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