btrfs-image.c

   1 /*
   2  * Copyright (C) 2008 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <pthread.h>
  20 #include <stdio.h>
  21 #include <stdlib.h>
  22 #include <sys/types.h>
  23 #include <sys/stat.h>
  24 #include <fcntl.h>
  25 #include <unistd.h>
  26 #include <dirent.h>
  27 #include <zlib.h>
  28 #include <getopt.h>
  29
  30 #include "kerncompat.h"
  31 #include "crc32c.h"
  32 #include "ctree.h"
  33 #include "disk-io.h"
  34 #include "transaction.h"
  35 #include "utils.h"
  36 #include "volumes.h"
  37 #include "extent_io.h"
  38
  39 #define HEADER_MAGIC            0xbd5c25e27295668bULL
  40 #define MAX_PENDING_SIZE        (256 * 1024)
  41 #define BLOCK_SIZE              1024
  42 #define BLOCK_MASK              (BLOCK_SIZE - 1)
  43
  44 #define COMPRESS_NONE           0
  45 #define COMPRESS_ZLIB           1
  46
  47 struct meta_cluster_item {
  48         __le64 bytenr;
  49         __le32 size;
  50 } __attribute__ ((__packed__));
  51
  52 struct meta_cluster_header {
  53         __le64 magic;
  54         __le64 bytenr;
  55         __le32 nritems;
  56         u8 compress;
  57 } __attribute__ ((__packed__));
  58
  59 /* cluster header + index items + buffers */
  60 struct meta_cluster {
  61         struct meta_cluster_header header;
  62         struct meta_cluster_item items[];
  63 } __attribute__ ((__packed__));
  64
  65 #define ITEMS_PER_CLUSTER ((BLOCK_SIZE - sizeof(struct meta_cluster)) / \
  66                            sizeof(struct meta_cluster_item))
  67
  68 struct fs_chunk {
  69         u64 logical;
  70         u64 physical;
  71         /*
  72          * physical_dup only store additonal physical for BTRFS_BLOCK_GROUP_DUP
  73          * currently restore only support single and DUP
  74          * TODO: modify this structure and the function related to this
  75          * structure for support RAID*
  76          */
  77         u64 physical_dup;
  78         u64 bytes;
  79         struct rb_node l;
  80         struct rb_node p;
  81         struct list_head list;
  82 };
  83
  84 struct async_work {
  85         struct list_head list;
  86         struct list_head ordered;
  87         u64 start;
  88         u64 size;
  89         u8 *buffer;
  90         size_t bufsize;
  91         int error;
  92 };
  93
  94 struct metadump_struct {
  95         struct btrfs_root *root;
  96         FILE *out;
  97
  98         struct meta_cluster *cluster;
  99
 100         pthread_t *threads;
 101         size_t num_threads;
 102         pthread_mutex_t mutex;
 103         pthread_cond_t cond;
 104         struct rb_root name_tree;
 105
 106         struct list_head list;
 107         struct list_head ordered;
 108         size_t num_items;
 109         size_t num_ready;
 110
 111         u64 pending_start;
 112         u64 pending_size;
 113
 114         int compress_level;
 115         int done;
 116         int data;
 117         int sanitize_names;
 118
 119         int error;
 120 };
 121
 122 struct name {
 123         struct rb_node n;
 124         char *val;
 125         char *sub;
 126         u32 len;
 127 };
 128
 129 struct mdrestore_struct {
 130         FILE *in;
 131         FILE *out;
 132
 133         pthread_t *threads;
 134         size_t num_threads;
 135         pthread_mutex_t mutex;
 136         pthread_cond_t cond;
 137
 138         struct rb_root chunk_tree;
 139         struct rb_root physical_tree;
 140         struct list_head list;
 141         struct list_head overlapping_chunks;
 142         size_t num_items;
 143         u32 nodesize;
 144         u64 devid;
 145         u64 alloced_chunks;
 146         u64 last_physical_offset;
 147         u8 uuid[BTRFS_UUID_SIZE];
 148         u8 fsid[BTRFS_FSID_SIZE];
 149
 150         int compress_method;
 151         int done;
 152         int error;
 153         int old_restore;
 154         int fixup_offset;
 155         int multi_devices;
 156         int clear_space_cache;
 157         struct btrfs_fs_info *info;
 158 };
 159
 160 static int search_for_chunk_blocks(struct mdrestore_struct *mdres,
 161                                    u64 search, u64 cluster_bytenr);
 162 static struct extent_buffer *alloc_dummy_eb(u64 bytenr, u32 size);
 163
 164 static void csum_block(u8 *buf, size_t len)
 165 {
 166         u8 result[BTRFS_CRC32_SIZE];
 167         u32 crc = ~(u32)0;
 168         crc = crc32c(crc, buf + BTRFS_CSUM_SIZE, len - BTRFS_CSUM_SIZE);
 169         btrfs_csum_final(crc, result);
 170         memcpy(buf, result, BTRFS_CRC32_SIZE);
 171 }
 172
 173 static int has_name(struct btrfs_key *key)
 174 {
 175         switch (key->type) {
 176         case BTRFS_DIR_ITEM_KEY:
 177         case BTRFS_DIR_INDEX_KEY:
 178         case BTRFS_INODE_REF_KEY:
 179         case BTRFS_INODE_EXTREF_KEY:
 180         case BTRFS_XATTR_ITEM_KEY:
 181                 return 1;
 182         default:
 183                 break;
 184         }
 185
 186         return 0;
 187 }
 188
 189 static char *generate_garbage(u32 name_len)
 190 {
 191         char *buf = malloc(name_len);
 192         int i;
 193
 194         if (!buf)
 195                 return NULL;
 196
 197         for (i = 0; i < name_len; i++) {
 198                 char c = rand_range(94) + 33;
 199
 200                 if (c == '/')
 201                         c++;
 202                 buf[i] = c;
 203         }
 204
 205         return buf;
 206 }
 207
 208 static int name_cmp(struct rb_node *a, struct rb_node *b, int fuzz)
 209 {
 210         struct name *entry = rb_entry(a, struct name, n);
 211         struct name *ins = rb_entry(b, struct name, n);
 212         u32 len;
 213
 214         len = min(ins->len, entry->len);
 215         return memcmp(ins->val, entry->val, len);
 216 }
 217
 218 static int chunk_cmp(struct rb_node *a, struct rb_node *b, int fuzz)
 219 {
 220         struct fs_chunk *entry = rb_entry(a, struct fs_chunk, l);
 221         struct fs_chunk *ins = rb_entry(b, struct fs_chunk, l);
 222
 223         if (fuzz && ins->logical >= entry->logical &&
 224             ins->logical < entry->logical + entry->bytes)
 225                 return 0;
 226
 227         if (ins->logical < entry->logical)
 228                 return -1;
 229         else if (ins->logical > entry->logical)
 230                 return 1;
 231         return 0;
 232 }
 233
 234 static int physical_cmp(struct rb_node *a, struct rb_node *b, int fuzz)
 235 {
 236         struct fs_chunk *entry = rb_entry(a, struct fs_chunk, p);
 237         struct fs_chunk *ins = rb_entry(b, struct fs_chunk, p);
 238
 239         if (fuzz && ins->physical >= entry->physical &&
 240             ins->physical < entry->physical + entry->bytes)
 241                 return 0;
 242
 243         if (fuzz && entry->physical >= ins->physical &&
 244             entry->physical < ins->physical + ins->bytes)
 245                 return 0;
 246
 247         if (ins->physical < entry->physical)
 248                 return -1;
 249         else if (ins->physical > entry->physical)
 250                 return 1;
 251         return 0;
 252 }
 253
 254 static void tree_insert(struct rb_root *root, struct rb_node *ins,
 255                         int (*cmp)(struct rb_node *a, struct rb_node *b,
 256                                    int fuzz))
 257 {
 258         struct rb_node ** p = &root->rb_node;
 259         struct rb_node * parent = NULL;
 260         int dir;
 261
 262         while(*p) {
 263                 parent = *p;
 264
 265                 dir = cmp(*p, ins, 1);
 266                 if (dir < 0)
 267                         p = &(*p)->rb_left;
 268                 else if (dir > 0)
 269                         p = &(*p)->rb_right;
 270                 else
 271                         BUG();
 272         }
 273
 274         rb_link_node(ins, parent, p);
 275         rb_insert_color(ins, root);
 276 }
 277
 278 static struct rb_node *tree_search(struct rb_root *root,
 279                                    struct rb_node *search,
 280                                    int (*cmp)(struct rb_node *a,
 281                                               struct rb_node *b, int fuzz),
 282                                    int fuzz)
 283 {
 284         struct rb_node *n = root->rb_node;
 285         int dir;
 286
 287         while (n) {
 288                 dir = cmp(n, search, fuzz);
 289                 if (dir < 0)
 290                         n = n->rb_left;
 291                 else if (dir > 0)
 292                         n = n->rb_right;
 293                 else
 294                         return n;
 295         }
 296
 297         return NULL;
 298 }
 299
 300 static u64 logical_to_physical(struct mdrestore_struct *mdres, u64 logical,
 301                                u64 *size, u64 *physical_dup)
 302 {
 303         struct fs_chunk *fs_chunk;
 304         struct rb_node *entry;
 305         struct fs_chunk search;
 306         u64 offset;
 307
 308         if (logical == BTRFS_SUPER_INFO_OFFSET)
 309                 return logical;
 310
 311         search.logical = logical;
 312         entry = tree_search(&mdres->chunk_tree, &search.l, chunk_cmp, 1);
 313         if (!entry) {
 314                 if (mdres->in != stdin)
 315                         warning("cannot find a chunk, using logical");
 316                 return logical;
 317         }
 318         fs_chunk = rb_entry(entry, struct fs_chunk, l);
 319         if (fs_chunk->logical > logical || fs_chunk->logical + fs_chunk->bytes < logical)
 320                 BUG();
 321         offset = search.logical - fs_chunk->logical;
 322
 323         if (physical_dup) {
 324                 /* Only in dup case, physical_dup is not equal to 0 */
 325                 if (fs_chunk->physical_dup)
 326                         *physical_dup = fs_chunk->physical_dup + offset;
 327                 else
 328                         *physical_dup = 0;
 329         }
 330
 331         *size = min(*size, fs_chunk->bytes + fs_chunk->logical - logical);
 332         return fs_chunk->physical + offset;
 333 }
 334
 335
 336 static char *find_collision(struct metadump_struct *md, char *name,
 337                             u32 name_len)
 338 {
 339         struct name *val;
 340         struct rb_node *entry;
 341         struct name tmp;
 342         unsigned long checksum;
 343         int found = 0;
 344         int i;
 345
 346         tmp.val = name;
 347         tmp.len = name_len;
 348         entry = tree_search(&md->name_tree, &tmp.n, name_cmp, 0);
 349         if (entry) {
 350                 val = rb_entry(entry, struct name, n);
 351                 free(name);
 352                 return val->sub;
 353         }
 354
 355         val = malloc(sizeof(struct name));
 356         if (!val) {
 357                 error("cannot sanitize name, not enough memory");
 358                 free(name);
 359                 return NULL;
 360         }
 361
 362         memset(val, 0, sizeof(*val));
 363
 364         val->val = name;
 365         val->len = name_len;
 366         val->sub = malloc(name_len);
 367         if (!val->sub) {
 368                 error("cannot sanitize name, not enough memory");
 369                 free(val);
 370                 free(name);
 371                 return NULL;
 372         }
 373
 374         checksum = crc32c(~1, val->val, name_len);
 375         memset(val->sub, ' ', name_len);
 376         i = 0;
 377         while (1) {
 378                 if (crc32c(~1, val->sub, name_len) == checksum &&
 379                     memcmp(val->sub, val->val, val->len)) {
 380                         found = 1;
 381                         break;
 382                 }
 383
 384                 if (val->sub[i] == 127) {
 385                         do {
 386                                 i++;
 387                                 if (i >= name_len)
 388                                         break;
 389                         } while (val->sub[i] == 127);
 390
 391                         if (i >= name_len)
 392                                 break;
 393                         val->sub[i]++;
 394                         if (val->sub[i] == '/')
 395                                 val->sub[i]++;
 396                         memset(val->sub, ' ', i);
 397                         i = 0;
 398                         continue;
 399                 } else {
 400                         val->sub[i]++;
 401                         if (val->sub[i] == '/')
 402                                 val->sub[i]++;
 403                 }
 404         }
 405
 406         if (!found) {
 407                 warning(
 408 "cannot find a hash collision for '%.*s', generating garbage, it won't match indexes",
 409                         val->len, val->val);
 410                 for (i = 0; i < name_len; i++) {
 411                         char c = rand_range(94) + 33;
 412
 413                         if (c == '/')
 414                                 c++;
 415                         val->sub[i] = c;
 416                 }
 417         }
 418
 419         tree_insert(&md->name_tree, &val->n, name_cmp);
 420         return val->sub;
 421 }
 422
 423 static void sanitize_dir_item(struct metadump_struct *md, struct extent_buffer *eb,
 424                               int slot)
 425 {
 426         struct btrfs_dir_item *dir_item;
 427         char *buf;
 428         char *garbage;
 429         unsigned long name_ptr;
 430         u32 total_len;
 431         u32 cur = 0;
 432         u32 this_len;
 433         u32 name_len;
 434         int free_garbage = (md->sanitize_names == 1);
 435
 436         dir_item = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
 437         total_len = btrfs_item_size_nr(eb, slot);
 438         while (cur < total_len) {
 439                 this_len = sizeof(*dir_item) +
 440                         btrfs_dir_name_len(eb, dir_item) +
 441                         btrfs_dir_data_len(eb, dir_item);
 442                 name_ptr = (unsigned long)(dir_item + 1);
 443                 name_len = btrfs_dir_name_len(eb, dir_item);
 444
 445                 if (md->sanitize_names > 1) {
 446                         buf = malloc(name_len);
 447                         if (!buf) {
 448                                 error("cannot sanitize name, not enough memory");
 449                                 return;
 450                         }
 451                         read_extent_buffer(eb, buf, name_ptr, name_len);
 452                         garbage = find_collision(md, buf, name_len);
 453                 } else {
 454                         garbage = generate_garbage(name_len);
 455                 }
 456                 if (!garbage) {
 457                         error("cannot sanitize name, not enough memory");
 458                         return;
 459                 }
 460                 write_extent_buffer(eb, garbage, name_ptr, name_len);
 461                 cur += this_len;
 462                 dir_item = (struct btrfs_dir_item *)((char *)dir_item +
 463                                                      this_len);
 464                 if (free_garbage)
 465                         free(garbage);
 466         }
 467 }
 468
 469 static void sanitize_inode_ref(struct metadump_struct *md,
 470                                struct extent_buffer *eb, int slot, int ext)
 471 {
 472         struct btrfs_inode_extref *extref;
 473         struct btrfs_inode_ref *ref;
 474         char *garbage, *buf;
 475         unsigned long ptr;
 476         unsigned long name_ptr;
 477         u32 item_size;
 478         u32 cur_offset = 0;
 479         int len;
 480         int free_garbage = (md->sanitize_names == 1);
 481
 482         item_size = btrfs_item_size_nr(eb, slot);
 483         ptr = btrfs_item_ptr_offset(eb, slot);
 484         while (cur_offset < item_size) {
 485                 if (ext) {
 486                         extref = (struct btrfs_inode_extref *)(ptr +
 487                                                                cur_offset);
 488                         name_ptr = (unsigned long)(&extref->name);
 489                         len = btrfs_inode_extref_name_len(eb, extref);
 490                         cur_offset += sizeof(*extref);
 491                 } else {
 492                         ref = (struct btrfs_inode_ref *)(ptr + cur_offset);
 493                         len = btrfs_inode_ref_name_len(eb, ref);
 494                         name_ptr = (unsigned long)(ref + 1);
 495                         cur_offset += sizeof(*ref);
 496                 }
 497                 cur_offset += len;
 498
 499                 if (md->sanitize_names > 1) {
 500                         buf = malloc(len);
 501                         if (!buf) {
 502                                 error("cannot sanitize name, not enough memory");
 503                                 return;
 504                         }
 505                         read_extent_buffer(eb, buf, name_ptr, len);
 506                         garbage = find_collision(md, buf, len);
 507                 } else {
 508                         garbage = generate_garbage(len);
 509                 }
 510
 511                 if (!garbage) {
 512                         error("cannot sanitize name, not enough memory");
 513                         return;
 514                 }
 515                 write_extent_buffer(eb, garbage, name_ptr, len);
 516                 if (free_garbage)
 517                         free(garbage);
 518         }
 519 }
 520
 521 static void sanitize_xattr(struct metadump_struct *md,
 522                            struct extent_buffer *eb, int slot)
 523 {
 524         struct btrfs_dir_item *dir_item;
 525         unsigned long data_ptr;
 526         u32 data_len;
 527
 528         dir_item = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
 529         data_len = btrfs_dir_data_len(eb, dir_item);
 530
 531         data_ptr = (unsigned long)((char *)(dir_item + 1) +
 532                                    btrfs_dir_name_len(eb, dir_item));
 533         memset_extent_buffer(eb, 0, data_ptr, data_len);
 534 }
 535
 536 static void sanitize_name(struct metadump_struct *md, u8 *dst,
 537                           struct extent_buffer *src, struct btrfs_key *key,
 538                           int slot)
 539 {
 540         struct extent_buffer *eb;
 541
 542         eb = alloc_dummy_eb(src->start, src->len);
 543         if (!eb) {
 544                 error("cannot sanitize name, not enough memory");
 545                 return;
 546         }
 547
 548         memcpy(eb->data, dst, eb->len);
 549
 550         switch (key->type) {
 551         case BTRFS_DIR_ITEM_KEY:
 552         case BTRFS_DIR_INDEX_KEY:
 553                 sanitize_dir_item(md, eb, slot);
 554                 break;
 555         case BTRFS_INODE_REF_KEY:
 556                 sanitize_inode_ref(md, eb, slot, 0);
 557                 break;
 558         case BTRFS_INODE_EXTREF_KEY:
 559                 sanitize_inode_ref(md, eb, slot, 1);
 560                 break;
 561         case BTRFS_XATTR_ITEM_KEY:
 562                 sanitize_xattr(md, eb, slot);
 563                 break;
 564         default:
 565                 break;
 566         }
 567
 568         memcpy(dst, eb->data, eb->len);
 569         free(eb);
 570 }
 571
 572 /*
 573  * zero inline extents and csum items
 574  */
 575 static void zero_items(struct metadump_struct *md, u8 *dst,
 576                        struct extent_buffer *src)
 577 {
 578         struct btrfs_file_extent_item *fi;
 579         struct btrfs_item *item;
 580         struct btrfs_key key;
 581         u32 nritems = btrfs_header_nritems(src);
 582         size_t size;
 583         unsigned long ptr;
 584         int i, extent_type;
 585
 586         for (i = 0; i < nritems; i++) {
 587                 item = btrfs_item_nr(i);
 588                 btrfs_item_key_to_cpu(src, &key, i);
 589                 if (key.type == BTRFS_CSUM_ITEM_KEY) {
 590                         size = btrfs_item_size_nr(src, i);
 591                         memset(dst + btrfs_leaf_data(src) +
 592                                btrfs_item_offset_nr(src, i), 0, size);
 593                         continue;
 594                 }
 595
 596                 if (md->sanitize_names && has_name(&key)) {
 597                         sanitize_name(md, dst, src, &key, i);
 598                         continue;
 599                 }
 600
 601                 if (key.type != BTRFS_EXTENT_DATA_KEY)
 602                         continue;
 603
 604                 fi = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
 605                 extent_type = btrfs_file_extent_type(src, fi);
 606                 if (extent_type != BTRFS_FILE_EXTENT_INLINE)
 607                         continue;
 608
 609                 ptr = btrfs_file_extent_inline_start(fi);
 610                 size = btrfs_file_extent_inline_item_len(src, item);
 611                 memset(dst + ptr, 0, size);
 612         }
 613 }
 614
 615 /*
 616  * copy buffer and zero useless data in the buffer
 617  */
 618 static void copy_buffer(struct metadump_struct *md, u8 *dst,
 619                         struct extent_buffer *src)
 620 {
 621         int level;
 622         size_t size;
 623         u32 nritems;
 624
 625         memcpy(dst, src->data, src->len);
 626         if (src->start == BTRFS_SUPER_INFO_OFFSET)
 627                 return;
 628
 629         level = btrfs_header_level(src);
 630         nritems = btrfs_header_nritems(src);
 631
 632         if (nritems == 0) {
 633                 size = sizeof(struct btrfs_header);
 634                 memset(dst + size, 0, src->len - size);
 635         } else if (level == 0) {
 636                 size = btrfs_leaf_data(src) +
 637                         btrfs_item_offset_nr(src, nritems - 1) -
 638                         btrfs_item_nr_offset(nritems);
 639                 memset(dst + btrfs_item_nr_offset(nritems), 0, size);
 640                 zero_items(md, dst, src);
 641         } else {
 642                 size = offsetof(struct btrfs_node, ptrs) +
 643                         sizeof(struct btrfs_key_ptr) * nritems;
 644                 memset(dst + size, 0, src->len - size);
 645         }
 646         csum_block(dst, src->len);
 647 }
 648
 649 static void *dump_worker(void *data)
 650 {
 651         struct metadump_struct *md = (struct metadump_struct *)data;
 652         struct async_work *async;
 653         int ret;
 654
 655         while (1) {
 656                 pthread_mutex_lock(&md->mutex);
 657                 while (list_empty(&md->list)) {
 658                         if (md->done) {
 659                                 pthread_mutex_unlock(&md->mutex);
 660                                 goto out;
 661                         }
 662                         pthread_cond_wait(&md->cond, &md->mutex);
 663                 }
 664                 async = list_entry(md->list.next, struct async_work, list);
 665                 list_del_init(&async->list);
 666                 pthread_mutex_unlock(&md->mutex);
 667
 668                 if (md->compress_level > 0) {
 669                         u8 *orig = async->buffer;
 670
 671                         async->bufsize = compressBound(async->size);
 672                         async->buffer = malloc(async->bufsize);
 673                         if (!async->buffer) {
 674                                 error("not enough memory for async buffer");
 675                                 pthread_mutex_lock(&md->mutex);
 676                                 if (!md->error)
 677                                         md->error = -ENOMEM;
 678                                 pthread_mutex_unlock(&md->mutex);
 679                                 pthread_exit(NULL);
 680                         }
 681
 682                         ret = compress2(async->buffer,
 683                                          (unsigned long *)&async->bufsize,
 684                                          orig, async->size, md->compress_level);
 685
 686                         if (ret != Z_OK)
 687                                 async->error = 1;
 688
 689                         free(orig);
 690                 }
 691
 692                 pthread_mutex_lock(&md->mutex);
 693                 md->num_ready++;
 694                 pthread_mutex_unlock(&md->mutex);
 695         }
 696 out:
 697         pthread_exit(NULL);
 698 }
 699
 700 static void meta_cluster_init(struct metadump_struct *md, u64 start)
 701 {
 702         struct meta_cluster_header *header;
 703
 704         md->num_items = 0;
 705         md->num_ready = 0;
 706         header = &md->cluster->header;
 707         header->magic = cpu_to_le64(HEADER_MAGIC);
 708         header->bytenr = cpu_to_le64(start);
 709         header->nritems = cpu_to_le32(0);
 710         header->compress = md->compress_level > 0 ?
 711                            COMPRESS_ZLIB : COMPRESS_NONE;
 712 }
 713
 714 static void metadump_destroy(struct metadump_struct *md, int num_threads)
 715 {
 716         int i;
 717         struct rb_node *n;
 718
 719         pthread_mutex_lock(&md->mutex);
 720         md->done = 1;
 721         pthread_cond_broadcast(&md->cond);
 722         pthread_mutex_unlock(&md->mutex);
 723
 724         for (i = 0; i < num_threads; i++)
 725                 pthread_join(md->threads[i], NULL);
 726
 727         pthread_cond_destroy(&md->cond);
 728         pthread_mutex_destroy(&md->mutex);
 729
 730         while ((n = rb_first(&md->name_tree))) {
 731                 struct name *name;
 732
 733                 name = rb_entry(n, struct name, n);
 734                 rb_erase(n, &md->name_tree);
 735                 free(name->val);
 736                 free(name->sub);
 737                 free(name);
 738         }
 739         free(md->threads);
 740         free(md->cluster);
 741 }
 742
 743 static int metadump_init(struct metadump_struct *md, struct btrfs_root *root,
 744                          FILE *out, int num_threads, int compress_level,
 745                          int sanitize_names)
 746 {
 747         int i, ret = 0;
 748
 749         memset(md, 0, sizeof(*md));
 750         md->cluster = calloc(1, BLOCK_SIZE);
 751         if (!md->cluster)
 752                 return -ENOMEM;
 753         md->threads = calloc(num_threads, sizeof(pthread_t));
 754         if (!md->threads) {
 755                 free(md->cluster);
 756                 return -ENOMEM;
 757         }
 758         INIT_LIST_HEAD(&md->list);
 759         INIT_LIST_HEAD(&md->ordered);
 760         md->root = root;
 761         md->out = out;
 762         md->pending_start = (u64)-1;
 763         md->compress_level = compress_level;
 764         md->sanitize_names = sanitize_names;
 765         if (sanitize_names > 1)
 766                 crc32c_optimization_init();
 767
 768         md->name_tree.rb_node = NULL;
 769         md->num_threads = num_threads;
 770         pthread_cond_init(&md->cond, NULL);
 771         pthread_mutex_init(&md->mutex, NULL);
 772         meta_cluster_init(md, 0);
 773
 774         if (!num_threads)
 775                 return 0;
 776
 777         for (i = 0; i < num_threads; i++) {
 778                 ret = pthread_create(md->threads + i, NULL, dump_worker, md);
 779                 if (ret)
 780                         break;
 781         }
 782
 783         if (ret)
 784                 metadump_destroy(md, i + 1);
 785
 786         return ret;
 787 }
 788
 789 static int write_zero(FILE *out, size_t size)
 790 {
 791         static char zero[BLOCK_SIZE];
 792         return fwrite(zero, size, 1, out);
 793 }
 794
 795 static int write_buffers(struct metadump_struct *md, u64 *next)
 796 {
 797         struct meta_cluster_header *header = &md->cluster->header;
 798         struct meta_cluster_item *item;
 799         struct async_work *async;
 800         u64 bytenr = 0;
 801         u32 nritems = 0;
 802         int ret;
 803         int err = 0;
 804
 805         if (list_empty(&md->ordered))
 806                 goto out;
 807
 808         /* wait until all buffers are compressed */
 809         while (!err && md->num_items > md->num_ready) {
 810                 struct timespec ts = {
 811                         .tv_sec = 0,
 812                         .tv_nsec = 10000000,
 813                 };
 814                 pthread_mutex_unlock(&md->mutex);
 815                 nanosleep(&ts, NULL);
 816                 pthread_mutex_lock(&md->mutex);
 817                 err = md->error;
 818         }
 819
 820         if (err) {
 821                 error("one of the threads failed: %s", strerror(-err));
 822                 goto out;
 823         }
 824
 825         /* setup and write index block */
 826         list_for_each_entry(async, &md->ordered, ordered) {
 827                 item = md->cluster->items + nritems;
 828                 item->bytenr = cpu_to_le64(async->start);
 829                 item->size = cpu_to_le32(async->bufsize);
 830                 nritems++;
 831         }
 832         header->nritems = cpu_to_le32(nritems);
 833
 834         ret = fwrite(md->cluster, BLOCK_SIZE, 1, md->out);
 835         if (ret != 1) {
 836                 error("unable to write out cluster: %s", strerror(errno));
 837                 return -errno;
 838         }
 839
 840         /* write buffers */
 841         bytenr += le64_to_cpu(header->bytenr) + BLOCK_SIZE;
 842         while (!list_empty(&md->ordered)) {
 843                 async = list_entry(md->ordered.next, struct async_work,
 844                                    ordered);
 845                 list_del_init(&async->ordered);
 846
 847                 bytenr += async->bufsize;
 848                 if (!err)
 849                         ret = fwrite(async->buffer, async->bufsize, 1,
 850                                      md->out);
 851                 if (ret != 1) {
 852                         error("unable to write out cluster: %s",
 853                                 strerror(errno));
 854                         err = -errno;
 855                         ret = 0;
 856                 }
 857
 858                 free(async->buffer);
 859                 free(async);
 860         }
 861
 862         /* zero unused space in the last block */
 863         if (!err && bytenr & BLOCK_MASK) {
 864                 size_t size = BLOCK_SIZE - (bytenr & BLOCK_MASK);
 865
 866                 bytenr += size;
 867                 ret = write_zero(md->out, size);
 868                 if (ret != 1) {
 869                         error("unable to zero out buffer: %s",
 870                                 strerror(errno));
 871                         err = -errno;
 872                 }
 873         }
 874 out:
 875         *next = bytenr;
 876         return err;
 877 }
 878
 879 static int read_data_extent(struct metadump_struct *md,
 880                             struct async_work *async)
 881 {
 882         struct btrfs_root *root = md->root;
 883         u64 bytes_left = async->size;
 884         u64 logical = async->start;
 885         u64 offset = 0;
 886         u64 read_len;
 887         int num_copies;
 888         int cur_mirror;
 889         int ret;
 890
 891         num_copies = btrfs_num_copies(&root->fs_info->mapping_tree, logical,
 892                                       bytes_left);
 893
 894         /* Try our best to read data, just like read_tree_block() */
 895         for (cur_mirror = 0; cur_mirror < num_copies; cur_mirror++) {
 896                 while (bytes_left) {
 897                         read_len = bytes_left;
 898                         ret = read_extent_data(root,
 899                                         (char *)(async->buffer + offset),
 900                                         logical, &read_len, cur_mirror);
 901                         if (ret < 0)
 902                                 break;
 903                         offset += read_len;
 904                         logical += read_len;
 905                         bytes_left -= read_len;
 906                 }
 907         }
 908         if (bytes_left)
 909                 return -EIO;
 910         return 0;
 911 }
 912
 913 static int get_dev_fd(struct btrfs_root *root)
 914 {
 915         struct btrfs_device *dev;
 916
 917         dev = list_first_entry(&root->fs_info->fs_devices->devices,
 918                                struct btrfs_device, dev_list);
 919         return dev->fd;
 920 }
 921
 922 static int flush_pending(struct metadump_struct *md, int done)
 923 {
 924         struct async_work *async = NULL;
 925         struct extent_buffer *eb;
 926         u64 blocksize = md->root->nodesize;
 927         u64 start = 0;
 928         u64 size;
 929         size_t offset;
 930         int ret = 0;
 931
 932         if (md->pending_size) {
 933                 async = calloc(1, sizeof(*async));
 934                 if (!async)
 935                         return -ENOMEM;
 936
 937                 async->start = md->pending_start;
 938                 async->size = md->pending_size;
 939                 async->bufsize = async->size;
 940                 async->buffer = malloc(async->bufsize);
 941                 if (!async->buffer) {
 942                         free(async);
 943                         return -ENOMEM;
 944                 }
 945                 offset = 0;
 946                 start = async->start;
 947                 size = async->size;
 948
 949                 if (md->data) {
 950                         ret = read_data_extent(md, async);
 951                         if (ret) {
 952                                 free(async->buffer);
 953                                 free(async);
 954                                 return ret;
 955                         }
 956                 }
 957
 958                 /*
 959                  * Balance can make the mapping not cover the super block, so
 960                  * just copy directly from one of the devices.
 961                  */
 962                 if (start == BTRFS_SUPER_INFO_OFFSET) {
 963                         int fd = get_dev_fd(md->root);
 964
 965                         ret = pread64(fd, async->buffer, size, start);
 966                         if (ret < size) {
 967                                 free(async->buffer);
 968                                 free(async);
 969                                 error("unable to read superblock at %llu: %s",
 970                                                 (unsigned long long)start,
 971                                                 strerror(errno));
 972                                 return -errno;
 973                         }
 974                         size = 0;
 975                         ret = 0;
 976                 }
 977
 978                 while (!md->data && size > 0) {
 979                         u64 this_read = min(blocksize, size);
 980                         eb = read_tree_block(md->root, start, this_read, 0);
 981                         if (!extent_buffer_uptodate(eb)) {
 982                                 free(async->buffer);
 983                                 free(async);
 984                                 error("unable to read metadata block %llu",
 985                                         (unsigned long long)start);
 986                                 return -EIO;
 987                         }
 988                         copy_buffer(md, async->buffer + offset, eb);
 989                         free_extent_buffer(eb);
 990                         start += this_read;
 991                         offset += this_read;
 992                         size -= this_read;
 993                 }
 994
 995                 md->pending_start = (u64)-1;
 996                 md->pending_size = 0;
 997         } else if (!done) {
 998                 return 0;
 999         }
1000
1001         pthread_mutex_lock(&md->mutex);
1002         if (async) {
1003                 list_add_tail(&async->ordered, &md->ordered);
1004                 md->num_items++;
1005                 if (md->compress_level > 0) {
1006                         list_add_tail(&async->list, &md->list);
1007                         pthread_cond_signal(&md->cond);
1008                 } else {
1009                         md->num_ready++;
1010                 }
1011         }
1012         if (md->num_items >= ITEMS_PER_CLUSTER || done) {
1013                 ret = write_buffers(md, &start);
1014                 if (ret)
1015                         error("unable to write buffers: %s", strerror(-ret));
1016                 else
1017                         meta_cluster_init(md, start);
1018         }
1019         pthread_mutex_unlock(&md->mutex);
1020         return ret;
1021 }
1022
1023 static int add_extent(u64 start, u64 size, struct metadump_struct *md,
1024                       int data)
1025 {
1026         int ret;
1027         if (md->data != data ||
1028             md->pending_size + size > MAX_PENDING_SIZE ||
1029             md->pending_start + md->pending_size != start) {
1030                 ret = flush_pending(md, 0);
1031                 if (ret)
1032                         return ret;
1033                 md->pending_start = start;
1034         }
1035         readahead_tree_block(md->root, start, size, 0);
1036         md->pending_size += size;
1037         md->data = data;
1038         return 0;
1039 }
1040
1041 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1042 static int is_tree_block(struct btrfs_root *extent_root,
1043                          struct btrfs_path *path, u64 bytenr)
1044 {
1045         struct extent_buffer *leaf;
1046         struct btrfs_key key;
1047         u64 ref_objectid;
1048         int ret;
1049
1050         leaf = path->nodes[0];
1051         while (1) {
1052                 struct btrfs_extent_ref_v0 *ref_item;
1053                 path->slots[0]++;
1054                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1055                         ret = btrfs_next_leaf(extent_root, path);
1056                         if (ret < 0)
1057                                 return ret;
1058                         if (ret > 0)
1059                                 break;
1060                         leaf = path->nodes[0];
1061                 }
1062                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1063                 if (key.objectid != bytenr)
1064                         break;
1065                 if (key.type != BTRFS_EXTENT_REF_V0_KEY)
1066                         continue;
1067                 ref_item = btrfs_item_ptr(leaf, path->slots[0],
1068                                           struct btrfs_extent_ref_v0);
1069                 ref_objectid = btrfs_ref_objectid_v0(leaf, ref_item);
1070                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID)
1071                         return 1;
1072                 break;
1073         }
1074         return 0;
1075 }
1076 #endif
1077
1078 static int copy_tree_blocks(struct btrfs_root *root, struct extent_buffer *eb,
1079                             struct metadump_struct *metadump, int root_tree)
1080 {
1081         struct extent_buffer *tmp;
1082         struct btrfs_root_item *ri;
1083         struct btrfs_key key;
1084         u64 bytenr;
1085         int level;
1086         int nritems = 0;
1087         int i = 0;
1088         int ret;
1089
1090         ret = add_extent(btrfs_header_bytenr(eb), root->nodesize, metadump, 0);
1091         if (ret) {
1092                 error("unable to add metadata block %llu: %d",
1093                                 btrfs_header_bytenr(eb), ret);
1094                 return ret;
1095         }
1096
1097         if (btrfs_header_level(eb) == 0 && !root_tree)
1098                 return 0;
1099
1100         level = btrfs_header_level(eb);
1101         nritems = btrfs_header_nritems(eb);
1102         for (i = 0; i < nritems; i++) {
1103                 if (level == 0) {
1104                         btrfs_item_key_to_cpu(eb, &key, i);
1105                         if (key.type != BTRFS_ROOT_ITEM_KEY)
1106                                 continue;
1107                         ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
1108                         bytenr = btrfs_disk_root_bytenr(eb, ri);
1109                         tmp = read_tree_block(root, bytenr, root->nodesize, 0);
1110                         if (!extent_buffer_uptodate(tmp)) {
1111                                 error("unable to read log root block");
1112                                 return -EIO;
1113                         }
1114                         ret = copy_tree_blocks(root, tmp, metadump, 0);
1115                         free_extent_buffer(tmp);
1116                         if (ret)
1117                                 return ret;
1118                 } else {
1119                         bytenr = btrfs_node_blockptr(eb, i);
1120                         tmp = read_tree_block(root, bytenr, root->nodesize, 0);
1121                         if (!extent_buffer_uptodate(tmp)) {
1122                                 error("unable to read log root block");
1123                                 return -EIO;
1124                         }
1125                         ret = copy_tree_blocks(root, tmp, metadump, root_tree);
1126                         free_extent_buffer(tmp);
1127                         if (ret)
1128                                 return ret;
1129                 }
1130         }
1131
1132         return 0;
1133 }
1134
1135 static int copy_log_trees(struct btrfs_root *root,
1136                           struct metadump_struct *metadump,
1137                           struct btrfs_path *path)
1138 {
1139         u64 blocknr = btrfs_super_log_root(root->fs_info->super_copy);
1140
1141         if (blocknr == 0)
1142                 return 0;
1143
1144         if (!root->fs_info->log_root_tree ||
1145             !root->fs_info->log_root_tree->node) {
1146                 error("unable to copy tree log, it has not been setup");
1147                 return -EIO;
1148         }
1149
1150         return copy_tree_blocks(root, root->fs_info->log_root_tree->node,
1151                                 metadump, 1);
1152 }
1153
1154 static int copy_space_cache(struct btrfs_root *root,
1155                             struct metadump_struct *metadump,
1156                             struct btrfs_path *path)
1157 {
1158         struct extent_buffer *leaf;
1159         struct btrfs_file_extent_item *fi;
1160         struct btrfs_key key;
1161         u64 bytenr, num_bytes;
1162         int ret;
1163
1164         root = root->fs_info->tree_root;
1165
1166         key.objectid = 0;
1167         key.type = BTRFS_EXTENT_DATA_KEY;
1168         key.offset = 0;
1169
1170         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1171         if (ret < 0) {
1172                 error("free space inode not found: %d", ret);
1173                 return ret;
1174         }
1175
1176         leaf = path->nodes[0];
1177
1178         while (1) {
1179                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1180                         ret = btrfs_next_leaf(root, path);
1181                         if (ret < 0) {
1182                                 error("cannot go to next leaf %d", ret);
1183                                 return ret;
1184                         }
1185                         if (ret > 0)
1186                                 break;
1187                         leaf = path->nodes[0];
1188                 }
1189
1190                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1191                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
1192                         path->slots[0]++;
1193                         continue;
1194                 }
1195
1196                 fi = btrfs_item_ptr(leaf, path->slots[0],
1197                                     struct btrfs_file_extent_item);
1198                 if (btrfs_file_extent_type(leaf, fi) !=
1199                     BTRFS_FILE_EXTENT_REG) {
1200                         path->slots[0]++;
1201                         continue;
1202                 }
1203
1204                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1205                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1206                 ret = add_extent(bytenr, num_bytes, metadump, 1);
1207                 if (ret) {
1208                         error("unable to add space cache blocks %d", ret);
1209                         btrfs_release_path(path);
1210                         return ret;
1211                 }
1212                 path->slots[0]++;
1213         }
1214
1215         return 0;
1216 }
1217
1218 static int copy_from_extent_tree(struct metadump_struct *metadump,
1219                                  struct btrfs_path *path)
1220 {
1221         struct btrfs_root *extent_root;
1222         struct extent_buffer *leaf;
1223         struct btrfs_extent_item *ei;
1224         struct btrfs_key key;
1225         u64 bytenr;
1226         u64 num_bytes;
1227         int ret;
1228
1229         extent_root = metadump->root->fs_info->extent_root;
1230         bytenr = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
1231         key.objectid = bytenr;
1232         key.type = BTRFS_EXTENT_ITEM_KEY;
1233         key.offset = 0;
1234
1235         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1236         if (ret < 0) {
1237                 error("extent root not found: %d", ret);
1238                 return ret;
1239         }
1240         ret = 0;
1241
1242         leaf = path->nodes[0];
1243
1244         while (1) {
1245                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1246                         ret = btrfs_next_leaf(extent_root, path);
1247                         if (ret < 0) {
1248                                 error("cannot go to next leaf %d", ret);
1249                                 break;
1250                         }
1251                         if (ret > 0) {
1252                                 ret = 0;
1253                                 break;
1254                         }
1255                         leaf = path->nodes[0];
1256                 }
1257
1258                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1259                 if (key.objectid < bytenr ||
1260                     (key.type != BTRFS_EXTENT_ITEM_KEY &&
1261                      key.type != BTRFS_METADATA_ITEM_KEY)) {
1262                         path->slots[0]++;
1263                         continue;
1264                 }
1265
1266                 bytenr = key.objectid;
1267                 if (key.type == BTRFS_METADATA_ITEM_KEY) {
1268                         num_bytes = extent_root->nodesize;
1269                 } else {
1270                         num_bytes = key.offset;
1271                 }
1272
1273                 if (num_bytes == 0) {
1274                         error("extent length 0 at bytenr %llu key type %d",
1275                                         (unsigned long long)bytenr, key.type);
1276                         ret = -EIO;
1277                         break;
1278                 }
1279
1280                 if (btrfs_item_size_nr(leaf, path->slots[0]) > sizeof(*ei)) {
1281                         ei = btrfs_item_ptr(leaf, path->slots[0],
1282                                             struct btrfs_extent_item);
1283                         if (btrfs_extent_flags(leaf, ei) &
1284                             BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1285                                 ret = add_extent(bytenr, num_bytes, metadump,
1286                                                  0);
1287                                 if (ret) {
1288                                         error("unable to add block %llu: %d",
1289                                                 (unsigned long long)bytenr, ret);
1290                                         break;
1291                                 }
1292                         }
1293                 } else {
1294 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1295                         ret = is_tree_block(extent_root, path, bytenr);
1296                         if (ret < 0) {
1297                                 error("failed to check tree block %llu: %d",
1298                                         (unsigned long long)bytenr, ret);
1299                                 break;
1300                         }
1301
1302                         if (ret) {
1303                                 ret = add_extent(bytenr, num_bytes, metadump,
1304                                                  0);
1305                                 if (ret) {
1306                                         error("unable to add block %llu: %d",
1307                                                 (unsigned long long)bytenr, ret);
1308                                         break;
1309                                 }
1310                         }
1311                         ret = 0;
1312 #else
1313                         error(
1314         "either extent tree is corrupted or you haven't built with V0 support");
1315                         ret = -EIO;
1316                         break;
1317 #endif
1318                 }
1319                 bytenr += num_bytes;
1320         }
1321
1322         btrfs_release_path(path);
1323
1324         return ret;
1325 }
1326
1327 static int create_metadump(const char *input, FILE *out, int num_threads,
1328                            int compress_level, int sanitize, int walk_trees)
1329 {
1330         struct btrfs_root *root;
1331         struct btrfs_path path;
1332         struct metadump_struct metadump;
1333         int ret;
1334         int err = 0;
1335
1336         root = open_ctree(input, 0, 0);
1337         if (!root) {
1338                 error("open ctree failed");
1339                 return -EIO;
1340         }
1341
1342         ret = metadump_init(&metadump, root, out, num_threads,
1343                             compress_level, sanitize);
1344         if (ret) {
1345                 error("failed to initialize metadump: %d", ret);
1346                 close_ctree(root);
1347                 return ret;
1348         }
1349
1350         ret = add_extent(BTRFS_SUPER_INFO_OFFSET, BTRFS_SUPER_INFO_SIZE,
1351                         &metadump, 0);
1352         if (ret) {
1353                 error("unable to add metadata: %d", ret);
1354                 err = ret;
1355                 goto out;
1356         }
1357
1358         btrfs_init_path(&path);
1359
1360         if (walk_trees) {
1361                 ret = copy_tree_blocks(root, root->fs_info->chunk_root->node,
1362                                        &metadump, 1);
1363                 if (ret) {
1364                         err = ret;
1365                         goto out;
1366                 }
1367
1368                 ret = copy_tree_blocks(root, root->fs_info->tree_root->node,
1369                                        &metadump, 1);
1370                 if (ret) {
1371                         err = ret;
1372                         goto out;
1373                 }
1374         } else {
1375                 ret = copy_from_extent_tree(&metadump, &path);
1376                 if (ret) {
1377                         err = ret;
1378                         goto out;
1379                 }
1380         }
1381
1382         ret = copy_log_trees(root, &metadump, &path);
1383         if (ret) {
1384                 err = ret;
1385                 goto out;
1386         }
1387
1388         ret = copy_space_cache(root, &metadump, &path);
1389 out:
1390         ret = flush_pending(&metadump, 1);
1391         if (ret) {
1392                 if (!err)
1393                         err = ret;
1394                 error("failed to flush pending data: %d", ret);
1395         }
1396
1397         metadump_destroy(&metadump, num_threads);
1398
1399         btrfs_release_path(&path);
1400         ret = close_ctree(root);
1401         return err ? err : ret;
1402 }
1403
1404 static void update_super_old(u8 *buffer)
1405 {
1406         struct btrfs_super_block *super = (struct btrfs_super_block *)buffer;
1407         struct btrfs_chunk *chunk;
1408         struct btrfs_disk_key *key;
1409         u32 sectorsize = btrfs_super_sectorsize(super);
1410         u64 flags = btrfs_super_flags(super);
1411
1412         flags |= BTRFS_SUPER_FLAG_METADUMP;
1413         btrfs_set_super_flags(super, flags);
1414
1415         key = (struct btrfs_disk_key *)(super->sys_chunk_array);
1416         chunk = (struct btrfs_chunk *)(super->sys_chunk_array +
1417                                        sizeof(struct btrfs_disk_key));
1418
1419         btrfs_set_disk_key_objectid(key, BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1420         btrfs_set_disk_key_type(key, BTRFS_CHUNK_ITEM_KEY);
1421         btrfs_set_disk_key_offset(key, 0);
1422
1423         btrfs_set_stack_chunk_length(chunk, (u64)-1);
1424         btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
1425         btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
1426         btrfs_set_stack_chunk_type(chunk, BTRFS_BLOCK_GROUP_SYSTEM);
1427         btrfs_set_stack_chunk_io_align(chunk, sectorsize);
1428         btrfs_set_stack_chunk_io_width(chunk, sectorsize);
1429         btrfs_set_stack_chunk_sector_size(chunk, sectorsize);
1430         btrfs_set_stack_chunk_num_stripes(chunk, 1);
1431         btrfs_set_stack_chunk_sub_stripes(chunk, 0);
1432         chunk->stripe.devid = super->dev_item.devid;
1433         btrfs_set_stack_stripe_offset(&chunk->stripe, 0);
1434         memcpy(chunk->stripe.dev_uuid, super->dev_item.uuid, BTRFS_UUID_SIZE);
1435         btrfs_set_super_sys_array_size(super, sizeof(*key) + sizeof(*chunk));
1436         csum_block(buffer, BTRFS_SUPER_INFO_SIZE);
1437 }
1438
1439 static int update_super(struct mdrestore_struct *mdres, u8 *buffer)
1440 {
1441         struct btrfs_super_block *super = (struct btrfs_super_block *)buffer;
1442         struct btrfs_chunk *chunk;
1443         struct btrfs_disk_key *disk_key;
1444         struct btrfs_key key;
1445         u64 flags = btrfs_super_flags(super);
1446         u32 new_array_size = 0;
1447         u32 array_size;
1448         u32 cur = 0;
1449         u8 *ptr, *write_ptr;
1450         int old_num_stripes;
1451
1452         write_ptr = ptr = super->sys_chunk_array;
1453         array_size = btrfs_super_sys_array_size(super);
1454
1455         while (cur < array_size) {
1456                 disk_key = (struct btrfs_disk_key *)ptr;
1457                 btrfs_disk_key_to_cpu(&key, disk_key);
1458
1459                 new_array_size += sizeof(*disk_key);
1460                 memmove(write_ptr, ptr, sizeof(*disk_key));
1461
1462                 write_ptr += sizeof(*disk_key);
1463                 ptr += sizeof(*disk_key);
1464                 cur += sizeof(*disk_key);
1465
1466                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1467                         u64 type, physical, physical_dup, size = 0;
1468
1469                         chunk = (struct btrfs_chunk *)ptr;
1470                         old_num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1471                         chunk = (struct btrfs_chunk *)write_ptr;
1472
1473                         memmove(write_ptr, ptr, sizeof(*chunk));
1474                         btrfs_set_stack_chunk_sub_stripes(chunk, 0);
1475                         type = btrfs_stack_chunk_type(chunk);
1476                         if (type & BTRFS_BLOCK_GROUP_DUP) {
1477                                 new_array_size += sizeof(struct btrfs_stripe);
1478                                 write_ptr += sizeof(struct btrfs_stripe);
1479                         } else {
1480                                 btrfs_set_stack_chunk_num_stripes(chunk, 1);
1481                                 btrfs_set_stack_chunk_type(chunk,
1482                                                 BTRFS_BLOCK_GROUP_SYSTEM);
1483                         }
1484                         chunk->stripe.devid = super->dev_item.devid;
1485                         physical = logical_to_physical(mdres, key.offset,
1486                                                        &size, &physical_dup);
1487                         if (size != (u64)-1)
1488                                 btrfs_set_stack_stripe_offset(&chunk->stripe,
1489                                                               physical);
1490                         memcpy(chunk->stripe.dev_uuid, super->dev_item.uuid,
1491                                BTRFS_UUID_SIZE);
1492                         new_array_size += sizeof(*chunk);
1493                 } else {
1494                         error("bogus key in the sys array %d", key.type);
1495                         return -EIO;
1496                 }
1497                 write_ptr += sizeof(*chunk);
1498                 ptr += btrfs_chunk_item_size(old_num_stripes);
1499                 cur += btrfs_chunk_item_size(old_num_stripes);
1500         }
1501
1502         if (mdres->clear_space_cache)
1503                 btrfs_set_super_cache_generation(super, 0);
1504
1505         flags |= BTRFS_SUPER_FLAG_METADUMP_V2;
1506         btrfs_set_super_flags(super, flags);
1507         btrfs_set_super_sys_array_size(super, new_array_size);
1508         csum_block(buffer, BTRFS_SUPER_INFO_SIZE);
1509
1510         return 0;
1511 }
1512
1513 static struct extent_buffer *alloc_dummy_eb(u64 bytenr, u32 size)
1514 {
1515         struct extent_buffer *eb;
1516
1517         eb = calloc(1, sizeof(struct extent_buffer) + size);
1518         if (!eb)
1519                 return NULL;
1520
1521         eb->start = bytenr;
1522         eb->len = size;
1523         return eb;
1524 }
1525
1526 static void truncate_item(struct extent_buffer *eb, int slot, u32 new_size)
1527 {
1528         struct btrfs_item *item;
1529         u32 nritems;
1530         u32 old_size;
1531         u32 old_data_start;
1532         u32 size_diff;
1533         u32 data_end;
1534         int i;
1535
1536         old_size = btrfs_item_size_nr(eb, slot);
1537         if (old_size == new_size)
1538                 return;
1539
1540         nritems = btrfs_header_nritems(eb);
1541         data_end = btrfs_item_offset_nr(eb, nritems - 1);
1542
1543         old_data_start = btrfs_item_offset_nr(eb, slot);
1544         size_diff = old_size - new_size;
1545
1546         for (i = slot; i < nritems; i++) {
1547                 u32 ioff;
1548                 item = btrfs_item_nr(i);
1549                 ioff = btrfs_item_offset(eb, item);
1550                 btrfs_set_item_offset(eb, item, ioff + size_diff);
1551         }
1552
1553         memmove_extent_buffer(eb, btrfs_leaf_data(eb) + data_end + size_diff,
1554                               btrfs_leaf_data(eb) + data_end,
1555                               old_data_start + new_size - data_end);
1556         item = btrfs_item_nr(slot);
1557         btrfs_set_item_size(eb, item, new_size);
1558 }
1559
1560 static int fixup_chunk_tree_block(struct mdrestore_struct *mdres,
1561                                   struct async_work *async, u8 *buffer,
1562                                   size_t size)
1563 {
1564         struct extent_buffer *eb;
1565         size_t size_left = size;
1566         u64 bytenr = async->start;
1567         int i;
1568
1569         if (size_left % mdres->nodesize)
1570                 return 0;
1571
1572         eb = alloc_dummy_eb(bytenr, mdres->nodesize);
1573         if (!eb)
1574                 return -ENOMEM;
1575
1576         while (size_left) {
1577                 eb->start = bytenr;
1578                 memcpy(eb->data, buffer, mdres->nodesize);
1579
1580                 if (btrfs_header_bytenr(eb) != bytenr)
1581                         break;
1582                 if (memcmp(mdres->fsid,
1583                            eb->data + offsetof(struct btrfs_header, fsid),
1584                            BTRFS_FSID_SIZE))
1585                         break;
1586
1587                 if (btrfs_header_owner(eb) != BTRFS_CHUNK_TREE_OBJECTID)
1588                         goto next;
1589
1590                 if (btrfs_header_level(eb) != 0)
1591                         goto next;
1592
1593                 for (i = 0; i < btrfs_header_nritems(eb); i++) {
1594                         struct btrfs_chunk *chunk;
1595                         struct btrfs_key key;
1596                         u64 type, physical, physical_dup, size = (u64)-1;
1597
1598                         btrfs_item_key_to_cpu(eb, &key, i);
1599                         if (key.type != BTRFS_CHUNK_ITEM_KEY)
1600                                 continue;
1601
1602                         size = 0;
1603                         physical = logical_to_physical(mdres, key.offset,
1604                                                        &size, &physical_dup);
1605
1606                         if (!physical_dup)
1607                                 truncate_item(eb, i, sizeof(*chunk));
1608                         chunk = btrfs_item_ptr(eb, i, struct btrfs_chunk);
1609
1610
1611                         /* Zero out the RAID profile */
1612                         type = btrfs_chunk_type(eb, chunk);
1613                         type &= (BTRFS_BLOCK_GROUP_DATA |
1614                                  BTRFS_BLOCK_GROUP_SYSTEM |
1615                                  BTRFS_BLOCK_GROUP_METADATA |
1616                                  BTRFS_BLOCK_GROUP_DUP);
1617                         btrfs_set_chunk_type(eb, chunk, type);
1618
1619                         if (!physical_dup)
1620                                 btrfs_set_chunk_num_stripes(eb, chunk, 1);
1621                         btrfs_set_chunk_sub_stripes(eb, chunk, 0);
1622                         btrfs_set_stripe_devid_nr(eb, chunk, 0, mdres->devid);
1623                         if (size != (u64)-1)
1624                                 btrfs_set_stripe_offset_nr(eb, chunk, 0,
1625                                                            physical);
1626                         /* update stripe 2 offset */
1627                         if (physical_dup)
1628                                 btrfs_set_stripe_offset_nr(eb, chunk, 1,
1629                                                            physical_dup);
1630
1631                         write_extent_buffer(eb, mdres->uuid,
1632                                         (unsigned long)btrfs_stripe_dev_uuid_nr(
1633                                                 chunk, 0),
1634                                         BTRFS_UUID_SIZE);
1635                 }
1636                 memcpy(buffer, eb->data, eb->len);
1637                 csum_block(buffer, eb->len);
1638 next:
1639                 size_left -= mdres->nodesize;
1640                 buffer += mdres->nodesize;
1641                 bytenr += mdres->nodesize;
1642         }
1643
1644         free(eb);
1645         return 0;
1646 }
1647
1648 static void write_backup_supers(int fd, u8 *buf)
1649 {
1650         struct btrfs_super_block *super = (struct btrfs_super_block *)buf;
1651         struct stat st;
1652         u64 size;
1653         u64 bytenr;
1654         int i;
1655         int ret;
1656
1657         if (fstat(fd, &st)) {
1658                 error(
1659         "cannot stat restore point, won't be able to write backup supers: %s",
1660                         strerror(errno));
1661                 return;
1662         }
1663
1664         size = btrfs_device_size(fd, &st);
1665
1666         for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1667                 bytenr = btrfs_sb_offset(i);
1668                 if (bytenr + BTRFS_SUPER_INFO_SIZE > size)
1669                         break;
1670                 btrfs_set_super_bytenr(super, bytenr);
1671                 csum_block(buf, BTRFS_SUPER_INFO_SIZE);
1672                 ret = pwrite64(fd, buf, BTRFS_SUPER_INFO_SIZE, bytenr);
1673                 if (ret < BTRFS_SUPER_INFO_SIZE) {
1674                         if (ret < 0)
1675                                 error(
1676                                 "problem writing out backup super block %d: %s",
1677                                                 i, strerror(errno));
1678                         else
1679                                 error("short write writing out backup super block");
1680                         break;
1681                 }
1682         }
1683 }
1684
1685 static void *restore_worker(void *data)
1686 {
1687         struct mdrestore_struct *mdres = (struct mdrestore_struct *)data;
1688         struct async_work *async;
1689         size_t size;
1690         u8 *buffer;
1691         u8 *outbuf;
1692         int outfd;
1693         int ret;
1694         int compress_size = MAX_PENDING_SIZE * 4;
1695
1696         outfd = fileno(mdres->out);
1697         buffer = malloc(compress_size);
1698         if (!buffer) {
1699                 error("not enough memory for restore worker buffer");
1700                 pthread_mutex_lock(&mdres->mutex);
1701                 if (!mdres->error)
1702                         mdres->error = -ENOMEM;
1703                 pthread_mutex_unlock(&mdres->mutex);
1704                 pthread_exit(NULL);
1705         }
1706
1707         while (1) {
1708                 u64 bytenr, physical_dup;
1709                 off_t offset = 0;
1710                 int err = 0;
1711
1712                 pthread_mutex_lock(&mdres->mutex);
1713                 while (!mdres->nodesize || list_empty(&mdres->list)) {
1714                         if (mdres->done) {
1715                                 pthread_mutex_unlock(&mdres->mutex);
1716                                 goto out;
1717                         }
1718                         pthread_cond_wait(&mdres->cond, &mdres->mutex);
1719                 }
1720                 async = list_entry(mdres->list.next, struct async_work, list);
1721                 list_del_init(&async->list);
1722                 pthread_mutex_unlock(&mdres->mutex);
1723
1724                 if (mdres->compress_method == COMPRESS_ZLIB) {
1725                         size = compress_size;
1726                         ret = uncompress(buffer, (unsigned long *)&size,
1727                                          async->buffer, async->bufsize);
1728                         if (ret != Z_OK) {
1729                                 error("decompressiion failed with %d", ret);
1730                                 err = -EIO;
1731                         }
1732                         outbuf = buffer;
1733                 } else {
1734                         outbuf = async->buffer;
1735                         size = async->bufsize;
1736                 }
1737
1738                 if (!mdres->multi_devices) {
1739                         if (async->start == BTRFS_SUPER_INFO_OFFSET) {
1740                                 if (mdres->old_restore) {
1741                                         update_super_old(outbuf);
1742                                 } else {
1743                                         ret = update_super(mdres, outbuf);
1744                                         if (ret)
1745                                                 err = ret;
1746                                 }
1747                         } else if (!mdres->old_restore) {
1748                                 ret = fixup_chunk_tree_block(mdres, async, outbuf, size);
1749                                 if (ret)
1750                                         err = ret;
1751                         }
1752                 }
1753
1754                 if (!mdres->fixup_offset) {
1755                         while (size) {
1756                                 u64 chunk_size = size;
1757                                 physical_dup = 0;
1758                                 if (!mdres->multi_devices && !mdres->old_restore)
1759                                         bytenr = logical_to_physical(mdres,
1760                                                      async->start + offset,
1761                                                      &chunk_size,
1762                                                      &physical_dup);
1763                                 else
1764                                         bytenr = async->start + offset;
1765
1766                                 ret = pwrite64(outfd, outbuf+offset, chunk_size,
1767                                                bytenr);
1768                                 if (ret != chunk_size)
1769                                         goto error;
1770
1771                                 if (physical_dup)
1772                                         ret = pwrite64(outfd, outbuf+offset,
1773                                                        chunk_size,
1774                                                        physical_dup);
1775                                 if (ret != chunk_size)
1776                                         goto error;
1777
1778                                 size -= chunk_size;
1779                                 offset += chunk_size;
1780                                 continue;
1781
1782 error:
1783                                 if (ret < 0) {
1784                                         error("unable to write to device: %s",
1785                                                         strerror(errno));
1786                                         err = errno;
1787                                 } else {
1788                                         error("short write");
1789                                         err = -EIO;
1790                                 }
1791                         }
1792                 } else if (async->start != BTRFS_SUPER_INFO_OFFSET) {
1793                         ret = write_data_to_disk(mdres->info, outbuf, async->start, size, 0);
1794                         if (ret) {
1795                                 error("failed to write data");
1796                                 exit(1);
1797                         }
1798                 }
1799
1800
1801                 /* backup super blocks are already there at fixup_offset stage */
1802                 if (!mdres->multi_devices && async->start == BTRFS_SUPER_INFO_OFFSET)
1803                         write_backup_supers(outfd, outbuf);
1804
1805                 pthread_mutex_lock(&mdres->mutex);
1806                 if (err && !mdres->error)
1807                         mdres->error = err;
1808                 mdres->num_items--;
1809                 pthread_mutex_unlock(&mdres->mutex);
1810
1811                 free(async->buffer);
1812                 free(async);
1813         }
1814 out:
1815         free(buffer);
1816         pthread_exit(NULL);
1817 }
1818
1819 static void mdrestore_destroy(struct mdrestore_struct *mdres, int num_threads)
1820 {
1821         struct rb_node *n;
1822         int i;
1823
1824         while ((n = rb_first(&mdres->chunk_tree))) {
1825                 struct fs_chunk *entry;
1826
1827                 entry = rb_entry(n, struct fs_chunk, l);
1828                 rb_erase(n, &mdres->chunk_tree);
1829                 rb_erase(&entry->p, &mdres->physical_tree);
1830                 free(entry);
1831         }
1832         pthread_mutex_lock(&mdres->mutex);
1833         mdres->done = 1;
1834         pthread_cond_broadcast(&mdres->cond);
1835         pthread_mutex_unlock(&mdres->mutex);
1836
1837         for (i = 0; i < num_threads; i++)
1838                 pthread_join(mdres->threads[i], NULL);
1839
1840         pthread_cond_destroy(&mdres->cond);
1841         pthread_mutex_destroy(&mdres->mutex);
1842         free(mdres->threads);
1843 }
1844
1845 static int mdrestore_init(struct mdrestore_struct *mdres,
1846                           FILE *in, FILE *out, int old_restore,
1847                           int num_threads, int fixup_offset,
1848                           struct btrfs_fs_info *info, int multi_devices)
1849 {
1850         int i, ret = 0;
1851
1852         memset(mdres, 0, sizeof(*mdres));
1853         pthread_cond_init(&mdres->cond, NULL);
1854         pthread_mutex_init(&mdres->mutex, NULL);
1855         INIT_LIST_HEAD(&mdres->list);
1856         INIT_LIST_HEAD(&mdres->overlapping_chunks);
1857         mdres->in = in;
1858         mdres->out = out;
1859         mdres->old_restore = old_restore;
1860         mdres->chunk_tree.rb_node = NULL;
1861         mdres->fixup_offset = fixup_offset;
1862         mdres->info = info;
1863         mdres->multi_devices = multi_devices;
1864         mdres->clear_space_cache = 0;
1865         mdres->last_physical_offset = 0;
1866         mdres->alloced_chunks = 0;
1867
1868         if (!num_threads)
1869                 return 0;
1870
1871         mdres->num_threads = num_threads;
1872         mdres->threads = calloc(num_threads, sizeof(pthread_t));
1873         if (!mdres->threads)
1874                 return -ENOMEM;
1875         for (i = 0; i < num_threads; i++) {
1876                 ret = pthread_create(mdres->threads + i, NULL, restore_worker,
1877                                      mdres);
1878                 if (ret) {
1879                         /* pthread_create returns errno directly */
1880                         ret = -ret;
1881                         break;
1882                 }
1883         }
1884         if (ret)
1885                 mdrestore_destroy(mdres, i + 1);
1886         return ret;
1887 }
1888
1889 static int fill_mdres_info(struct mdrestore_struct *mdres,
1890                            struct async_work *async)
1891 {
1892         struct btrfs_super_block *super;
1893         u8 *buffer = NULL;
1894         u8 *outbuf;
1895         int ret;
1896
1897         /* We've already been initialized */
1898         if (mdres->nodesize)
1899                 return 0;
1900
1901         if (mdres->compress_method == COMPRESS_ZLIB) {
1902                 size_t size = MAX_PENDING_SIZE * 2;
1903
1904                 buffer = malloc(MAX_PENDING_SIZE * 2);
1905                 if (!buffer)
1906                         return -ENOMEM;
1907                 ret = uncompress(buffer, (unsigned long *)&size,
1908                                  async->buffer, async->bufsize);
1909                 if (ret != Z_OK) {
1910                         error("decompressiion failed with %d", ret);
1911                         free(buffer);
1912                         return -EIO;
1913                 }
1914                 outbuf = buffer;
1915         } else {
1916                 outbuf = async->buffer;
1917         }
1918
1919         super = (struct btrfs_super_block *)outbuf;
1920         mdres->nodesize = btrfs_super_nodesize(super);
1921         memcpy(mdres->fsid, super->fsid, BTRFS_FSID_SIZE);
1922         memcpy(mdres->uuid, super->dev_item.uuid,
1923                        BTRFS_UUID_SIZE);
1924         mdres->devid = le64_to_cpu(super->dev_item.devid);
1925         free(buffer);
1926         return 0;
1927 }
1928
1929 static int add_cluster(struct meta_cluster *cluster,
1930                        struct mdrestore_struct *mdres, u64 *next)
1931 {
1932         struct meta_cluster_item *item;
1933         struct meta_cluster_header *header = &cluster->header;
1934         struct async_work *async;
1935         u64 bytenr;
1936         u32 i, nritems;
1937         int ret;
1938
1939         mdres->compress_method = header->compress;
1940
1941         bytenr = le64_to_cpu(header->bytenr) + BLOCK_SIZE;
1942         nritems = le32_to_cpu(header->nritems);
1943         for (i = 0; i < nritems; i++) {
1944                 item = &cluster->items[i];
1945                 async = calloc(1, sizeof(*async));
1946                 if (!async) {
1947                         error("not enough memory for async data");
1948                         return -ENOMEM;
1949                 }
1950                 async->start = le64_to_cpu(item->bytenr);
1951                 async->bufsize = le32_to_cpu(item->size);
1952                 async->buffer = malloc(async->bufsize);
1953                 if (!async->buffer) {
1954                         error("not enough memory for async buffer");
1955                         free(async);
1956                         return -ENOMEM;
1957                 }
1958                 ret = fread(async->buffer, async->bufsize, 1, mdres->in);
1959                 if (ret != 1) {
1960                         error("unable to read buffer: %s", strerror(errno));
1961                         free(async->buffer);
1962                         free(async);
1963                         return -EIO;
1964                 }
1965                 bytenr += async->bufsize;
1966
1967                 pthread_mutex_lock(&mdres->mutex);
1968                 if (async->start == BTRFS_SUPER_INFO_OFFSET) {
1969                         ret = fill_mdres_info(mdres, async);
1970                         if (ret) {
1971                                 error("unable to set up restore state");
1972                                 pthread_mutex_unlock(&mdres->mutex);
1973                                 free(async->buffer);
1974                                 free(async);
1975                                 return ret;
1976                         }
1977                 }
1978                 list_add_tail(&async->list, &mdres->list);
1979                 mdres->num_items++;
1980                 pthread_cond_signal(&mdres->cond);
1981                 pthread_mutex_unlock(&mdres->mutex);
1982         }
1983         if (bytenr & BLOCK_MASK) {
1984                 char buffer[BLOCK_MASK];
1985                 size_t size = BLOCK_SIZE - (bytenr & BLOCK_MASK);
1986
1987                 bytenr += size;
1988                 ret = fread(buffer, size, 1, mdres->in);
1989                 if (ret != 1) {
1990                         error("failed to read buffer: %s", strerror(errno));
1991                         return -EIO;
1992                 }
1993         }
1994         *next = bytenr;
1995         return 0;
1996 }
1997
1998 static int wait_for_worker(struct mdrestore_struct *mdres)
1999 {
2000         int ret = 0;
2001
2002         pthread_mutex_lock(&mdres->mutex);
2003         ret = mdres->error;
2004         while (!ret && mdres->num_items > 0) {
2005                 struct timespec ts = {
2006                         .tv_sec = 0,
2007                         .tv_nsec = 10000000,
2008                 };
2009                 pthread_mutex_unlock(&mdres->mutex);
2010                 nanosleep(&ts, NULL);
2011                 pthread_mutex_lock(&mdres->mutex);
2012                 ret = mdres->error;
2013         }
2014         pthread_mutex_unlock(&mdres->mutex);
2015         return ret;
2016 }
2017
2018 static int read_chunk_block(struct mdrestore_struct *mdres, u8 *buffer,
2019                             u64 bytenr, u64 item_bytenr, u32 bufsize,
2020                             u64 cluster_bytenr)
2021 {
2022         struct extent_buffer *eb;
2023         int ret = 0;
2024         int i;
2025
2026         eb = alloc_dummy_eb(bytenr, mdres->nodesize);
2027         if (!eb) {
2028                 ret = -ENOMEM;
2029                 goto out;
2030         }
2031
2032         while (item_bytenr != bytenr) {
2033                 buffer += mdres->nodesize;
2034                 item_bytenr += mdres->nodesize;
2035         }
2036
2037         memcpy(eb->data, buffer, mdres->nodesize);
2038         if (btrfs_header_bytenr(eb) != bytenr) {
2039                 error("eb bytenr does not match found bytenr: %llu != %llu",
2040                                 (unsigned long long)btrfs_header_bytenr(eb),
2041                                 (unsigned long long)bytenr);
2042                 ret = -EIO;
2043                 goto out;
2044         }
2045
2046         if (memcmp(mdres->fsid, eb->data + offsetof(struct btrfs_header, fsid),
2047                    BTRFS_FSID_SIZE)) {
2048                 error("filesystem UUID of eb %llu does not match",
2049                                 (unsigned long long)bytenr);
2050                 ret = -EIO;
2051                 goto out;
2052         }
2053
2054         if (btrfs_header_owner(eb) != BTRFS_CHUNK_TREE_OBJECTID) {
2055                 error("wrong eb %llu owner %llu",
2056                                 (unsigned long long)bytenr,
2057                                 (unsigned long long)btrfs_header_owner(eb));
2058                 ret = -EIO;
2059                 goto out;
2060         }
2061
2062         for (i = 0; i < btrfs_header_nritems(eb); i++) {
2063                 struct btrfs_chunk *chunk;
2064                 struct fs_chunk *fs_chunk;
2065                 struct btrfs_key key;
2066                 u64 type;
2067
2068                 if (btrfs_header_level(eb)) {
2069                         u64 blockptr = btrfs_node_blockptr(eb, i);
2070
2071                         ret = search_for_chunk_blocks(mdres, blockptr,
2072                                                       cluster_bytenr);
2073                         if (ret)
2074                                 break;
2075                         continue;
2076                 }
2077
2078                 /* Yay a leaf!  We loves leafs! */
2079                 btrfs_item_key_to_cpu(eb, &key, i);
2080                 if (key.type != BTRFS_CHUNK_ITEM_KEY)
2081                         continue;
2082
2083                 fs_chunk = malloc(sizeof(struct fs_chunk));
2084                 if (!fs_chunk) {
2085                         error("not enough memory to allocate chunk");
2086                         ret = -ENOMEM;
2087                         break;
2088                 }
2089                 memset(fs_chunk, 0, sizeof(*fs_chunk));
2090                 chunk = btrfs_item_ptr(eb, i, struct btrfs_chunk);
2091
2092                 fs_chunk->logical = key.offset;
2093                 fs_chunk->physical = btrfs_stripe_offset_nr(eb, chunk, 0);
2094                 fs_chunk->bytes = btrfs_chunk_length(eb, chunk);
2095                 INIT_LIST_HEAD(&fs_chunk->list);
2096                 if (tree_search(&mdres->physical_tree, &fs_chunk->p,
2097                                 physical_cmp, 1) != NULL)
2098                         list_add(&fs_chunk->list, &mdres->overlapping_chunks);
2099                 else
2100                         tree_insert(&mdres->physical_tree, &fs_chunk->p,
2101                                     physical_cmp);
2102
2103                 type = btrfs_chunk_type(eb, chunk);
2104                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2105                         fs_chunk->physical_dup =
2106                                         btrfs_stripe_offset_nr(eb, chunk, 1);
2107                 }
2108
2109                 if (fs_chunk->physical_dup + fs_chunk->bytes >
2110                     mdres->last_physical_offset)
2111                         mdres->last_physical_offset = fs_chunk->physical_dup +
2112                                 fs_chunk->bytes;
2113                 else if (fs_chunk->physical + fs_chunk->bytes >
2114                     mdres->last_physical_offset)
2115                         mdres->last_physical_offset = fs_chunk->physical +
2116                                 fs_chunk->bytes;
2117                 mdres->alloced_chunks += fs_chunk->bytes;
2118                 /* in dup case, fs_chunk->bytes should add twice */
2119                 if (fs_chunk->physical_dup)
2120                         mdres->alloced_chunks += fs_chunk->bytes;
2121                 tree_insert(&mdres->chunk_tree, &fs_chunk->l, chunk_cmp);
2122         }
2123 out:
2124         free(eb);
2125         return ret;
2126 }
2127
2128 /* If you have to ask you aren't worthy */
2129 static int search_for_chunk_blocks(struct mdrestore_struct *mdres,
2130                                    u64 search, u64 cluster_bytenr)
2131 {
2132         struct meta_cluster *cluster;
2133         struct meta_cluster_header *header;
2134         struct meta_cluster_item *item;
2135         u64 current_cluster = cluster_bytenr, bytenr;
2136         u64 item_bytenr;
2137         u32 bufsize, nritems, i;
2138         u32 max_size = MAX_PENDING_SIZE * 2;
2139         u8 *buffer, *tmp = NULL;
2140         int ret = 0;
2141
2142         cluster = malloc(BLOCK_SIZE);
2143         if (!cluster) {
2144                 error("not enough memory for cluster");
2145                 return -ENOMEM;
2146         }
2147
2148         buffer = malloc(max_size);
2149         if (!buffer) {
2150                 error("not enough memory for buffer");
2151                 free(cluster);
2152                 return -ENOMEM;
2153         }
2154
2155         if (mdres->compress_method == COMPRESS_ZLIB) {
2156                 tmp = malloc(max_size);
2157                 if (!tmp) {
2158                         error("not enough memory for buffer");
2159                         free(cluster);
2160                         free(buffer);
2161                         return -ENOMEM;
2162                 }
2163         }
2164
2165         bytenr = current_cluster;
2166         while (1) {
2167                 if (fseek(mdres->in, current_cluster, SEEK_SET)) {
2168                         error("seek failed: %s\n", strerror(errno));
2169                         ret = -EIO;
2170                         break;
2171                 }
2172
2173                 ret = fread(cluster, BLOCK_SIZE, 1, mdres->in);
2174                 if (ret == 0) {
2175                         if (cluster_bytenr != 0) {
2176                                 cluster_bytenr = 0;
2177                                 current_cluster = 0;
2178                                 bytenr = 0;
2179                                 continue;
2180                         }
2181                         error(
2182         "unknown state after reading cluster at %llu, probably crrupted data",
2183                                         cluster_bytenr);
2184                         ret = -EIO;
2185                         break;
2186                 } else if (ret < 0) {
2187                         error("unable to read image at %llu: %s",
2188                                         (unsigned long long)cluster_bytenr,
2189                                         strerror(errno));
2190                         break;
2191                 }
2192                 ret = 0;
2193
2194                 header = &cluster->header;
2195                 if (le64_to_cpu(header->magic) != HEADER_MAGIC ||
2196                     le64_to_cpu(header->bytenr) != current_cluster) {
2197                         error("bad header in metadump image");
2198                         ret = -EIO;
2199                         break;
2200                 }
2201
2202                 bytenr += BLOCK_SIZE;
2203                 nritems = le32_to_cpu(header->nritems);
2204                 for (i = 0; i < nritems; i++) {
2205                         size_t size;
2206
2207                         item = &cluster->items[i];
2208                         bufsize = le32_to_cpu(item->size);
2209                         item_bytenr = le64_to_cpu(item->bytenr);
2210
2211                         if (bufsize > max_size) {
2212                                 error("item %u too big: %u > %u", i, bufsize,
2213                                                 max_size);
2214                                 ret = -EIO;
2215                                 break;
2216                         }
2217
2218                         if (mdres->compress_method == COMPRESS_ZLIB) {
2219                                 ret = fread(tmp, bufsize, 1, mdres->in);
2220                                 if (ret != 1) {
2221                                         error("read error: %s", strerror(errno));
2222                                         ret = -EIO;
2223                                         break;
2224                                 }
2225
2226                                 size = max_size;
2227                                 ret = uncompress(buffer,
2228                                                  (unsigned long *)&size, tmp,
2229                                                  bufsize);
2230                                 if (ret != Z_OK) {
2231                                         error("decompressiion failed with %d",
2232                                                         ret);
2233                                         ret = -EIO;
2234                                         break;
2235                                 }
2236                         } else {
2237                                 ret = fread(buffer, bufsize, 1, mdres->in);
2238                                 if (ret != 1) {
2239                                         error("read error: %s",
2240                                                         strerror(errno));
2241                                         ret = -EIO;
2242                                         break;
2243                                 }
2244                                 size = bufsize;
2245                         }
2246                         ret = 0;
2247
2248                         if (item_bytenr <= search &&
2249                             item_bytenr + size > search) {
2250                                 ret = read_chunk_block(mdres, buffer, search,
2251                                                        item_bytenr, size,
2252                                                        current_cluster);
2253                                 if (!ret)
2254                                         ret = 1;
2255                                 break;
2256                         }
2257                         bytenr += bufsize;
2258                 }
2259                 if (ret) {
2260                         if (ret > 0)
2261                                 ret = 0;
2262                         break;
2263                 }
2264                 if (bytenr & BLOCK_MASK)
2265                         bytenr += BLOCK_SIZE - (bytenr & BLOCK_MASK);
2266                 current_cluster = bytenr;
2267         }
2268
2269         free(tmp);
2270         free(buffer);
2271         free(cluster);
2272         return ret;
2273 }
2274
2275 static int build_chunk_tree(struct mdrestore_struct *mdres,
2276                             struct meta_cluster *cluster)
2277 {
2278         struct btrfs_super_block *super;
2279         struct meta_cluster_header *header;
2280         struct meta_cluster_item *item = NULL;
2281         u64 chunk_root_bytenr = 0;
2282         u32 i, nritems;
2283         u64 bytenr = 0;
2284         u8 *buffer;
2285         int ret;
2286
2287         /* We can't seek with stdin so don't bother doing this */
2288         if (mdres->in == stdin)
2289                 return 0;
2290
2291         ret = fread(cluster, BLOCK_SIZE, 1, mdres->in);
2292         if (ret <= 0) {
2293                 error("unable to read cluster: %s", strerror(errno));
2294                 return -EIO;
2295         }
2296         ret = 0;
2297
2298         header = &cluster->header;
2299         if (le64_to_cpu(header->magic) != HEADER_MAGIC ||
2300             le64_to_cpu(header->bytenr) != 0) {
2301                 error("bad header in metadump image");
2302                 return -EIO;
2303         }
2304
2305         bytenr += BLOCK_SIZE;
2306         mdres->compress_method = header->compress;
2307         nritems = le32_to_cpu(header->nritems);
2308         for (i = 0; i < nritems; i++) {
2309                 item = &cluster->items[i];
2310
2311                 if (le64_to_cpu(item->bytenr) == BTRFS_SUPER_INFO_OFFSET)
2312                         break;
2313                 bytenr += le32_to_cpu(item->size);
2314                 if (fseek(mdres->in, le32_to_cpu(item->size), SEEK_CUR)) {
2315                         error("seek failed: %s\n", strerror(errno));
2316                         return -EIO;
2317                 }
2318         }
2319
2320         if (!item || le64_to_cpu(item->bytenr) != BTRFS_SUPER_INFO_OFFSET) {
2321                 error("did not find superblock at %llu",
2322                                 le64_to_cpu(item->bytenr));
2323                 return -EINVAL;
2324         }
2325
2326         buffer = malloc(le32_to_cpu(item->size));
2327         if (!buffer) {
2328                 error("not enough memory to allocate buffer");
2329                 return -ENOMEM;
2330         }
2331
2332         ret = fread(buffer, le32_to_cpu(item->size), 1, mdres->in);
2333         if (ret != 1) {
2334                 error("unable to read buffer: %s", strerror(errno));
2335                 free(buffer);
2336                 return -EIO;
2337         }
2338
2339         if (mdres->compress_method == COMPRESS_ZLIB) {
2340                 size_t size = MAX_PENDING_SIZE * 2;
2341                 u8 *tmp;
2342
2343                 tmp = malloc(MAX_PENDING_SIZE * 2);
2344                 if (!tmp) {
2345                         free(buffer);
2346                         return -ENOMEM;
2347                 }
2348                 ret = uncompress(tmp, (unsigned long *)&size,
2349                                  buffer, le32_to_cpu(item->size));
2350                 if (ret != Z_OK) {
2351                         error("decompressiion failed with %d", ret);
2352                         free(buffer);
2353                         free(tmp);
2354                         return -EIO;
2355                 }
2356                 free(buffer);
2357                 buffer = tmp;
2358         }
2359
2360         pthread_mutex_lock(&mdres->mutex);
2361         super = (struct btrfs_super_block *)buffer;
2362         chunk_root_bytenr = btrfs_super_chunk_root(super);
2363         mdres->nodesize = btrfs_super_nodesize(super);
2364         memcpy(mdres->fsid, super->fsid, BTRFS_FSID_SIZE);
2365         memcpy(mdres->uuid, super->dev_item.uuid,
2366                        BTRFS_UUID_SIZE);
2367         mdres->devid = le64_to_cpu(super->dev_item.devid);
2368         free(buffer);
2369         pthread_mutex_unlock(&mdres->mutex);
2370
2371         return search_for_chunk_blocks(mdres, chunk_root_bytenr, 0);
2372 }
2373
2374 static int range_contains_super(u64 physical, u64 bytes)
2375 {
2376         u64 super_bytenr;
2377         int i;
2378
2379         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
2380                 super_bytenr = btrfs_sb_offset(i);
2381                 if (super_bytenr >= physical &&
2382                     super_bytenr < physical + bytes)
2383                         return 1;
2384         }
2385
2386         return 0;
2387 }
2388
2389 static void remap_overlapping_chunks(struct mdrestore_struct *mdres)
2390 {
2391         struct fs_chunk *fs_chunk;
2392
2393         while (!list_empty(&mdres->overlapping_chunks)) {
2394                 fs_chunk = list_first_entry(&mdres->overlapping_chunks,
2395                                             struct fs_chunk, list);
2396                 list_del_init(&fs_chunk->list);
2397                 if (range_contains_super(fs_chunk->physical,
2398                                          fs_chunk->bytes)) {
2399                         warning(
2400 "remapping a chunk that had a super mirror inside of it, clearing space cache so we don't end up with corruption");
2401                         mdres->clear_space_cache = 1;
2402                 }
2403                 fs_chunk->physical = mdres->last_physical_offset;
2404                 tree_insert(&mdres->physical_tree, &fs_chunk->p, physical_cmp);
2405                 mdres->last_physical_offset += fs_chunk->bytes;
2406         }
2407 }
2408
2409 static int fixup_devices(struct btrfs_fs_info *fs_info,
2410                          struct mdrestore_struct *mdres, off_t dev_size)
2411 {
2412         struct btrfs_trans_handle *trans;
2413         struct btrfs_dev_item *dev_item;
2414         struct btrfs_path *path;
2415         struct extent_buffer *leaf;
2416         struct btrfs_root *root = fs_info->chunk_root;
2417         struct btrfs_key key;
2418         u64 devid, cur_devid;
2419         int ret;
2420
2421         path = btrfs_alloc_path();
2422         if (!path) {
2423                 error("not enough memory to allocate path");
2424                 return -ENOMEM;
2425         }
2426
2427         trans = btrfs_start_transaction(fs_info->tree_root, 1);
2428         if (IS_ERR(trans)) {
2429                 error("cannot starting transaction %ld", PTR_ERR(trans));
2430                 btrfs_free_path(path);
2431                 return PTR_ERR(trans);
2432         }
2433
2434         dev_item = &fs_info->super_copy->dev_item;
2435
2436         devid = btrfs_stack_device_id(dev_item);
2437
2438         btrfs_set_stack_device_total_bytes(dev_item, dev_size);
2439         btrfs_set_stack_device_bytes_used(dev_item, mdres->alloced_chunks);
2440
2441         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2442         key.type = BTRFS_DEV_ITEM_KEY;
2443         key.offset = 0;
2444
2445 again:
2446         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2447         if (ret < 0) {
2448                 error("search failed: %d", ret);
2449                 exit(1);
2450         }
2451
2452         while (1) {
2453                 leaf = path->nodes[0];
2454                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2455                         ret = btrfs_next_leaf(root, path);
2456                         if (ret < 0) {
2457                                 error("cannot go to next leaf %d", ret);
2458                                 exit(1);
2459                         }
2460                         if (ret > 0) {
2461                                 ret = 0;
2462                                 break;
2463                         }
2464                         leaf = path->nodes[0];
2465                 }
2466
2467                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2468                 if (key.type > BTRFS_DEV_ITEM_KEY)
2469                         break;
2470                 if (key.type != BTRFS_DEV_ITEM_KEY) {
2471                         path->slots[0]++;
2472                         continue;
2473                 }
2474
2475                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2476                                           struct btrfs_dev_item);
2477                 cur_devid = btrfs_device_id(leaf, dev_item);
2478                 if (devid != cur_devid) {
2479                         ret = btrfs_del_item(trans, root, path);
2480                         if (ret) {
2481                                 error("cannot delete item: %d", ret);
2482                                 exit(1);
2483                         }
2484                         btrfs_release_path(path);
2485                         goto again;
2486                 }
2487
2488                 btrfs_set_device_total_bytes(leaf, dev_item, dev_size);
2489                 btrfs_set_device_bytes_used(leaf, dev_item,
2490                                             mdres->alloced_chunks);
2491                 btrfs_mark_buffer_dirty(leaf);
2492                 path->slots[0]++;
2493         }
2494
2495         btrfs_free_path(path);
2496         ret = btrfs_commit_transaction(trans, fs_info->tree_root);
2497         if (ret) {
2498                 error("unable to commit transaction: %d", ret);
2499                 return ret;
2500         }
2501         return 0;
2502 }
2503
2504 static int restore_metadump(const char *input, FILE *out, int old_restore,
2505                             int num_threads, int fixup_offset,
2506                             const char *target, int multi_devices)
2507 {
2508         struct meta_cluster *cluster = NULL;
2509         struct meta_cluster_header *header;
2510         struct mdrestore_struct mdrestore;
2511         struct btrfs_fs_info *info = NULL;
2512         u64 bytenr = 0;
2513         FILE *in = NULL;
2514         int ret = 0;
2515
2516         if (!strcmp(input, "-")) {
2517                 in = stdin;
2518         } else {
2519                 in = fopen(input, "r");
2520                 if (!in) {
2521                         error("unable to open metadump image: %s",
2522                                         strerror(errno));
2523                         return 1;
2524                 }
2525         }
2526
2527         /* NOTE: open with write mode */
2528         if (fixup_offset) {
2529                 info = open_ctree_fs_info(target, 0, 0, 0,
2530                                           OPEN_CTREE_WRITES |
2531                                           OPEN_CTREE_RESTORE |
2532                                           OPEN_CTREE_PARTIAL);
2533                 if (!info) {
2534                         error("open ctree failed");
2535                         ret = -EIO;
2536                         goto failed_open;
2537                 }
2538         }
2539
2540         cluster = malloc(BLOCK_SIZE);
2541         if (!cluster) {
2542                 error("not enough memory for cluster");
2543                 ret = -ENOMEM;
2544                 goto failed_info;
2545         }
2546
2547         ret = mdrestore_init(&mdrestore, in, out, old_restore, num_threads,
2548                              fixup_offset, info, multi_devices);
2549         if (ret) {
2550                 error("failed to intialize metadata restore state: %d", ret);
2551                 goto failed_cluster;
2552         }
2553
2554         if (!multi_devices && !old_restore) {
2555                 ret = build_chunk_tree(&mdrestore, cluster);
2556                 if (ret)
2557                         goto out;
2558                 if (!list_empty(&mdrestore.overlapping_chunks))
2559                         remap_overlapping_chunks(&mdrestore);
2560         }
2561
2562         if (in != stdin && fseek(in, 0, SEEK_SET)) {
2563                 error("seek failed: %s\n", strerror(errno));
2564                 goto out;
2565         }
2566
2567         while (!mdrestore.error) {
2568                 ret = fread(cluster, BLOCK_SIZE, 1, in);
2569                 if (!ret)
2570                         break;
2571
2572                 header = &cluster->header;
2573                 if (le64_to_cpu(header->magic) != HEADER_MAGIC ||
2574                     le64_to_cpu(header->bytenr) != bytenr) {
2575                         error("bad header in metadump image");
2576                         ret = -EIO;
2577                         break;
2578                 }
2579                 ret = add_cluster(cluster, &mdrestore, &bytenr);
2580                 if (ret) {
2581                         error("failed to add cluster: %d", ret);
2582                         break;
2583                 }
2584         }
2585         ret = wait_for_worker(&mdrestore);
2586
2587         if (!ret && !multi_devices && !old_restore) {
2588                 struct btrfs_root *root;
2589                 struct stat st;
2590
2591                 root = open_ctree_fd(fileno(out), target, 0,
2592                                           OPEN_CTREE_PARTIAL |
2593                                           OPEN_CTREE_WRITES |
2594                                           OPEN_CTREE_NO_DEVICES);
2595                 if (!root) {
2596                         error("open ctree failed in %s", target);
2597                         ret = -EIO;
2598                         goto out;
2599                 }
2600                 info = root->fs_info;
2601
2602                 if (stat(target, &st)) {
2603                         error("stat %s failed: %s", target, strerror(errno));
2604                         close_ctree(info->chunk_root);
2605                         free(cluster);
2606                         return 1;
2607                 }
2608
2609                 ret = fixup_devices(info, &mdrestore, st.st_size);
2610                 close_ctree(info->chunk_root);
2611                 if (ret)
2612                         goto out;
2613         }
2614 out:
2615         mdrestore_destroy(&mdrestore, num_threads);
2616 failed_cluster:
2617         free(cluster);
2618 failed_info:
2619         if (fixup_offset && info)
2620                 close_ctree(info->chunk_root);
2621 failed_open:
2622         if (in != stdin)
2623                 fclose(in);
2624         return ret;
2625 }
2626
2627 static int update_disk_super_on_device(struct btrfs_fs_info *info,
2628                                        const char *other_dev, u64 cur_devid)
2629 {
2630         struct btrfs_key key;
2631         struct extent_buffer *leaf;
2632         struct btrfs_path path;
2633         struct btrfs_dev_item *dev_item;
2634         struct btrfs_super_block *disk_super;
2635         char dev_uuid[BTRFS_UUID_SIZE];
2636         char fs_uuid[BTRFS_UUID_SIZE];
2637         u64 devid, type, io_align, io_width;
2638         u64 sector_size, total_bytes, bytes_used;
2639         char buf[BTRFS_SUPER_INFO_SIZE];
2640         int fp = -1;
2641         int ret;
2642
2643         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2644         key.type = BTRFS_DEV_ITEM_KEY;
2645         key.offset = cur_devid;
2646
2647         btrfs_init_path(&path);
2648         ret = btrfs_search_slot(NULL, info->chunk_root, &key, &path, 0, 0);
2649         if (ret) {
2650                 error("search key failed: %d", ret);
2651                 ret = -EIO;
2652                 goto out;
2653         }
2654
2655         leaf = path.nodes[0];
2656         dev_item = btrfs_item_ptr(leaf, path.slots[0],
2657                                   struct btrfs_dev_item);
2658
2659         devid = btrfs_device_id(leaf, dev_item);
2660         if (devid != cur_devid) {
2661                 error("devid mismatch: %llu != %llu",
2662                                 (unsigned long long)devid,
2663                                 (unsigned long long)cur_devid);
2664                 ret = -EIO;
2665                 goto out;
2666         }
2667
2668         type = btrfs_device_type(leaf, dev_item);
2669         io_align = btrfs_device_io_align(leaf, dev_item);
2670         io_width = btrfs_device_io_width(leaf, dev_item);
2671         sector_size = btrfs_device_sector_size(leaf, dev_item);
2672         total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2673         bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2674         read_extent_buffer(leaf, dev_uuid, (unsigned long)btrfs_device_uuid(dev_item), BTRFS_UUID_SIZE);
2675         read_extent_buffer(leaf, fs_uuid, (unsigned long)btrfs_device_fsid(dev_item), BTRFS_UUID_SIZE);
2676
2677         btrfs_release_path(&path);
2678
2679         printf("update disk super on %s devid=%llu\n", other_dev, devid);
2680
2681         /* update other devices' super block */
2682         fp = open(other_dev, O_CREAT | O_RDWR, 0600);
2683         if (fp < 0) {
2684                 error("could not open %s: %s", other_dev, strerror(errno));
2685                 ret = -EIO;
2686                 goto out;
2687         }
2688
2689         memcpy(buf, info->super_copy, BTRFS_SUPER_INFO_SIZE);
2690
2691         disk_super = (struct btrfs_super_block *)buf;
2692         dev_item = &disk_super->dev_item;
2693
2694         btrfs_set_stack_device_type(dev_item, type);
2695         btrfs_set_stack_device_id(dev_item, devid);
2696         btrfs_set_stack_device_total_bytes(dev_item, total_bytes);
2697         btrfs_set_stack_device_bytes_used(dev_item, bytes_used);
2698         btrfs_set_stack_device_io_align(dev_item, io_align);
2699         btrfs_set_stack_device_io_width(dev_item, io_width);
2700         btrfs_set_stack_device_sector_size(dev_item, sector_size);
2701         memcpy(dev_item->uuid, dev_uuid, BTRFS_UUID_SIZE);
2702         memcpy(dev_item->fsid, fs_uuid, BTRFS_UUID_SIZE);
2703         csum_block((u8 *)buf, BTRFS_SUPER_INFO_SIZE);
2704
2705         ret = pwrite64(fp, buf, BTRFS_SUPER_INFO_SIZE, BTRFS_SUPER_INFO_OFFSET);
2706         if (ret != BTRFS_SUPER_INFO_SIZE) {
2707                 if (ret < 0)
2708                         error("cannot write superblock: %s", strerror(ret));
2709                 else
2710                         error("cannot write superblock");
2711                 ret = -EIO;
2712                 goto out;
2713         }
2714
2715         write_backup_supers(fp, (u8 *)buf);
2716
2717 out:
2718         if (fp != -1)
2719                 close(fp);
2720         return ret;
2721 }
2722
2723 static void print_usage(int ret)
2724 {
2725         printf("usage: btrfs-image [options] source target\n");
2726         printf("\t-r      \trestore metadump image\n");
2727         printf("\t-c value\tcompression level (0 ~ 9)\n");
2728         printf("\t-t value\tnumber of threads (1 ~ 32)\n");
2729         printf("\t-o      \tdon't mess with the chunk tree when restoring\n");
2730         printf("\t-s      \tsanitize file names, use once to just use garbage, use twice if you want crc collisions\n");
2731         printf("\t-w      \twalk all trees instead of using extent tree, do this if your extent tree is broken\n");
2732         printf("\t-m       \trestore for multiple devices\n");
2733         printf("\n");
2734         printf("\tIn the dump mode, source is the btrfs device and target is the output file (use '-' for stdout).\n");
2735         printf("\tIn the restore mode, source is the dumped image and target is the btrfs device/file.\n");
2736         exit(ret);
2737 }
2738
2739 int main(int argc, char *argv[])
2740 {
2741         char *source;
2742         char *target;
2743         u64 num_threads = 0;
2744         u64 compress_level = 0;
2745         int create = 1;
2746         int old_restore = 0;
2747         int walk_trees = 0;
2748         int multi_devices = 0;
2749         int ret;
2750         int sanitize = 0;
2751         int dev_cnt = 0;
2752         int usage_error = 0;
2753         FILE *out;
2754
2755         while (1) {
2756                 static const struct option long_options[] = {
2757                         { "help", no_argument, NULL, GETOPT_VAL_HELP},
2758                         { NULL, 0, NULL, 0 }
2759                 };
2760                 int c = getopt_long(argc, argv, "rc:t:oswm", long_options, NULL);
2761                 if (c < 0)
2762                         break;
2763                 switch (c) {
2764                 case 'r':
2765                         create = 0;
2766                         break;
2767                 case 't':
2768                         num_threads = arg_strtou64(optarg);
2769                         if (num_threads > 32) {
2770                                 error("number of threads out of range: %llu",
2771                                         (unsigned long long)num_threads);
2772                                 return 1;
2773                         }
2774                         break;
2775                 case 'c':
2776                         compress_level = arg_strtou64(optarg);
2777                         if (compress_level > 9) {
2778                                 error("compression level out of range: %llu",
2779                                         (unsigned long long)compress_level);
2780                                 return 1;
2781                         }
2782                         break;
2783                 case 'o':
2784                         old_restore = 1;
2785                         break;
2786                 case 's':
2787                         sanitize++;
2788                         break;
2789                 case 'w':
2790                         walk_trees = 1;
2791                         break;
2792                 case 'm':
2793                         create = 0;
2794                         multi_devices = 1;
2795                         break;
2796                         case GETOPT_VAL_HELP:
2797                 default:
2798                         print_usage(c != GETOPT_VAL_HELP);
2799                 }
2800         }
2801
2802         set_argv0(argv);
2803         if (check_argc_min(argc - optind, 2))
2804                 print_usage(1);
2805
2806         dev_cnt = argc - optind - 1;
2807
2808         if (create) {
2809                 if (old_restore) {
2810                         error(
2811                         "create and restore cannot be used at the same time");
2812                         usage_error++;
2813                 }
2814         } else {
2815                 if (walk_trees || sanitize || compress_level) {
2816                         error(
2817                         "useing -w, -s, -c options for restore makes no sense");
2818                         usage_error++;
2819                 }
2820                 if (multi_devices && dev_cnt < 2) {
2821                         error("not enough devices specified for -m option");
2822                         usage_error++;
2823                 }
2824                 if (!multi_devices && dev_cnt != 1) {
2825                         error("accepts only 1 device without -m option");
2826                         usage_error++;
2827                 }
2828         }
2829
2830         if (usage_error)
2831                 print_usage(1);
2832
2833         source = argv[optind];
2834         target = argv[optind + 1];
2835
2836         if (create && !strcmp(target, "-")) {
2837                 out = stdout;
2838         } else {
2839                 out = fopen(target, "w+");
2840                 if (!out) {
2841                         error("unable to create target file %s", target);
2842                         exit(1);
2843                 }
2844         }
2845
2846         if (compress_level > 0 || create == 0) {
2847                 if (num_threads == 0) {
2848                         long tmp = sysconf(_SC_NPROCESSORS_ONLN);
2849
2850                         if (tmp <= 0)
2851                                 tmp = 1;
2852                         num_threads = tmp;
2853                 }
2854         } else {
2855                 num_threads = 0;
2856         }
2857
2858         if (create) {
2859                 ret = check_mounted(source);
2860                 if (ret < 0) {
2861                         warning("unable to check mount status of: %s",
2862                                         strerror(-ret));
2863                 } else if (ret) {
2864                         warning("%s already mounted, results may be inaccurate",
2865                                         source);
2866                 }
2867
2868                 ret = create_metadump(source, out, num_threads,
2869                                       compress_level, sanitize, walk_trees);
2870         } else {
2871                 ret = restore_metadump(source, out, old_restore, num_threads,
2872                                        0, target, multi_devices);
2873         }
2874         if (ret) {
2875                 error("%s failed: %s", (create) ? "create" : "restore",
2876                        strerror(errno));
2877                 goto out;
2878         }
2879
2880          /* extended support for multiple devices */
2881         if (!create && multi_devices) {
2882                 struct btrfs_fs_info *info;
2883                 u64 total_devs;
2884                 int i;
2885
2886                 info = open_ctree_fs_info(target, 0, 0, 0,
2887                                           OPEN_CTREE_PARTIAL |
2888                                           OPEN_CTREE_RESTORE);
2889                 if (!info) {
2890                         error("open ctree failed at %s", target);
2891                         return 1;
2892                 }
2893
2894                 total_devs = btrfs_super_num_devices(info->super_copy);
2895                 if (total_devs != dev_cnt) {
2896                         error("it needs %llu devices but has only %d",
2897                                 total_devs, dev_cnt);
2898                         close_ctree(info->chunk_root);
2899                         goto out;
2900                 }
2901
2902                 /* update super block on other disks */
2903                 for (i = 2; i <= dev_cnt; i++) {
2904                         ret = update_disk_super_on_device(info,
2905                                         argv[optind + i], (u64)i);
2906                         if (ret) {
2907                                 error("update disk superblock failed devid %d: %d",
2908                                         i, ret);
2909                                 close_ctree(info->chunk_root);
2910                                 exit(1);
2911                         }
2912                 }
2913
2914                 close_ctree(info->chunk_root);
2915
2916                 /* fix metadata block to map correct chunk */
2917                 ret = restore_metadump(source, out, 0, num_threads, 1,
2918                                        target, 1);
2919                 if (ret) {
2920                         error("unable to fixup metadump: %d", ret);
2921                         exit(1);
2922                 }
2923         }
2924 out:
2925         if (out == stdout) {
2926                 fflush(out);
2927         } else {
2928                 fclose(out);
2929                 if (ret && create) {
2930                         int unlink_ret;
2931
2932                         unlink_ret = unlink(target);
2933                         if (unlink_ret)
2934                                 error("unlink output file %s failed: %s",
2935                                                 target, strerror(errno));
2936                 }
2937         }
2938
2939         btrfs_close_all_devices();
2940
2941         return !!ret;
2942 }