks are already there at fixup_offset stage */
1796                 if (!mdres->multi_devices && async->start == BTRFS_SUPER_INFO_OFFSET)
1797                         write_backup_supers(outfd, outbuf);
1798
1799                 pthread_mutex_lock(&mdres->mutex);
1800                 if (err && !mdres->error)
1801                         mdres->error = err;
1802                 mdres->num_items--;
1803                 pthread_mutex_unlock(&mdres->mutex);
1804
1805                 free(async->buffer);
1806                 free(async);
1807         }
1808 out:
1809         free(buffer);
1810         pthread_exit(NULL);
1811 }
1812
1813 static void mdrestore_destroy(struct mdrestore_struct *mdres, int num_threads)
1814 {
1815         struct rb_node *n;
1816         int i;
1817
1818         while ((n = rb_first(&mdres->chunk_tree))) {
1819                 struct fs_chunk *entry;
1820
1821                 entry = rb_entry(n, struct fs_chunk, l);
1822                 rb_erase(n, &mdres->chunk_tree);
1823                 rb_erase(&entry->p, &mdres->physical_tree);
1824                 free(entry);
1825         }
1826         pthread_mutex_lock(&mdres->mutex);
1827         mdres->done = 1;
1828         pthread_cond_broadcast(&mdres->cond);
1829         pthread_mutex_unlock(&mdres->mutex);
1830
1831         for (i = 0; i < num_threads; i++)
1832                 pthread_join(mdres->threads[i], NULL);
1833
1834         pthread_cond_destroy(&mdres->cond);
1835         pthread_mutex_destroy(&mdres->mutex);
1836         free(mdres->threads);
1837 }
1838
1839 static int mdrestore_init(struct mdrestore_struct *mdres,
1840                           FILE *in, FILE *out, int old_restore,
1841                           int num_threads, int fixup_offset,
1842                           struct btrfs_fs_info *info, int multi_devices)
1843 {
1844         int i, ret = 0;
1845
1846         memset(mdres, 0, sizeof(*mdres));
1847         pthread_cond_init(&mdres->cond, NULL);
1848         pthread_mutex_init(&mdres->mutex, NULL);
1849         INIT_LIST_HEAD(&mdres->list);
1850         INIT_LIST_HEAD(&mdres->overlapping_chunks);
1851         mdres->in = in;
1852         mdres->out = out;
1853         mdres->old_restore = old_restore;
1854         mdres->chunk_tree.rb_node = NULL;
1855         mdres->fixup_offset = fixup_offset;
1856         mdres->info = info;
1857         mdres->multi_devices = multi_devices;
1858         mdres->clear_space_cache = 0;
1859         mdres->last_physical_offset = 0;
1860         mdres->alloced_chunks = 0;
1861
1862         if (!num_threads)
1863                 return 0;
1864
1865         mdres->num_threads = num_threads;
1866         mdres->threads = calloc(num_threads, sizeof(pthread_t));
1867         if (!mdres->threads)
1868                 return -ENOMEM;
1869         for (i = 0; i < num_threads; i++) {
1870                 ret = pthread_create(mdres->threads + i, NULL, restore_worker,
1871                                      mdres);
1872                 if (ret)
1873                         break;
1874         }
1875         if (ret)
1876                 mdrestore_destroy(mdres, i + 1);
1877         return ret;
1878 }
1879
1880 static int fill_mdres_info(struct mdrestore_struct *mdres,
1881                            struct async_work *async)
1882 {
1883         struct btrfs_super_block *super;
1884         u8 *buffer = NULL;
1885         u8 *outbuf;
1886         int ret;
1887
1888         /* We've already been initialized */
1889         if (mdres->leafsize)
1890                 return 0;
1891
1892         if (mdres->compress_method == COMPRESS_ZLIB) {
1893                 size_t size = MAX_PENDING_SIZE * 2;
1894
1895                 buffer = malloc(MAX_PENDING_SIZE * 2);
1896                 if (!buffer)
1897                         return -ENOMEM;
1898                 ret = uncompress(buffer, (unsigned long *)&size,
1899                                  async->buffer, async->bufsize);
1900                 if (ret != Z_OK) {
1901                         fprintf(stderr, "Error decompressing %d\n", ret);
1902                         free(buffer);
1903                         return -EIO;
1904                 }
1905                 outbuf = buffer;
1906         } else {
1907                 outbuf = async->buffer;
1908         }
1909
1910         super = (struct btrfs_super_block *)outbuf;
1911         mdres->leafsize = btrfs_super_leafsize(super);
1912         memcpy(mdres->fsid, super->fsid, BTRFS_FSID_SIZE);
1913         memcpy(mdres->uuid, super->dev_item.uuid,
1914                        BTRFS_UUID_SIZE);
1915         mdres->devid = le64_to_cpu(super->dev_item.devid);
1916         free(buffer);
1917         return 0;
1918 }
1919
1920 static int add_cluster(struct meta_cluster *cluster,
1921                        struct mdrestore_struct *mdres, u64 *next)
1922 {
1923         struct meta_cluster_item *item;
1924         struct meta_cluster_header *header = &cluster->header;
1925         struct async_work *async;
1926         u64 bytenr;
1927         u32 i, nritems;
1928         int ret;
1929
1930         mdres->compress_method = header->compress;
1931
1932         bytenr = le64_to_cpu(header->bytenr) + BLOCK_SIZE;
1933         nritems = le32_to_cpu(header->nritems);
1934         for (i = 0; i < nritems; i++) {
1935                 item = &cluster->items[i];
1936                 async = calloc(1, sizeof(*async));
1937                 if (!async) {
1938                         fprintf(stderr, "Error allocating async\n");
1939                         return -ENOMEM;
1940                 }
1941                 async->start = le64_to_cpu(item->bytenr);
1942                 async->bufsize = le32_to_cpu(item->size);
1943                 async->buffer = malloc(async->bufsize);
1944                 if (!async->buffer) {
1945                         fprintf(stderr, "Error allocing async buffer\n");
1946                         free(async);
1947                         return -ENOMEM;
1948                 }
1949                 ret = fread(async->buffer, async->bufsize, 1, mdres->in);
1950                 if (ret != 1) {
1951                         fprintf(stderr, "Error reading buffer %d\n", errno);
1952                         free(async->buffer);
1953                         free(async);
1954                         return -EIO;
1955                 }
1956                 bytenr += async->bufsize;
1957
1958                 pthread_mutex_lock(&mdres->mutex);
1959                 if (async->start == BTRFS_SUPER_INFO_OFFSET) {
1960                         ret = fill_mdres_info(mdres, async);
1961                         if (ret) {
1962                                 fprintf(stderr, "Error setting up restore\n");
1963                                 pthread_mutex_unlock(&mdres->mutex);
1964                                 free(async->buffer);
1965                                 free(async);
1966                                 return ret;
1967                         }
1968                 }
1969                 list_add_tail(&async->list, &mdres->list);
1970                 mdres->num_items++;
1971                 pthread_cond_signal(&mdres->cond);
1972                 pthread_mutex_unlock(&mdres->mutex);
1973         }
1974         if (bytenr & BLOCK_MASK) {
1975                 char buffer[BLOCK_MASK];
1976                 size_t size = BLOCK_SIZE - (bytenr & BLOCK_MASK);
1977
1978                 bytenr += size;
1979                 ret = fread(buffer, size, 1, mdres->in);
1980                 if (ret != 1) {
1981                         fprintf(stderr, "Error reading in buffer %d\n", errno);
1982                         return -EIO;
1983                 }
1984         }
1985         *next = bytenr;
1986         return 0;
1987 }
1988
1989 static int wait_for_worker(struct mdrestore_struct *mdres)
1990 {
1991         int ret = 0;
1992
1993         pthread_mutex_lock(&mdres->mutex);
1994         ret = mdres->error;
1995         while (!ret && mdres->num_items > 0) {
1996                 struct timespec ts = {
1997                         .tv_sec = 0,
1998                         .tv_nsec = 10000000,
1999                 };
2000                 pthread_mutex_unlock(&mdres->mutex);
2001                 nanosleep(&ts, NULL);
2002                 pthread_mutex_lock(&mdres->mutex);
2003                 ret = mdres->error;
2004         }
2005         pthread_mutex_unlock(&mdres->mutex);
2006         return ret;
2007 }
2008
2009 static int read_chunk_block(struct mdrestore_struct *mdres, u8 *buffer,
2010                             u64 bytenr, u64 item_bytenr, u32 bufsize,
2011                             u64 cluster_bytenr)
2012 {
2013         struct extent_buffer *eb;
2014         int ret = 0;
2015         int i;
2016
2017         eb = alloc_dummy_eb(bytenr, mdres->leafsize);
2018         if (!eb) {
2019                 ret = -ENOMEM;
2020                 goto out;
2021         }
2022
2023         while (item_bytenr != bytenr) {
2024                 buffer += mdres->leafsize;
2025                 item_bytenr += mdres->leafsize;
2026         }
2027
2028         memcpy(eb->data, buffer, mdres->leafsize);
2029         if (btrfs_header_bytenr(eb) != bytenr) {
2030                 fprintf(stderr, "Eb bytenr doesn't match found bytenr\n");
2031                 ret = -EIO;
2032                 goto out;
2033         }
2034
2035         if (memcmp(mdres->fsid, eb->data + offsetof(struct btrfs_header, fsid),
2036                    BTRFS_FSID_SIZE)) {
2037                 fprintf(stderr, "Fsid doesn't match\n");
2038                 ret = -EIO;
2039                 goto out;
2040         }
2041
2042         if (btrfs_header_owner(eb) != BTRFS_CHUNK_TREE_OBJECTID) {
2043                 fprintf(stderr, "Does not belong to the chunk tree\n");
2044                 ret = -EIO;
2045                 goto out;
2046         }
2047
2048         for (i = 0; i < btrfs_header_nritems(eb); i++) {
2049                 struct btrfs_chunk chunk;
2050                 struct fs_chunk *fs_chunk;
2051                 struct btrfs_key key;
2052
2053                 if (btrfs_header_level(eb)) {
2054                         u64 blockptr = btrfs_node_blockptr(eb, i);
2055
2056                         ret = search_for_chunk_blocks(mdres, blockptr,
2057                                                       cluster_bytenr);
2058                         if (ret)
2059                                 break;
2060                         continue;
2061                 }
2062
2063                 /* Yay a leaf!  We loves leafs! */
2064                 btrfs_item_key_to_cpu(eb, &key, i);
2065                 if (key.type != BTRFS_CHUNK_ITEM_KEY)
2066                         continue;
2067
2068                 fs_chunk = malloc(sizeof(struct fs_chunk));
2069                 if (!fs_chunk) {
2070                         fprintf(stderr, "Erorr allocating chunk\n");
2071                         ret = -ENOMEM;
2072                         break;
2073                 }
2074                 memset(fs_chunk, 0, sizeof(*fs_chunk));
2075                 read_extent_buffer(eb, &chunk, btrfs_item_ptr_offset(eb, i),
2076                                    sizeof(chunk));
2077
2078                 fs_chunk->logical = key.offset;
2079                 fs_chunk->physical = btrfs_stack_stripe_offset(&chunk.stripe);
2080                 fs_chunk->bytes = btrfs_stack_chunk_length(&chunk);
2081                 INIT_LIST_HEAD(&fs_chunk->list);
2082                 if (tree_search(&mdres->physical_tree, &fs_chunk->p,
2083                                 physical_cmp, 1) != NULL)
2084                         list_add(&fs_chunk->list, &mdres->overlapping_chunks);
2085                 else
2086                         tree_insert(&mdres->physical_tree, &fs_chunk->p,
2087                                     physical_cmp);
2088                 if (fs_chunk->physical + fs_chunk->bytes >
2089                     mdres->last_physical_offset)
2090                         mdres->last_physical_offset = fs_chunk->physical +
2091                                 fs_chunk->bytes;
2092                 mdres->alloced_chunks += fs_chunk->bytes;
2093                 tree_insert(&mdres->chunk_tree, &fs_chunk->l, chunk_cmp);
2094         }
2095 out:
2096         free(eb);
2097         return ret;
2098 }
2099
2100 /* If you have to ask you aren't worthy */
2101 static int search_for_chunk_blocks(struct mdrestore_struct *mdres,
2102                                    u64 search, u64 cluster_bytenr)
2103 {
2104         struct meta_cluster *cluster;
2105         struct meta_cluster_header *header;
2106         struct meta_cluster_item *item;
2107         u64 current_cluster = cluster_bytenr, bytenr;
2108         u64 item_bytenr;
2109         u32 bufsize, nritems, i;
2110         u32 max_size = MAX_PENDING_SIZE * 2;
2111         u8 *buffer, *tmp = NULL;
2112         int ret = 0;
2113
2114         cluster = malloc(BLOCK_SIZE);
2115         if (!cluster) {
2116                 fprintf(stderr, "Error allocating cluster\n");
2117                 return -ENOMEM;
2118         }
2119
2120         buffer = malloc(max_size);
2121         if (!buffer) {
2122                 fprintf(stderr, "Error allocing buffer\n");
2123                 free(cluster);
2124                 return -ENOMEM;
2125         }
2126
2127         if (mdres->compress_method == COMPRESS_ZLIB) {
2128                 tmp = malloc(max_size);
2129                 if (!tmp) {
2130                         fprintf(stderr, "Error allocing tmp buffer\n");
2131                         free(cluster);
2132                         free(buffer);
2133                         return -ENOMEM;
2134                 }
2135         }
2136
2137         bytenr = current_cluster;
2138         while (1) {
2139                 if (fseek(mdres->in, current_cluster, SEEK_SET)) {
2140                         fprintf(stderr, "Error seeking: %d\n", errno);
2141                         ret = -EIO;
2142                         break;
2143                 }
2144
2145                 ret = fread(cluster, BLOCK_SIZE, 1, mdres->in);
2146                 if (ret == 0) {
2147                         if (cluster_bytenr != 0) {
2148                                 cluster_bytenr = 0;
2149                                 current_cluster = 0;
2150                                 bytenr = 0;
2151                                 continue;
2152                         }
2153                         printf("ok this is where we screwed up?\n");
2154                         ret = -EIO;
2155                         break;
2156                 } else if (ret < 0) {
2157                         fprintf(stderr, "Error reading image\n");
2158                         break;
2159                 }
2160                 ret = 0;
2161
2162                 header = &cluster->header;
2163                 if (le64_to_cpu(header->magic) != HEADER_MAGIC ||
2164                     le64_to_cpu(header->bytenr) != current_cluster) {
2165                         fprintf(stderr, "bad header in metadump image\n");
2166                         ret = -EIO;
2167                         break;
2168                 }
2169
2170                 bytenr += BLOCK_SIZE;
2171                 nritems = le32_to_cpu(header->nritems);
2172                 for (i = 0; i < nritems; i++) {
2173                         size_t size;
2174
2175                         item = &cluster->items[i];
2176                         bufsize = le32_to_cpu(item->size);
2177                         item_bytenr = le64_to_cpu(item->bytenr);
2178
2179                         if (bufsize > max_size) {
2180                                 fprintf(stderr, "item %u size %u too big\n",
2181                                         i, bufsize);
2182                                 ret = -EIO;
2183                                 break;
2184                         }
2185
2186                         if (mdres->compress_method == COMPRESS_ZLIB) {
2187                                 ret = fread(tmp, bufsize, 1, mdres->in);
2188                                 if (ret != 1) {
2189                                         fprintf(stderr, "Error reading: %d\n",
2190                                                 errno);
2191                                         ret = -EIO;
2192                                         break;
2193                                 }
2194
2195                                 size = max_size;
2196                                 ret = uncompress(buffer,
2197                                                  (unsigned long *)&size, tmp,
2198                                                  bufsize);
2199                                 if (ret != Z_OK) {
2200                                         fprintf(stderr, "Error decompressing "
2201                                                 "%d\n", ret);
2202                                         ret = -EIO;
2203                                         break;
2204                                 }
2205                         } else {
2206                                 ret = fread(buffer, bufsize, 1, mdres->in);
2207                                 if (ret != 1) {
2208                                         fprintf(stderr, "Error reading: %d\n",
2209                                                 errno);
2210                                         ret = -EIO;
2211                                         break;
2212                                 }
2213                                 size = bufsize;
2214                         }
2215                         ret = 0;
2216
2217                         if (item_bytenr <= search &&
2218                             item_bytenr + size > search) {
2219                                 ret = read_chunk_block(mdres, buffer, search,
2220                                                        item_bytenr, size,
2221                                                        current_cluster);
2222                                 if (!ret)
2223                                         ret = 1;
2224                                 break;
2225                         }
2226                         bytenr += bufsize;
2227                 }
2228                 if (ret) {
2229                         if (ret > 0)
2230                                 ret = 0;
2231                         break;
2232                 }
2233                 if (bytenr & BLOCK_MASK)
2234                         bytenr += BLOCK_SIZE - (bytenr & BLOCK_MASK);
2235                 current_cluster = bytenr;
2236         }
2237
2238         free(tmp);
2239         free(buffer);
2240         free(cluster);
2241         return ret;
2242 }
2243
2244 static int build_chunk_tree(struct mdrestore_struct *mdres,
2245                             struct meta_cluster *cluster)
2246 {
2247         struct btrfs_super_block *super;
2248         struct meta_cluster_header *header;
2249         struct meta_cluster_item *item = NULL;
2250         u64 chunk_root_bytenr = 0;
2251         u32 i, nritems;
2252         u64 bytenr = 0;
2253         u8 *buffer;
2254         int ret;
2255
2256         /* We can't seek with stdin so don't bother doing this */
2257         if (mdres->in == stdin)
2258                 return 0;
2259
2260         ret = fread(cluster, BLOCK_SIZE, 1, mdres->in);
2261         if (ret <= 0) {
2262                 fprintf(stderr, "Error reading in cluster: %d\n", errno);
2263                 return -EIO;
2264         }
2265         ret = 0;
2266
2267         header = &cluster->header;
2268         if (le64_to_cpu(header->magic) != HEADER_MAGIC ||
2269             le64_to_cpu(header->bytenr) != 0) {
2270                 fprintf(stderr, "bad header in metadump image\n");
2271                 return -EIO;
2272         }
2273
2274         bytenr += BLOCK_SIZE;
2275         mdres->compress_method = header->compress;
2276         nritems = le32_to_cpu(header->nritems);
2277         for (i = 0; i < nritems; i++) {
2278                 item = &cluster->items[i];
2279
2280                 if (le64_to_cpu(item->bytenr) == BTRFS_SUPER_INFO_OFFSET)
2281                         break;
2282                 bytenr += le32_to_cpu(item->size);
2283                 if (fseek(mdres->in, le32_to_cpu(item->size), SEEK_CUR)) {
2284                         fprintf(stderr, "Error seeking: %d\n", errno);
2285                         return -EIO;
2286                 }
2287         }
2288
2289         if (!item || le64_to_cpu(item->bytenr) != BTRFS_SUPER_INFO_OFFSET) {
2290                 fprintf(stderr, "Huh, didn't find the super?\n");
2291                 return -EINVAL;
2292         }
2293
2294         buffer = malloc(le32_to_cpu(item->size));
2295         if (!buffer) {
2296                 fprintf(stderr, "Error allocing buffer\n");
2297                 return -ENOMEM;
2298         }
2299
2300         ret = fread(buffer, le32_to_cpu(item->size), 1, mdres->in);
2301         if (ret != 1) {
2302                 fprintf(stderr, "Error reading buffer: %d\n", errno);
2303                 free(buffer);
2304                 return -EIO;
2305         }
2306
2307         if (mdres->compress_method == COMPRESS_ZLIB) {
2308                 size_t size = MAX_PENDING_SIZE * 2;
2309                 u8 *tmp;
2310
2311                 tmp = malloc(MAX_PENDING_SIZE * 2);
2312                 if (!tmp) {
2313                         free(buffer);
2314                         return -ENOMEM;
2315                 }
2316                 ret = uncompress(tmp, (unsigned long *)&size,
2317                                  buffer, le32_to_cpu(item->size));
2318                 if (ret != Z_OK) {
2319                         fprintf(stderr, "Error decompressing %d\n", ret);
2320                         free(buffer);
2321                         free(tmp);
2322                         return -EIO;
2323                 }
2324                 free(buffer);
2325                 buffer = tmp;
2326         }
2327
2328         pthread_mutex_lock(&mdres->mutex);
2329         super = (struct btrfs_super_block *)buffer;
2330         chunk_root_bytenr = btrfs_super_chunk_root(super);
2331         mdres->leafsize = btrfs_super_leafsize(super);
2332         memcpy(mdres->fsid, super->fsid, BTRFS_FSID_SIZE);
2333         memcpy(mdres->uuid, super->dev_item.uuid,
2334                        BTRFS_UUID_SIZE);
2335         mdres->devid = le64_to_cpu(super->dev_item.devid);
2336         free(buffer);
2337         pthread_mutex_unlock(&mdres->mutex);
2338
2339         return search_for_chunk_blocks(mdres, chunk_root_bytenr, 0);
2340 }
2341
2342 static int range_contains_super(u64 physical, u64 bytes)
2343 {
2344         u64 super_bytenr;
2345         int i;
2346
2347         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
2348                 super_bytenr = btrfs_sb_offset(i);
2349                 if (super_bytenr >= physical &&
2350                     super_bytenr < physical + bytes)
2351                         return 1;
2352         }
2353
2354         return 0;
2355 }
2356
2357 static void remap_overlapping_chunks(struct mdrestore_struct *mdres)
2358 {
2359         struct fs_chunk *fs_chunk;
2360
2361         while (!list_empty(&mdres->overlapping_chunks)) {
2362                 fs_chunk = list_first_entry(&mdres->overlapping_chunks,
2363                                             struct fs_chunk, list);
2364                 list_del_init(&fs_chunk->list);
2365                 if (range_contains_super(fs_chunk->physical,
2366                                          fs_chunk->bytes)) {
2367                         fprintf(stderr, "Remapping a chunk that had a super "
2368                                 "mirror inside of it, clearing space cache "
2369                                 "so we don't end up with corruption\n");
2370                         mdres->clear_space_cache = 1;
2371                 }
2372                 fs_chunk->physical = mdres->last_physical_offset;
2373                 tree_insert(&mdres->physical_tree, &fs_chunk->p, physical_cmp);
2374                 mdres->last_physical_offset += fs_chunk->bytes;
2375         }
2376 }
2377
2378 static int fixup_devices(struct btrfs_fs_info *fs_info,
2379                          struct mdrestore_struct *mdres, off_t dev_size)
2380 {
2381         struct btrfs_trans_handle *trans;
2382         struct btrfs_dev_item *dev_item;
2383         struct btrfs_path *path;
2384         struct extent_buffer *leaf;
2385         struct btrfs_root *root = fs_info->chunk_root;
2386         struct btrfs_key key;
2387         u64 devid, cur_devid;
2388         int ret;
2389
2390         path = btrfs_alloc_path();
2391         if (!path) {
2392                 fprintf(stderr, "Error alloc'ing path\n");
2393                 return -ENOMEM;
2394         }
2395
2396         trans = btrfs_start_transaction(fs_info->tree_root, 1);
2397         if (IS_ERR(trans)) {
2398                 fprintf(stderr, "Error starting transaction %ld\n",
2399                         PTR_ERR(trans));
2400                 btrfs_free_path(path);
2401                 return PTR_ERR(trans);
2402         }
2403
2404         dev_item = &fs_info->super_copy->dev_item;
2405
2406         devid = btrfs_stack_device_id(dev_item);
2407
2408         btrfs_set_stack_device_total_bytes(dev_item, dev_size);
2409         btrfs_set_stack_device_bytes_used(dev_item, mdres->alloced_chunks);
2410
2411         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2412         key.type = BTRFS_DEV_ITEM_KEY;
2413         key.offset = 0;
2414
2415 again:
2416         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2417         if (ret < 0) {
2418                 fprintf(stderr, "search failed %d\n", ret);
2419                 exit(1);
2420         }
2421
2422         while (1) {
2423                 leaf = path->nodes[0];
2424                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2425                         ret = btrfs_next_leaf(root, path);
2426                         if (ret < 0) {
2427                                 fprintf(stderr, "Error going to next leaf "
2428                                         "%d\n", ret);
2429                                 exit(1);
2430                         }
2431                         if (ret > 0) {
2432                                 ret = 0;
2433                                 break;
2434                         }
2435                         leaf = path->nodes[0];
2436                 }
2437
2438                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2439                 if (key.type > BTRFS_DEV_ITEM_KEY)
2440                         break;
2441                 if (key.type != BTRFS_DEV_ITEM_KEY) {
2442                         path->slots[0]++;
2443                         continue;
2444                 }
2445
2446                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2447                                           struct btrfs_dev_item);
2448                 cur_devid = btrfs_device_id(leaf, dev_item);
2449                 if (devid != cur_devid) {
2450                         ret = btrfs_del_item(trans, root, path);
2451                         if (ret) {
2452                                 fprintf(stderr, "Error deleting item %d\n",
2453                                         ret);
2454                                 exit(1);
2455                         }
2456                         btrfs_release_path(path);
2457                         goto again;
2458                 }
2459
2460                 btrfs_set_device_total_bytes(leaf, dev_item, dev_size);
2461                 btrfs_set_device_bytes_used(leaf, dev_item,
2462                                             mdres->alloced_chunks);
2463                 btrfs_mark_buffer_dirty(leaf);
2464                 path->slots[0]++;
2465         }
2466
2467         btrfs_free_path(path);
2468         ret = btrfs_commit_transaction(trans, fs_info->tree_root);
2469         if (ret) {
2470                 fprintf(stderr, "Commit failed %d\n", ret);
2471                 return ret;
2472         }
2473         return 0;
2474 }
2475
2476 static int restore_metadump(const char *input, FILE *out, int old_restore,
2477                             int num_threads, int fixup_offset,
2478                             const char *target, int multi_devices)
2479 {
2480         struct meta_cluster *cluster = NULL;
2481         struct meta_cluster_header *header;
2482         struct mdrestore_struct mdrestore;
2483         struct btrfs_fs_info *info = NULL;
2484         u64 bytenr = 0;
2485         FILE *in = NULL;
2486         int ret = 0;
2487
2488         if (!strcmp(input, "-")) {
2489                 in = stdin;
2490         } else {
2491                 in = fopen(input, "r");
2492                 if (!in) {
2493                         perror("unable to open metadump image");
2494                         return 1;
2495                 }
2496         }
2497
2498         /* NOTE: open with write mode */
2499         if (fixup_offset) {
2500                 BUG_ON(!target);
2501                 info = open_ctree_fs_info(target, 0, 0,
2502                                           OPEN_CTREE_WRITES |
2503                                           OPEN_CTREE_RESTORE |
2504                                           OPEN_CTREE_PARTIAL);
2505                 if (!info) {
2506                         fprintf(stderr, "%s: open ctree failed\n", __func__);
2507                         ret = -EIO;
2508                         goto failed_open;
2509                 }
2510         }
2511
2512         cluster = malloc(BLOCK_SIZE);
2513         if (!cluster) {
2514                 fprintf(stderr, "Error allocating cluster\n");
2515                 ret = -ENOMEM;
2516                 goto failed_info;
2517         }
2518
2519         ret = mdrestore_init(&mdrestore, in, out, old_restore, num_threads,
2520                              fixup_offset, info, multi_devices);
2521         if (ret) {
2522                 fprintf(stderr, "Error initing mdrestore %d\n", ret);
2523                 goto failed_cluster;
2524         }
2525
2526         if (!multi_devices && !old_restore) {
2527                 ret = build_chunk_tree(&mdrestore, cluster);
2528                 if (ret)
2529                         goto out;
2530                 if (!list_empty(&mdrestore.overlapping_chunks))
2531                         remap_overlapping_chunks(&mdrestore);
2532         }
2533
2534         if (in != stdin && fseek(in, 0, SEEK_SET)) {
2535                 fprintf(stderr, "Error seeking %d\n", errno);
2536                 goto out;
2537         }
2538
2539         while (!mdrestore.error) {
2540                 ret = fread(cluster, BLOCK_SIZE, 1, in);
2541                 if (!ret)
2542                         break;
2543
2544                 header = &cluster->header;
2545                 if (le64_to_cpu(header->magic) != HEADER_MAGIC ||
2546                     le64_to_cpu(header->bytenr) != bytenr) {
2547                         fprintf(stderr, "bad header in metadump image\n");
2548                         ret = -EIO;
2549                         break;
2550                 }
2551                 ret = add_cluster(cluster, &mdrestore, &bytenr);
2552                 if (ret) {
2553                         fprintf(stderr, "Error adding cluster\n");
2554                         break;
2555                 }
2556         }
2557         ret = wait_for_worker(&mdrestore);
2558
2559         if (!ret && !multi_devices && !old_restore) {
2560                 struct btrfs_root *root;
2561                 struct stat st;
2562
2563                 root = open_ctree_fd(fileno(out), target, 0,
2564                                           OPEN_CTREE_PARTIAL |
2565                                           OPEN_CTREE_WRITES |
2566                                           OPEN_CTREE_NO_DEVICES);
2567                 if (!root) {
2568                         fprintf(stderr, "unable to open %s\n", target);
2569                         ret = -EIO;
2570                         goto out;
2571                 }
2572                 info = root->fs_info;
2573
2574                 if (stat(target, &st)) {
2575                         fprintf(stderr, "statting %s failed\n", target);
2576                         close_ctree(info->chunk_root);
2577                         return 1;
2578                 }
2579
2580                 ret = fixup_devices(info, &mdrestore, st.st_size);
2581                 close_ctree(info->chunk_root);
2582                 if (ret)
2583                         goto out;
2584         }
2585 out:
2586         mdrestore_destroy(&mdrestore, num_threads);
2587 failed_cluster:
2588         free(cluster);
2589 failed_info:
2590         if (fixup_offset && info)
2591                 close_ctree(info->chunk_root);
2592 failed_open:
2593         if (in != stdin)
2594                 fclose(in);
2595         return ret;
2596 }
2597
2598 static int update_disk_super_on_device(struct btrfs_fs_info *info,
2599                                        const char *other_dev, u64 cur_devid)
2600 {
2601         struct btrfs_key key;
2602         struct extent_buffer *leaf;
2603         struct btrfs_path path;
2604         struct btrfs_dev_item *dev_item;
2605         struct btrfs_super_block *disk_super;
2606         char dev_uuid[BTRFS_UUID_SIZE];
2607         char fs_uuid[BTRFS_UUID_SIZE];
2608         u64 devid, type, io_align, io_width;
2609         u64 sector_size, total_bytes, bytes_used;
2610         char *buf;
2611         int fp;
2612         int ret;
2613
2614         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2615         key.type = BTRFS_DEV_ITEM_KEY;
2616         key.offset = cur_devid;
2617
2618         btrfs_init_path(&path);
2619         ret = btrfs_search_slot(NULL, info->chunk_root, &key, &path, 0, 0);
2620         if (ret) {
2621                 fprintf(stderr, "search key fails\n");
2622                 exit(1);
2623         }
2624
2625         leaf = path.nodes[0];
2626         dev_item = btrfs_item_ptr(leaf, path.slots[0],
2627                                   struct btrfs_dev_item);
2628
2629         devid = btrfs_device_id(leaf, dev_item);
2630         if (devid != cur_devid) {
2631                 printk("devid %llu mismatch with %llu\n", devid, cur_devid);
2632                 exit(1);
2633         }
2634
2635         type = btrfs_device_type(leaf, dev_item);
2636         io_align = btrfs_device_io_align(leaf, dev_item);
2637         io_width = btrfs_device_io_width(leaf, dev_item);
2638         sector_size = btrfs_device_sector_size(leaf, dev_item);
2639         total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2640         bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2641         read_extent_buffer(leaf, dev_uuid, (unsigned long)btrfs_device_uuid(dev_item), BTRFS_UUID_SIZE);
2642         read_extent_buffer(leaf, fs_uuid, (unsigned long)btrfs_device_fsid(dev_item), BTRFS_UUID_SIZE);
2643
2644         btrfs_release_path(&path);
2645
2646         printk("update disk super on %s devid=%llu\n", other_dev, devid);
2647
2648         /* update other devices' super block */
2649         fp = open(other_dev, O_CREAT | O_RDWR, 0600);
2650         if (fp < 0) {
2651                 fprintf(stderr, "could not open %s\n", other_dev);
2652                 exit(1);
2653         }
2654
2655         buf = malloc(BTRFS_SUPER_INFO_SIZE);
2656         if (!buf) {
2657                 ret = -ENOMEM;
2658                 close(fp);
2659                 return ret;
2660         }
2661
2662         memcpy(buf, info->super_copy, BTRFS_SUPER_INFO_SIZE);
2663
2664         disk_super = (struct btrfs_super_block *)buf;
2665         dev_item = &disk_super->dev_item;
2666
2667         btrfs_set_stack_device_type(dev_item, type);
2668         btrfs_set_stack_device_id(dev_item, devid);
2669         btrfs_set_stack_device_total_bytes(dev_item, total_bytes);
2670         btrfs_set_stack_device_bytes_used(dev_item, bytes_used);
2671         btrfs_set_stack_device_io_align(dev_item, io_align);
2672         btrfs_set_stack_device_io_width(dev_item, io_width);
2673         btrfs_set_stack_device_sector_size(dev_item, sector_size);
2674         memcpy(dev_item->uuid, dev_uuid, BTRFS_UUID_SIZE);
2675         memcpy(dev_item->fsid, fs_uuid, BTRFS_UUID_SIZE);
2676         csum_block((u8 *)buf, BTRFS_SUPER_INFO_SIZE);
2677
2678         ret = pwrite64(fp, buf, BTRFS_SUPER_INFO_SIZE, BTRFS_SUPER_INFO_OFFSET);
2679         if (ret != BTRFS_SUPER_INFO_SIZE) {
2680                 ret = -EIO;
2681                 goto out;
2682         }
2683
2684         write_backup_supers(fp, (u8 *)buf);
2685
2686 out:
2687         free(buf);
2688         close(fp);
2689         return 0;
2690 }
2691
2692 static void print_usage(void)
2693 {
2694         fprintf(stderr, "usage: btrfs-image [options] source target\n");
2695         fprintf(stderr, "\t-r      \trestore metadump image\n");
2696         fprintf(stderr, "\t-c value\tcompression level (0 ~ 9)\n");
2697         fprintf(stderr, "\t-t value\tnumber of threads (1 ~ 32)\n");
2698         fprintf(stderr, "\t-o      \tdon't mess with the chunk tree when restoring\n");
2699         fprintf(stderr, "\t-s      \tsanitize file names, use once to just use garbage, use twice if you want crc collisions\n");
2700         fprintf(stderr, "\t-w      \twalk all trees instead of using extent tree, do this if your extent tree is broken\n");
2701         fprintf(stderr, "\t-m      \trestore for multiple devices\n");
2702         fprintf(stderr, "\n");
2703         fprintf(stderr, "\tIn the dump mode, source is the btrfs device and target is the output file (use '-' for stdout).\n");
2704         fprintf(stderr, "\tIn the restore mode, source is the dumped image and target is the btrfs device/file.\n");
2705         exit(1);
2706 }
2707
2708 int main(int argc, char *argv[])
2709 {
2710         char *source;
2711         char *target;
2712         u64 num_threads = 1;
2713         u64 compress_level = 0;
2714         int create = 1;
2715         int old_restore = 0;
2716         int walk_trees = 0;
2717         int multi_devices = 0;
2718         int ret;
2719         int sanitize = 0;
2720         int dev_cnt = 0;
2721         int usage_error = 0;
2722         FILE *out;
2723
2724         while (1) {
2725                 int c = getopt(argc, argv, "rc:t:oswm");
2726                 if (c < 0)
2727                         break;
2728                 switch (c) {
2729                 case 'r':
2730                         create = 0;
2731                         break;
2732                 case 't':
2733                         num_threads = arg_strtou64(optarg);
2734                         if (num_threads > 32)
2735                                 print_usage();
2736                         break;
2737                 case 'c':
2738                         compress_level = arg_strtou64(optarg);
2739                         if (compress_level > 9)
2740                                 print_usage();
2741                         break;
2742                 case 'o':
2743                         old_restore = 1;
2744                         break;
2745                 case 's':
2746                         sanitize++;
2747                         break;
2748                 case 'w':
2749                         walk_trees = 1;
2750                         break;
2751                 case 'm':
2752                         create = 0;
2753                         multi_devices = 1;
2754                         break;
2755                 default:
2756                         print_usage();
2757                 }
2758         }
2759
2760         argc = argc - optind;
2761         set_argv0(argv);
2762         if (check_argc_min(argc, 2))
2763                 print_usage();
2764
2765         dev_cnt = argc - 1;
2766
2767         if (create) {
2768                 if (old_restore) {
2769                         fprintf(stderr, "Usage error: create and restore cannot be used at the same time\n");
2770                         usage_error++;
2771                 }
2772         } else {
2773                 if (walk_trees || sanitize || compress_level) {
2774                         fprintf(stderr, "Usage error: use -w, -s, -c options for restore makes no sense\n");
2775                         usage_error++;
2776                 }
2777                 if (multi_devices && dev_cnt < 2) {
2778                         fprintf(stderr, "Usage error: not enough devices specified for -m option\n");
2779                         usage_error++;
2780                 }
2781                 if (!multi_devices && dev_cnt != 1) {
2782                         fprintf(stderr, "Usage error: accepts only 1 device without -m option\n");
2783                         usage_error++;
2784                 }
2785         }
2786
2787         if (usage_error)
2788                 print_usage();
2789
2790         source = argv[optind];
2791         target = argv[optind + 1];
2792
2793         if (create && !strcmp(target, "-")) {
2794                 out = stdout;
2795         } else {
2796                 out = fopen(target, "w+");
2797                 if (!out) {
2798                         perror("unable to create target file");
2799                         exit(1);
2800                 }
2801         }
2802
2803         if (num_threads == 1 && compress_level > 0) {
2804                 num_threads = sysconf(_SC_NPROCESSORS_ONLN);
2805                 if (num_threads <= 0)
2806                         num_threads = 1;
2807         }
2808
2809         if (create) {
2810                 ret = check_mounted(source);
2811                 if (ret < 0) {
2812                         fprintf(stderr, "Could not check mount status: %s\n",
2813                                 strerror(-ret));
2814                         exit(1);
2815                 } else if (ret)
2816                         fprintf(stderr,
2817                 "WARNING: The device is mounted. Make sure the filesystem is quiescent.\n");
2818
2819                 ret = create_metadump(source, out, num_threads,
2820                                       compress_level, sanitize, walk_trees);
2821         } else {
2822                 ret = restore_metadump(source, out, old_restore, num_threads,
2823                                        0, target, multi_devices);
2824         }
2825         if (ret) {
2826                 printk("%s failed (%s)\n", (create) ? "create" : "restore",
2827                        strerror(errno));
2828                 goto out;
2829         }
2830
2831          /* extended support for multiple devices */
2832         if (!create && multi_devices) {
2833                 struct btrfs_fs_info *info;
2834                 u64 total_devs;
2835                 int i;
2836
2837                 info = open_ctree_fs_info(target, 0, 0,
2838                                           OPEN_CTREE_PARTIAL |
2839                                           OPEN_CTREE_RESTORE);
2840                 if (!info) {
2841                         int e = errno;
2842                         fprintf(stderr, "unable to open %s error = %s\n",
2843                                 target, strerror(e));
2844                         return 1;
2845                 }
2846
2847                 total_devs = btrfs_super_num_devices(info->super_copy);
2848                 if (total_devs != dev_cnt) {
2849                         printk("it needs %llu devices but has only %d\n",
2850                                 total_devs, dev_cnt);
2851                         close_ctree(info->chunk_root);
2852                         goto out;
2853                 }
2854
2855                 /* update super block on other disks */
2856                 for (i = 2; i <= dev_cnt; i++) {
2857                         ret = update_disk_super_on_device(info,
2858                                         argv[optind + i], (u64)i);
2859                         if (ret) {
2860                                 printk("update disk super failed devid=%d (error=%d)\n",
2861                                         i, ret);
2862                                 close_ctree(info->chunk_root);
2863                                 exit(1);
2864                         }
2865                 }
2866
2867                 close_ctree(info->chunk_root);
2868
2869                 /* fix metadata block to map correct chunk */
2870                 ret = restore_metadump(source, out, 0, num_threads, 1,
2871                                        target, 1);
2872                 if (ret) {
2873                         fprintf(stderr, "fix metadump failed (error=%d)\n",
2874                                 ret);
2875                         exit(1);
2876                 }
2877         }
2878 out:
2879         if (out == stdout) {
2880                 fflush(out);
2881         } else {
2882                 fclose(out);
2883                 if (ret && create) {
2884                         int unlink_ret;
2885
2886                         unlink_ret = unlink(target);
2887                         if (unlink_ret)
2888                                 fprintf(stderr,
2889                                         "unlink output file failed : %s\n",
2890                                         strerror(errno));
2891                 }
2892         }
2893
2894         return !!ret;
2895 }