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