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 #define _XOPEN_SOURCE 500
  20 #define _GNU_SOURCE 1
  21 #include <pthread.h>
  22 #include <stdio.h>
  23 #include <stdlib.h>
  24 #include <sys/types.h>
  25 #include <sys/stat.h>
  26 #include <fcntl.h>
  27 #include <unistd.h>
  28 #include <dirent.h>
  29 #include <zlib.h>
  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 "version.h"
  37
  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 async_work {
  69         struct list_head list;
  70         struct list_head ordered;
  71         u64 start;
  72         u64 size;
  73         u8 *buffer;
  74         size_t bufsize;
  75         int error;
  76 };
  77
  78 struct metadump_struct {
  79         struct btrfs_root *root;
  80         FILE *out;
  81
  82         struct meta_cluster *cluster;
  83
  84         pthread_t *threads;
  85         size_t num_threads;
  86         pthread_mutex_t mutex;
  87         pthread_cond_t cond;
  88
  89         struct list_head list;
  90         struct list_head ordered;
  91         size_t num_items;
  92         size_t num_ready;
  93
  94         u64 pending_start;
  95         u64 pending_size;
  96
  97         int compress_level;
  98         int done;
  99 };
 100
 101 struct mdrestore_struct {
 102         FILE *in;
 103         FILE *out;
 104
 105         pthread_t *threads;
 106         size_t num_threads;
 107         pthread_mutex_t mutex;
 108         pthread_cond_t cond;
 109
 110         struct list_head list;
 111         size_t num_items;
 112
 113         int compress_method;
 114         int done;
 115         int error;
 116 };
 117
 118 static void csum_block(u8 *buf, size_t len)
 119 {
 120         char result[BTRFS_CRC32_SIZE];
 121         u32 crc = ~(u32)0;
 122         crc = crc32c(crc, buf + BTRFS_CSUM_SIZE, len - BTRFS_CSUM_SIZE);
 123         btrfs_csum_final(crc, result);
 124         memcpy(buf, result, BTRFS_CRC32_SIZE);
 125 }
 126
 127 /*
 128  * zero inline extents and csum items
 129  */
 130 static void zero_items(u8 *dst, struct extent_buffer *src)
 131 {
 132         struct btrfs_file_extent_item *fi;
 133         struct btrfs_item *item;
 134         struct btrfs_key key;
 135         u32 nritems = btrfs_header_nritems(src);
 136         size_t size;
 137         unsigned long ptr;
 138         int i, extent_type;
 139
 140         for (i = 0; i < nritems; i++) {
 141                 item = btrfs_item_nr(src, i);
 142                 btrfs_item_key_to_cpu(src, &key, i);
 143                 if (key.type == BTRFS_CSUM_ITEM_KEY) {
 144                         size = btrfs_item_size_nr(src, i);
 145                         memset(dst + btrfs_leaf_data(src) +
 146                                btrfs_item_offset_nr(src, i), 0, size);
 147                         continue;
 148                 }
 149                 if (key.type != BTRFS_EXTENT_DATA_KEY)
 150                         continue;
 151
 152                 fi = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
 153                 extent_type = btrfs_file_extent_type(src, fi);
 154                 if (extent_type != BTRFS_FILE_EXTENT_INLINE)
 155                         continue;
 156
 157                 ptr = btrfs_file_extent_inline_start(fi);
 158                 size = btrfs_file_extent_inline_item_len(src, item);
 159                 memset(dst + ptr, 0, size);
 160         }
 161 }
 162
 163 /*
 164  * copy buffer and zero useless data in the buffer
 165  */
 166 static void copy_buffer(u8 *dst, struct extent_buffer *src)
 167 {
 168         int level;
 169         size_t size;
 170         u32 nritems;
 171
 172         memcpy(dst, src->data, src->len);
 173         if (src->start == BTRFS_SUPER_INFO_OFFSET)
 174                 return;
 175
 176         level = btrfs_header_level(src);
 177         nritems = btrfs_header_nritems(src);
 178
 179         if (nritems == 0) {
 180                 size = sizeof(struct btrfs_header);
 181                 memset(dst + size, 0, src->len - size);
 182         } else if (level == 0) {
 183                 size = btrfs_leaf_data(src) +
 184                         btrfs_item_offset_nr(src, nritems - 1) -
 185                         btrfs_item_nr_offset(nritems);
 186                 memset(dst + btrfs_item_nr_offset(nritems), 0, size);
 187                 zero_items(dst, src);
 188         } else {
 189                 size = offsetof(struct btrfs_node, ptrs) +
 190                         sizeof(struct btrfs_key_ptr) * nritems;
 191                 memset(dst + size, 0, src->len - size);
 192         }
 193         csum_block(dst, src->len);
 194 }
 195
 196 static void *dump_worker(void *data)
 197 {
 198         struct metadump_struct *md = (struct metadump_struct *)data;
 199         struct async_work *async;
 200         int ret;
 201
 202         while (1) {
 203                 pthread_mutex_lock(&md->mutex);
 204                 while (list_empty(&md->list)) {
 205                         if (md->done) {
 206                                 pthread_mutex_unlock(&md->mutex);
 207                                 goto out;
 208                         }
 209                         pthread_cond_wait(&md->cond, &md->mutex);
 210                 }
 211                 async = list_entry(md->list.next, struct async_work, list);
 212                 list_del_init(&async->list);
 213                 pthread_mutex_unlock(&md->mutex);
 214
 215                 if (md->compress_level > 0) {
 216                         u8 *orig = async->buffer;
 217
 218                         async->bufsize = compressBound(async->size);
 219                         async->buffer = malloc(async->bufsize);
 220
 221                         ret = compress2(async->buffer,
 222                                          (unsigned long *)&async->bufsize,
 223                                          orig, async->size, md->compress_level);
 224
 225                         if (ret != Z_OK)
 226                                 async->error = 1;
 227
 228                         free(orig);
 229                 }
 230
 231                 pthread_mutex_lock(&md->mutex);
 232                 md->num_ready++;
 233                 pthread_mutex_unlock(&md->mutex);
 234         }
 235 out:
 236         pthread_exit(NULL);
 237 }
 238
 239 static void meta_cluster_init(struct metadump_struct *md, u64 start)
 240 {
 241         struct meta_cluster_header *header;
 242
 243         md->num_items = 0;
 244         md->num_ready = 0;
 245         header = &md->cluster->header;
 246         header->magic = cpu_to_le64(HEADER_MAGIC);
 247         header->bytenr = cpu_to_le64(start);
 248         header->nritems = cpu_to_le32(0);
 249         header->compress = md->compress_level > 0 ?
 250                            COMPRESS_ZLIB : COMPRESS_NONE;
 251 }
 252
 253 static int metadump_init(struct metadump_struct *md, struct btrfs_root *root,
 254                          FILE *out, int num_threads, int compress_level)
 255 {
 256         int i, ret;
 257
 258         memset(md, 0, sizeof(*md));
 259         pthread_cond_init(&md->cond, NULL);
 260         pthread_mutex_init(&md->mutex, NULL);
 261         INIT_LIST_HEAD(&md->list);
 262         INIT_LIST_HEAD(&md->ordered);
 263         md->root = root;
 264         md->out = out;
 265         md->pending_start = (u64)-1;
 266         md->compress_level = compress_level;
 267         md->cluster = calloc(1, BLOCK_SIZE);
 268         if (!md->cluster) {
 269                 pthread_cond_destroy(&md->cond);
 270                 pthread_mutex_destroy(&md->mutex);
 271                 return -ENOMEM;
 272         }
 273
 274         meta_cluster_init(md, 0);
 275         if (!num_threads)
 276                 return 0;
 277
 278         md->num_threads = num_threads;
 279         md->threads = calloc(num_threads, sizeof(pthread_t));
 280         if (!md->threads) {
 281                 free(md->cluster);
 282                 pthread_cond_destroy(&md->cond);
 283                 pthread_mutex_destroy(&md->mutex);
 284                 return -ENOMEM;
 285         }
 286
 287         for (i = 0; i < num_threads; i++) {
 288                 ret = pthread_create(md->threads + i, NULL, dump_worker, md);
 289                 if (ret)
 290                         break;
 291         }
 292
 293         if (ret) {
 294                 pthread_mutex_lock(&md->mutex);
 295                 md->done = 1;
 296                 pthread_cond_broadcast(&md->cond);
 297                 pthread_mutex_unlock(&md->mutex);
 298
 299                 for (i--; i >= 0; i--)
 300                         pthread_join(md->threads[i], NULL);
 301
 302                 pthread_cond_destroy(&md->cond);
 303                 pthread_mutex_destroy(&md->mutex);
 304                 free(md->cluster);
 305                 free(md->threads);
 306         }
 307
 308         return ret;
 309 }
 310
 311 static void metadump_destroy(struct metadump_struct *md)
 312 {
 313         int i;
 314         pthread_mutex_lock(&md->mutex);
 315         md->done = 1;
 316         pthread_cond_broadcast(&md->cond);
 317         pthread_mutex_unlock(&md->mutex);
 318
 319         for (i = 0; i < md->num_threads; i++)
 320                 pthread_join(md->threads[i], NULL);
 321
 322         pthread_cond_destroy(&md->cond);
 323         pthread_mutex_destroy(&md->mutex);
 324         free(md->threads);
 325         free(md->cluster);
 326 }
 327
 328 static int write_zero(FILE *out, size_t size)
 329 {
 330         static char zero[BLOCK_SIZE];
 331         return fwrite(zero, size, 1, out);
 332 }
 333
 334 static int write_buffers(struct metadump_struct *md, u64 *next)
 335 {
 336         struct meta_cluster_header *header = &md->cluster->header;
 337         struct meta_cluster_item *item;
 338         struct async_work *async;
 339         u64 bytenr = 0;
 340         u32 nritems = 0;
 341         int ret;
 342         int err = 0;
 343
 344         if (list_empty(&md->ordered))
 345                 goto out;
 346
 347         /* wait until all buffers are compressed */
 348         while (md->num_items > md->num_ready) {
 349                 struct timespec ts = {
 350                         .tv_sec = 0,
 351                         .tv_nsec = 10000000,
 352                 };
 353                 pthread_mutex_unlock(&md->mutex);
 354                 nanosleep(&ts, NULL);
 355                 pthread_mutex_lock(&md->mutex);
 356         }
 357
 358         /* setup and write index block */
 359         list_for_each_entry(async, &md->ordered, ordered) {
 360                 item = md->cluster->items + nritems;
 361                 item->bytenr = cpu_to_le64(async->start);
 362                 item->size = cpu_to_le32(async->bufsize);
 363                 nritems++;
 364         }
 365         header->nritems = cpu_to_le32(nritems);
 366
 367         ret = fwrite(md->cluster, BLOCK_SIZE, 1, md->out);
 368         if (ret != 1) {
 369                 fprintf(stderr, "Error writing out cluster: %d\n", errno);
 370                 return -EIO;
 371         }
 372
 373         /* write buffers */
 374         bytenr += le64_to_cpu(header->bytenr) + BLOCK_SIZE;
 375         while (!list_empty(&md->ordered)) {
 376                 async = list_entry(md->ordered.next, struct async_work,
 377                                    ordered);
 378                 list_del_init(&async->ordered);
 379
 380                 bytenr += async->bufsize;
 381                 if (!err)
 382                         ret = fwrite(async->buffer, async->bufsize, 1,
 383                                      md->out);
 384                 if (ret != 1) {
 385                         err = -EIO;
 386                         ret = 0;
 387                         fprintf(stderr, "Error writing out cluster: %d\n",
 388                                 errno);
 389                 }
 390
 391                 free(async->buffer);
 392                 free(async);
 393         }
 394
 395         /* zero unused space in the last block */
 396         if (!err && bytenr & BLOCK_MASK) {
 397                 size_t size = BLOCK_SIZE - (bytenr & BLOCK_MASK);
 398
 399                 bytenr += size;
 400                 ret = write_zero(md->out, size);
 401                 if (ret != 1) {
 402                         fprintf(stderr, "Error zeroing out buffer: %d\n",
 403                                 errno);
 404                         err = -EIO;
 405                 }
 406         }
 407 out:
 408         *next = bytenr;
 409         return err;
 410 }
 411
 412 static int flush_pending(struct metadump_struct *md, int done)
 413 {
 414         struct async_work *async = NULL;
 415         struct extent_buffer *eb;
 416         u64 blocksize = md->root->nodesize;
 417         u64 start;
 418         u64 size;
 419         size_t offset;
 420         int ret = 0;
 421
 422         if (md->pending_size) {
 423                 async = calloc(1, sizeof(*async));
 424                 if (!async)
 425                         return -ENOMEM;
 426
 427                 async->start = md->pending_start;
 428                 async->size = md->pending_size;
 429                 async->bufsize = async->size;
 430                 async->buffer = malloc(async->bufsize);
 431                 if (!async->buffer) {
 432                         free(async);
 433                         return -ENOMEM;
 434                 }
 435                 offset = 0;
 436                 start = async->start;
 437                 size = async->size;
 438                 while (size > 0) {
 439                         eb = read_tree_block(md->root, start, blocksize, 0);
 440                         if (!eb) {
 441                                 free(async->buffer);
 442                                 free(async);
 443                                 fprintf(stderr,
 444                                         "Error reading metadata block\n");
 445                                 return -EIO;
 446                         }
 447                         copy_buffer(async->buffer + offset, eb);
 448                         free_extent_buffer(eb);
 449                         start += blocksize;
 450                         offset += blocksize;
 451                         size -= blocksize;
 452                 }
 453
 454                 md->pending_start = (u64)-1;
 455                 md->pending_size = 0;
 456         } else if (!done) {
 457                 return 0;
 458         }
 459
 460         pthread_mutex_lock(&md->mutex);
 461         if (async) {
 462                 list_add_tail(&async->ordered, &md->ordered);
 463                 md->num_items++;
 464                 if (md->compress_level > 0) {
 465                         list_add_tail(&async->list, &md->list);
 466                         pthread_cond_signal(&md->cond);
 467                 } else {
 468                         md->num_ready++;
 469                 }
 470         }
 471         if (md->num_items >= ITEMS_PER_CLUSTER || done) {
 472                 ret = write_buffers(md, &start);
 473                 if (ret)
 474                         fprintf(stderr, "Error writing buffers %d\n",
 475                                 errno);
 476                 else
 477                         meta_cluster_init(md, start);
 478         }
 479         pthread_mutex_unlock(&md->mutex);
 480         return ret;
 481 }
 482
 483 static int add_metadata(u64 start, u64 size, struct metadump_struct *md)
 484 {
 485         int ret;
 486         if (md->pending_size + size > MAX_PENDING_SIZE ||
 487             md->pending_start + md->pending_size != start) {
 488                 ret = flush_pending(md, 0);
 489                 if (ret)
 490                         return ret;
 491                 md->pending_start = start;
 492         }
 493         readahead_tree_block(md->root, start, size, 0);
 494         md->pending_size += size;
 495         return 0;
 496 }
 497
 498 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 499 static int is_tree_block(struct btrfs_root *extent_root,
 500                          struct btrfs_path *path, u64 bytenr)
 501 {
 502         struct extent_buffer *leaf;
 503         struct btrfs_key key;
 504         u64 ref_objectid;
 505         int ret;
 506
 507         leaf = path->nodes[0];
 508         while (1) {
 509                 struct btrfs_extent_ref_v0 *ref_item;
 510                 path->slots[0]++;
 511                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 512                         ret = btrfs_next_leaf(extent_root, path);
 513                         if (ret < 0)
 514                                 return ret;
 515                         if (ret > 0)
 516                                 break;
 517                         leaf = path->nodes[0];
 518                 }
 519                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 520                 if (key.objectid != bytenr)
 521                         break;
 522                 if (key.type != BTRFS_EXTENT_REF_V0_KEY)
 523                         continue;
 524                 ref_item = btrfs_item_ptr(leaf, path->slots[0],
 525                                           struct btrfs_extent_ref_v0);
 526                 ref_objectid = btrfs_ref_objectid_v0(leaf, ref_item);
 527                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID)
 528                         return 1;
 529                 break;
 530         }
 531         return 0;
 532 }
 533 #endif
 534
 535 static int create_metadump(const char *input, FILE *out, int num_threads,
 536                            int compress_level)
 537 {
 538         struct btrfs_root *root;
 539         struct btrfs_root *extent_root;
 540         struct btrfs_path *path = NULL;
 541         struct extent_buffer *leaf;
 542         struct btrfs_extent_item *ei;
 543         struct btrfs_key key;
 544         struct metadump_struct metadump;
 545         u64 bytenr;
 546         u64 num_bytes;
 547         int ret;
 548         int err = 0;
 549
 550         root = open_ctree(input, 0, 0);
 551         if (!root) {
 552                 fprintf(stderr, "Open ctree failed\n");
 553                 return -EIO;
 554         }
 555
 556         BUG_ON(root->nodesize != root->leafsize);
 557
 558         ret = metadump_init(&metadump, root, out, num_threads,
 559                             compress_level);
 560         if (ret) {
 561                 fprintf(stderr, "Error initing metadump %d\n", ret);
 562                 close_ctree(root);
 563                 return ret;
 564         }
 565
 566         ret = add_metadata(BTRFS_SUPER_INFO_OFFSET, 4096, &metadump);
 567         if (ret) {
 568                 fprintf(stderr, "Error adding metadata %d\n", ret);
 569                 err = ret;
 570                 goto out;
 571         }
 572
 573         extent_root = root->fs_info->extent_root;
 574         path = btrfs_alloc_path();
 575         if (!path) {
 576                 fprintf(stderr, "Out of memory allocing path\n");
 577                 err = -ENOMEM;
 578                 goto out;
 579         }
 580         bytenr = BTRFS_SUPER_INFO_OFFSET + 4096;
 581         key.objectid = bytenr;
 582         key.type = BTRFS_EXTENT_ITEM_KEY;
 583         key.offset = 0;
 584
 585         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 586         if (ret < 0) {
 587                 fprintf(stderr, "Error searching extent root %d\n", ret);
 588                 err = ret;
 589                 goto out;
 590         }
 591
 592         while (1) {
 593                 leaf = path->nodes[0];
 594                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 595                         ret = btrfs_next_leaf(extent_root, path);
 596                         if (ret < 0) {
 597                                 fprintf(stderr, "Error going to next leaf %d"
 598                                         "\n", ret);
 599                                 err = ret;
 600                                 goto out;
 601                         }
 602                         if (ret > 0)
 603                                 break;
 604                         leaf = path->nodes[0];
 605                 }
 606
 607                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 608                 if (key.objectid < bytenr ||
 609                     key.type != BTRFS_EXTENT_ITEM_KEY) {
 610                         path->slots[0]++;
 611                         continue;
 612                 }
 613
 614                 bytenr = key.objectid;
 615                 num_bytes = key.offset;
 616
 617                 if (btrfs_item_size_nr(leaf, path->slots[0]) > sizeof(*ei)) {
 618                         ei = btrfs_item_ptr(leaf, path->slots[0],
 619                                             struct btrfs_extent_item);
 620                         if (btrfs_extent_flags(leaf, ei) &
 621                             BTRFS_EXTENT_FLAG_TREE_BLOCK) {
 622                                 ret = add_metadata(bytenr, num_bytes,
 623                                                    &metadump);
 624                                 if (ret) {
 625                                         fprintf(stderr, "Error adding block "
 626                                                 "%d\n", ret);
 627                                         err = ret;
 628                                         goto out;
 629                                 }
 630                         }
 631                 } else {
 632 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 633                         ret = is_tree_block(extent_root, path, bytenr);
 634                         if (ret < 0) {
 635                                 fprintf(stderr, "Error checking tree block "
 636                                         "%d\n", ret);
 637                                 err = ret;
 638                                 goto out;
 639                         }
 640
 641                         if (ret) {
 642                                 ret = add_metadata(bytenr, num_bytes,
 643                                                    &metadump);
 644                                 if (ret) {
 645                                         fprintf(stderr, "Error adding block "
 646                                                 "%d\n", ret);
 647                                         err = ret;
 648                                         goto out;
 649                                 }
 650                         }
 651 #else
 652                         fprintf(stderr, "Either extent tree corruption or "
 653                                 "you haven't built with V0 support\n");
 654                         err = -EIO;
 655                         goto out;
 656 #endif
 657                 }
 658                 bytenr += num_bytes;
 659         }
 660
 661 out:
 662         ret = flush_pending(&metadump, 1);
 663         if (ret) {
 664                 if (!err)
 665                         ret = err;
 666                 fprintf(stderr, "Error flushing pending %d\n", ret);
 667         }
 668
 669         metadump_destroy(&metadump);
 670
 671         btrfs_free_path(path);
 672         ret = close_ctree(root);
 673         return err ? err : ret;
 674 }
 675
 676 static void update_super(u8 *buffer)
 677 {
 678         struct btrfs_super_block *super = (struct btrfs_super_block *)buffer;
 679         struct btrfs_chunk *chunk;
 680         struct btrfs_disk_key *key;
 681         u32 sectorsize = btrfs_super_sectorsize(super);
 682         u64 flags = btrfs_super_flags(super);
 683
 684         flags |= BTRFS_SUPER_FLAG_METADUMP;
 685         btrfs_set_super_flags(super, flags);
 686
 687         key = (struct btrfs_disk_key *)(super->sys_chunk_array);
 688         chunk = (struct btrfs_chunk *)(super->sys_chunk_array +
 689                                        sizeof(struct btrfs_disk_key));
 690
 691         btrfs_set_disk_key_objectid(key, BTRFS_FIRST_CHUNK_TREE_OBJECTID);
 692         btrfs_set_disk_key_type(key, BTRFS_CHUNK_ITEM_KEY);
 693         btrfs_set_disk_key_offset(key, 0);
 694
 695         btrfs_set_stack_chunk_length(chunk, (u64)-1);
 696         btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
 697         btrfs_set_stack_chunk_stripe_len(chunk, 64 * 1024);
 698         btrfs_set_stack_chunk_type(chunk, BTRFS_BLOCK_GROUP_SYSTEM);
 699         btrfs_set_stack_chunk_io_align(chunk, sectorsize);
 700         btrfs_set_stack_chunk_io_width(chunk, sectorsize);
 701         btrfs_set_stack_chunk_sector_size(chunk, sectorsize);
 702         btrfs_set_stack_chunk_num_stripes(chunk, 1);
 703         btrfs_set_stack_chunk_sub_stripes(chunk, 0);
 704         chunk->stripe.devid = super->dev_item.devid;
 705         chunk->stripe.offset = cpu_to_le64(0);
 706         memcpy(chunk->stripe.dev_uuid, super->dev_item.uuid, BTRFS_UUID_SIZE);
 707         btrfs_set_super_sys_array_size(super, sizeof(*key) + sizeof(*chunk));
 708         csum_block(buffer, 4096);
 709 }
 710
 711 static void *restore_worker(void *data)
 712 {
 713         struct mdrestore_struct *mdres = (struct mdrestore_struct *)data;
 714         struct async_work *async;
 715         size_t size;
 716         u8 *buffer;
 717         u8 *outbuf;
 718         int outfd;
 719         int ret;
 720
 721         outfd = fileno(mdres->out);
 722         buffer = malloc(MAX_PENDING_SIZE * 2);
 723         if (!buffer) {
 724                 fprintf(stderr, "Error allocing buffer\n");
 725                 pthread_mutex_lock(&mdres->mutex);
 726                 if (!mdres->error)
 727                         mdres->error = -ENOMEM;
 728                 pthread_mutex_unlock(&mdres->mutex);
 729                 goto out;
 730         }
 731
 732         while (1) {
 733                 int err = 0;
 734
 735                 pthread_mutex_lock(&mdres->mutex);
 736                 while (list_empty(&mdres->list)) {
 737                         if (mdres->done) {
 738                                 pthread_mutex_unlock(&mdres->mutex);
 739                                 goto out;
 740                         }
 741                         pthread_cond_wait(&mdres->cond, &mdres->mutex);
 742                 }
 743                 async = list_entry(mdres->list.next, struct async_work, list);
 744                 list_del_init(&async->list);
 745                 pthread_mutex_unlock(&mdres->mutex);
 746
 747                 if (mdres->compress_method == COMPRESS_ZLIB) {
 748                         size = MAX_PENDING_SIZE * 2;
 749                         ret = uncompress(buffer, (unsigned long *)&size,
 750                                          async->buffer, async->bufsize);
 751                         if (ret != Z_OK) {
 752                                 fprintf(stderr, "Error decompressing %d\n",
 753                                         ret);
 754                                 err = -EIO;
 755                         }
 756                         outbuf = buffer;
 757                 } else {
 758                         outbuf = async->buffer;
 759                         size = async->bufsize;
 760                 }
 761
 762                 if (async->start == BTRFS_SUPER_INFO_OFFSET)
 763                         update_super(outbuf);
 764
 765                 ret = pwrite64(outfd, outbuf, size, async->start);
 766                 if (ret < size) {
 767                         if (ret < 0) {
 768                                 fprintf(stderr, "Error writing to device %d\n",
 769                                         errno);
 770                                 err = errno;
 771                         } else {
 772                                 fprintf(stderr, "Short write\n");
 773                                 err = -EIO;
 774                         }
 775                 }
 776
 777                 pthread_mutex_lock(&mdres->mutex);
 778                 if (err && !mdres->error)
 779                         mdres->error = err;
 780                 mdres->num_items--;
 781                 pthread_mutex_unlock(&mdres->mutex);
 782
 783                 free(async->buffer);
 784                 free(async);
 785         }
 786 out:
 787         free(buffer);
 788         pthread_exit(NULL);
 789 }
 790
 791 static void mdrestore_destroy(struct mdrestore_struct *mdres)
 792 {
 793         int i;
 794         pthread_mutex_lock(&mdres->mutex);
 795         mdres->done = 1;
 796         pthread_cond_broadcast(&mdres->cond);
 797         pthread_mutex_unlock(&mdres->mutex);
 798
 799         for (i = 0; i < mdres->num_threads; i++)
 800                 pthread_join(mdres->threads[i], NULL);
 801
 802         pthread_cond_destroy(&mdres->cond);
 803         pthread_mutex_destroy(&mdres->mutex);
 804         free(mdres->threads);
 805 }
 806
 807 static int mdrestore_init(struct mdrestore_struct *mdres,
 808                           FILE *in, FILE *out, int num_threads)
 809 {
 810         int i, ret = 0;
 811
 812         memset(mdres, 0, sizeof(*mdres));
 813         pthread_cond_init(&mdres->cond, NULL);
 814         pthread_mutex_init(&mdres->mutex, NULL);
 815         INIT_LIST_HEAD(&mdres->list);
 816         mdres->in = in;
 817         mdres->out = out;
 818
 819         if (!num_threads)
 820                 return 0;
 821
 822         mdres->num_threads = num_threads;
 823         mdres->threads = calloc(num_threads, sizeof(pthread_t));
 824         if (!mdres->threads)
 825                 return -ENOMEM;
 826         for (i = 0; i < num_threads; i++) {
 827                 ret = pthread_create(mdres->threads + i, NULL, restore_worker,
 828                                      mdres);
 829                 if (ret)
 830                         break;
 831         }
 832         if (ret)
 833                 mdrestore_destroy(mdres);
 834         return ret;
 835 }
 836
 837 static int add_cluster(struct meta_cluster *cluster,
 838                        struct mdrestore_struct *mdres, u64 *next)
 839 {
 840         struct meta_cluster_item *item;
 841         struct meta_cluster_header *header = &cluster->header;
 842         struct async_work *async;
 843         u64 bytenr;
 844         u32 i, nritems;
 845         int ret;
 846
 847         BUG_ON(mdres->num_items);
 848         mdres->compress_method = header->compress;
 849
 850         bytenr = le64_to_cpu(header->bytenr) + BLOCK_SIZE;
 851         nritems = le32_to_cpu(header->nritems);
 852         for (i = 0; i < nritems; i++) {
 853                 item = &cluster->items[i];
 854                 async = calloc(1, sizeof(*async));
 855                 if (!async) {
 856                         fprintf(stderr, "Error allocating async\n");
 857                         return -ENOMEM;
 858                 }
 859                 async->start = le64_to_cpu(item->bytenr);
 860                 async->bufsize = le32_to_cpu(item->size);
 861                 async->buffer = malloc(async->bufsize);
 862                 if (!async->buffer) {
 863                         fprintf(stderr, "Error allocing async buffer\n");
 864                         free(async);
 865                         return -ENOMEM;
 866                 }
 867                 ret = fread(async->buffer, async->bufsize, 1, mdres->in);
 868                 if (ret != 1) {
 869                         fprintf(stderr, "Error reading buffer %d\n", errno);
 870                         free(async->buffer);
 871                         free(async);
 872                         return -EIO;
 873                 }
 874                 bytenr += async->bufsize;
 875
 876                 pthread_mutex_lock(&mdres->mutex);
 877                 list_add_tail(&async->list, &mdres->list);
 878                 mdres->num_items++;
 879                 pthread_cond_signal(&mdres->cond);
 880                 pthread_mutex_unlock(&mdres->mutex);
 881         }
 882         if (bytenr & BLOCK_MASK) {
 883                 char buffer[BLOCK_MASK];
 884                 size_t size = BLOCK_SIZE - (bytenr & BLOCK_MASK);
 885
 886                 bytenr += size;
 887                 ret = fread(buffer, size, 1, mdres->in);
 888                 if (ret != 1) {
 889                         fprintf(stderr, "Error reading in buffer %d\n", errno);
 890                         return -EIO;
 891                 }
 892         }
 893         *next = bytenr;
 894         return 0;
 895 }
 896
 897 static int wait_for_worker(struct mdrestore_struct *mdres)
 898 {
 899         int ret = 0;
 900
 901         pthread_mutex_lock(&mdres->mutex);
 902         ret = mdres->error;
 903         while (!ret && mdres->num_items > 0) {
 904                 struct timespec ts = {
 905                         .tv_sec = 0,
 906                         .tv_nsec = 10000000,
 907                 };
 908                 pthread_mutex_unlock(&mdres->mutex);
 909                 nanosleep(&ts, NULL);
 910                 pthread_mutex_lock(&mdres->mutex);
 911                 ret = mdres->error;
 912         }
 913         pthread_mutex_unlock(&mdres->mutex);
 914         return ret;
 915 }
 916
 917 static int restore_metadump(const char *input, FILE *out, int num_threads)
 918 {
 919         struct meta_cluster *cluster = NULL;
 920         struct meta_cluster_header *header;
 921         struct mdrestore_struct mdrestore;
 922         u64 bytenr = 0;
 923         FILE *in = NULL;
 924         int ret = 0;
 925
 926         if (!strcmp(input, "-")) {
 927                 in = stdin;
 928         } else {
 929                 in = fopen(input, "r");
 930                 if (!in) {
 931                         perror("unable to open metadump image");
 932                         return 1;
 933                 }
 934         }
 935
 936         cluster = malloc(BLOCK_SIZE);
 937         if (!cluster) {
 938                 fprintf(stderr, "Error allocating cluster\n");
 939                 if (in != stdin)
 940                         fclose(in);
 941                 return -ENOMEM;
 942         }
 943
 944         ret = mdrestore_init(&mdrestore, in, out, num_threads);
 945         if (ret) {
 946                 fprintf(stderr, "Error initing mdrestore %d\n", ret);
 947                 if (in != stdin)
 948                         fclose(in);
 949                 free(cluster);
 950                 return ret;
 951         }
 952
 953         while (1) {
 954                 ret = fread(cluster, BLOCK_SIZE, 1, in);
 955                 if (!ret)
 956                         break;
 957
 958                 header = &cluster->header;
 959                 if (le64_to_cpu(header->magic) != HEADER_MAGIC ||
 960                     le64_to_cpu(header->bytenr) != bytenr) {
 961                         fprintf(stderr, "bad header in metadump image\n");
 962                         ret = -EIO;
 963                         break;
 964                 }
 965                 ret = add_cluster(cluster, &mdrestore, &bytenr);
 966                 if (ret) {
 967                         fprintf(stderr, "Error adding cluster\n");
 968                         break;
 969                 }
 970
 971                 ret = wait_for_worker(&mdrestore);
 972                 if (ret) {
 973                         fprintf(stderr, "One of the threads errored out %d\n",
 974                                 ret);
 975                         break;
 976                 }
 977         }
 978
 979         mdrestore_destroy(&mdrestore);
 980         free(cluster);
 981         if (in != stdin)
 982                 fclose(in);
 983         return ret;
 984 }
 985
 986 static void print_usage(void)
 987 {
 988         fprintf(stderr, "usage: btrfs-image [options] source target\n");
 989         fprintf(stderr, "\t-r      \trestore metadump image\n");
 990         fprintf(stderr, "\t-c value\tcompression level (0 ~ 9)\n");
 991         fprintf(stderr, "\t-t value\tnumber of threads (1 ~ 32)\n");
 992         exit(1);
 993 }
 994
 995 int main(int argc, char *argv[])
 996 {
 997         char *source;
 998         char *target;
 999         int num_threads = 0;
1000         int compress_level = 0;
1001         int create = 1;
1002         int ret;
1003         FILE *out;
1004
1005         while (1) {
1006                 int c = getopt(argc, argv, "rc:t:");
1007                 if (c < 0)
1008                         break;
1009                 switch (c) {
1010                 case 'r':
1011                         create = 0;
1012                         break;
1013                 case 't':
1014                         num_threads = atoi(optarg);
1015                         if (num_threads <= 0 || num_threads > 32)
1016                                 print_usage();
1017                         break;
1018                 case 'c':
1019                         compress_level = atoi(optarg);
1020                         if (compress_level < 0 || compress_level > 9)
1021                                 print_usage();
1022                         break;
1023                 default:
1024                         print_usage();
1025                 }
1026         }
1027
1028         argc = argc - optind;
1029         if (argc != 2)
1030                 print_usage();
1031         source = argv[optind];
1032         target = argv[optind + 1];
1033
1034         if (create && !strcmp(target, "-")) {
1035                 out = stdout;
1036         } else {
1037                 out = fopen(target, "w+");
1038                 if (!out) {
1039                         perror("unable to create target file");
1040                         exit(1);
1041                 }
1042         }
1043
1044         if (num_threads == 0 && compress_level > 0) {
1045                 num_threads = sysconf(_SC_NPROCESSORS_ONLN);
1046                 if (num_threads <= 0)
1047                         num_threads = 1;
1048         }
1049
1050         if (create)
1051                 ret = create_metadump(source, out, num_threads,
1052                                       compress_level);
1053         else
1054                 ret = restore_metadump(source, out, 1);
1055
1056         if (out == stdout)
1057                 fflush(out);
1058         else
1059                 fclose(out);
1060
1061         return ret;
1062 }