mkfs/main.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include "kerncompat.h"
  20 #include "androidcompat.h"
  21
  22 #include <sys/ioctl.h>
  23 #include <sys/mount.h>
  24 #include "ioctl.h"
  25 #include <stdio.h>
  26 #include <stdlib.h>
  27 #include <sys/types.h>
  28 #include <sys/stat.h>
  29 /* #include <sys/dir.h> included via androidcompat.h */
  30 #include <fcntl.h>
  31 #include <unistd.h>
  32 #include <getopt.h>
  33 #include <uuid/uuid.h>
  34 #include <ctype.h>
  35 #include <sys/xattr.h>
  36 #include <limits.h>
  37 #include <linux/limits.h>
  38 #include <blkid/blkid.h>
  39 #include <ftw.h>
  40 #include "ctree.h"
  41 #include "disk-io.h"
  42 #include "volumes.h"
  43 #include "transaction.h"
  44 #include "utils.h"
  45 #include "list_sort.h"
  46 #include "help.h"
  47 #include "mkfs/common.h"
  48 #include "fsfeatures.h"
  49
  50 static u64 index_cnt = 2;
  51 static int verbose = 1;
  52
  53 struct directory_name_entry {
  54         const char *dir_name;
  55         const char *path;
  56         ino_t inum;
  57         struct list_head list;
  58 };
  59
  60 struct mkfs_allocation {
  61         u64 data;
  62         u64 metadata;
  63         u64 mixed;
  64         u64 system;
  65 };
  66
  67 static int create_metadata_block_groups(struct btrfs_root *root, int mixed,
  68                                 struct mkfs_allocation *allocation)
  69 {
  70         struct btrfs_fs_info *fs_info = root->fs_info;
  71         struct btrfs_trans_handle *trans;
  72         u64 bytes_used;
  73         u64 chunk_start = 0;
  74         u64 chunk_size = 0;
  75         int ret;
  76
  77         trans = btrfs_start_transaction(root, 1);
  78         BUG_ON(IS_ERR(trans));
  79         bytes_used = btrfs_super_bytes_used(fs_info->super_copy);
  80
  81         root->fs_info->system_allocs = 1;
  82         ret = btrfs_make_block_group(trans, fs_info, bytes_used,
  83                                      BTRFS_BLOCK_GROUP_SYSTEM,
  84                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
  85                                      0, BTRFS_MKFS_SYSTEM_GROUP_SIZE);
  86         allocation->system += BTRFS_MKFS_SYSTEM_GROUP_SIZE;
  87         if (ret)
  88                 return ret;
  89
  90         if (mixed) {
  91                 ret = btrfs_alloc_chunk(trans, fs_info,
  92                                         &chunk_start, &chunk_size,
  93                                         BTRFS_BLOCK_GROUP_METADATA |
  94                                         BTRFS_BLOCK_GROUP_DATA);
  95                 if (ret == -ENOSPC) {
  96                         error("no space to allocate data/metadata chunk");
  97                         goto err;
  98                 }
  99                 if (ret)
 100                         return ret;
 101                 ret = btrfs_make_block_group(trans, fs_info, 0,
 102                                              BTRFS_BLOCK_GROUP_METADATA |
 103                                              BTRFS_BLOCK_GROUP_DATA,
 104                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 105                                              chunk_start, chunk_size);
 106                 if (ret)
 107                         return ret;
 108                 allocation->mixed += chunk_size;
 109         } else {
 110                 ret = btrfs_alloc_chunk(trans, fs_info,
 111                                         &chunk_start, &chunk_size,
 112                                         BTRFS_BLOCK_GROUP_METADATA);
 113                 if (ret == -ENOSPC) {
 114                         error("no space to allocate metadata chunk");
 115                         goto err;
 116                 }
 117                 if (ret)
 118                         return ret;
 119                 ret = btrfs_make_block_group(trans, fs_info, 0,
 120                                              BTRFS_BLOCK_GROUP_METADATA,
 121                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 122                                              chunk_start, chunk_size);
 123                 allocation->metadata += chunk_size;
 124                 if (ret)
 125                         return ret;
 126         }
 127
 128         root->fs_info->system_allocs = 0;
 129         ret = btrfs_commit_transaction(trans, root);
 130
 131 err:
 132         return ret;
 133 }
 134
 135 static int create_data_block_groups(struct btrfs_trans_handle *trans,
 136                 struct btrfs_root *root, int mixed,
 137                 struct mkfs_allocation *allocation)
 138 {
 139         struct btrfs_fs_info *fs_info = root->fs_info;
 140         u64 chunk_start = 0;
 141         u64 chunk_size = 0;
 142         int ret = 0;
 143
 144         if (!mixed) {
 145                 ret = btrfs_alloc_chunk(trans, fs_info,
 146                                         &chunk_start, &chunk_size,
 147                                         BTRFS_BLOCK_GROUP_DATA);
 148                 if (ret == -ENOSPC) {
 149                         error("no space to allocate data chunk");
 150                         goto err;
 151                 }
 152                 if (ret)
 153                         return ret;
 154                 ret = btrfs_make_block_group(trans, fs_info, 0,
 155                                              BTRFS_BLOCK_GROUP_DATA,
 156                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 157                                              chunk_start, chunk_size);
 158                 allocation->data += chunk_size;
 159                 if (ret)
 160                         return ret;
 161         }
 162
 163 err:
 164         return ret;
 165 }
 166
 167 static int make_root_dir(struct btrfs_trans_handle *trans,
 168                 struct btrfs_root *root)
 169 {
 170         struct btrfs_key location;
 171         int ret;
 172
 173         ret = btrfs_make_root_dir(trans, root->fs_info->tree_root,
 174                               BTRFS_ROOT_TREE_DIR_OBJECTID);
 175         if (ret)
 176                 goto err;
 177         ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
 178         if (ret)
 179                 goto err;
 180         memcpy(&location, &root->fs_info->fs_root->root_key, sizeof(location));
 181         location.offset = (u64)-1;
 182         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
 183                         "default", 7,
 184                         btrfs_super_root_dir(root->fs_info->super_copy),
 185                         &location, BTRFS_FT_DIR, 0);
 186         if (ret)
 187                 goto err;
 188
 189         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 190                              "default", 7, location.objectid,
 191                              BTRFS_ROOT_TREE_DIR_OBJECTID, 0);
 192         if (ret)
 193                 goto err;
 194
 195 err:
 196         return ret;
 197 }
 198
 199 static int __recow_root(struct btrfs_trans_handle *trans,
 200                          struct btrfs_root *root)
 201 {
 202         struct extent_buffer *tmp;
 203         int ret;
 204
 205         if (trans->transid != btrfs_root_generation(&root->root_item)) {
 206                 extent_buffer_get(root->node);
 207                 ret = __btrfs_cow_block(trans, root, root->node,
 208                                         NULL, 0, &tmp, 0, 0);
 209                 if (ret)
 210                         return ret;
 211                 free_extent_buffer(tmp);
 212         }
 213
 214         return 0;
 215 }
 216
 217 static int recow_roots(struct btrfs_trans_handle *trans,
 218                        struct btrfs_root *root)
 219 {
 220         struct btrfs_fs_info *info = root->fs_info;
 221         int ret;
 222
 223         ret = __recow_root(trans, info->fs_root);
 224         if (ret)
 225                 return ret;
 226         ret = __recow_root(trans, info->tree_root);
 227         if (ret)
 228                 return ret;
 229         ret = __recow_root(trans, info->extent_root);
 230         if (ret)
 231                 return ret;
 232         ret = __recow_root(trans, info->chunk_root);
 233         if (ret)
 234                 return ret;
 235         ret = __recow_root(trans, info->dev_root);
 236         if (ret)
 237                 return ret;
 238         ret = __recow_root(trans, info->csum_root);
 239         if (ret)
 240                 return ret;
 241
 242         return 0;
 243 }
 244
 245 static int create_one_raid_group(struct btrfs_trans_handle *trans,
 246                               struct btrfs_root *root, u64 type,
 247                               struct mkfs_allocation *allocation)
 248
 249 {
 250         struct btrfs_fs_info *fs_info = root->fs_info;
 251         u64 chunk_start;
 252         u64 chunk_size;
 253         int ret;
 254
 255         ret = btrfs_alloc_chunk(trans, fs_info,
 256                                 &chunk_start, &chunk_size, type);
 257         if (ret == -ENOSPC) {
 258                 error("not enough free space to allocate chunk");
 259                 exit(1);
 260         }
 261         if (ret)
 262                 return ret;
 263
 264         ret = btrfs_make_block_group(trans, fs_info, 0,
 265                                      type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 266                                      chunk_start, chunk_size);
 267
 268         type &= BTRFS_BLOCK_GROUP_TYPE_MASK;
 269         if (type == BTRFS_BLOCK_GROUP_DATA) {
 270                 allocation->data += chunk_size;
 271         } else if (type == BTRFS_BLOCK_GROUP_METADATA) {
 272                 allocation->metadata += chunk_size;
 273         } else if (type == BTRFS_BLOCK_GROUP_SYSTEM) {
 274                 allocation->system += chunk_size;
 275         } else if (type ==
 276                         (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA)) {
 277                 allocation->mixed += chunk_size;
 278         } else {
 279                 error("unrecognized profile type: 0x%llx",
 280                                 (unsigned long long)type);
 281                 ret = -EINVAL;
 282         }
 283
 284         return ret;
 285 }
 286
 287 static int create_raid_groups(struct btrfs_trans_handle *trans,
 288                               struct btrfs_root *root, u64 data_profile,
 289                               u64 metadata_profile, int mixed,
 290                               struct mkfs_allocation *allocation)
 291 {
 292         int ret;
 293
 294         if (metadata_profile) {
 295                 u64 meta_flags = BTRFS_BLOCK_GROUP_METADATA;
 296
 297                 ret = create_one_raid_group(trans, root,
 298                                             BTRFS_BLOCK_GROUP_SYSTEM |
 299                                             metadata_profile, allocation);
 300                 if (ret)
 301                         return ret;
 302
 303                 if (mixed)
 304                         meta_flags |= BTRFS_BLOCK_GROUP_DATA;
 305
 306                 ret = create_one_raid_group(trans, root, meta_flags |
 307                                             metadata_profile, allocation);
 308                 if (ret)
 309                         return ret;
 310
 311         }
 312         if (!mixed && data_profile) {
 313                 ret = create_one_raid_group(trans, root,
 314                                             BTRFS_BLOCK_GROUP_DATA |
 315                                             data_profile, allocation);
 316                 if (ret)
 317                         return ret;
 318         }
 319         ret = recow_roots(trans, root);
 320
 321         return ret;
 322 }
 323
 324 static int create_data_reloc_tree(struct btrfs_trans_handle *trans,
 325                                   struct btrfs_root *root)
 326 {
 327         struct btrfs_key location;
 328         struct btrfs_root_item root_item;
 329         struct extent_buffer *tmp;
 330         u64 objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
 331         int ret;
 332
 333         ret = btrfs_copy_root(trans, root, root->node, &tmp, objectid);
 334         if (ret)
 335                 return ret;
 336
 337         memcpy(&root_item, &root->root_item, sizeof(root_item));
 338         btrfs_set_root_bytenr(&root_item, tmp->start);
 339         btrfs_set_root_level(&root_item, btrfs_header_level(tmp));
 340         btrfs_set_root_generation(&root_item, trans->transid);
 341         free_extent_buffer(tmp);
 342
 343         location.objectid = objectid;
 344         location.type = BTRFS_ROOT_ITEM_KEY;
 345         location.offset = 0;
 346         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
 347                                 &location, &root_item);
 348
 349         return ret;
 350 }
 351
 352 static void print_usage(int ret)
 353 {
 354         printf("Usage: mkfs.btrfs [options] dev [ dev ... ]\n");
 355         printf("Options:\n");
 356         printf("  allocation profiles:\n");
 357         printf("\t-d|--data PROFILE       data profile, raid0, raid1, raid5, raid6, raid10, dup or single\n");
 358         printf("\t-m|--metadata PROFILE   metadata profile, values like for data profile\n");
 359         printf("\t-M|--mixed              mix metadata and data together\n");
 360         printf("  features:\n");
 361         printf("\t-n|--nodesize SIZE      size of btree nodes\n");
 362         printf("\t-s|--sectorsize SIZE    data block size (may not be mountable by current kernel)\n");
 363         printf("\t-O|--features LIST      comma separated list of filesystem features (use '-O list-all' to list features)\n");
 364         printf("\t-L|--label LABEL        set the filesystem label\n");
 365         printf("\t-U|--uuid UUID          specify the filesystem UUID (must be unique)\n");
 366         printf("  creation:\n");
 367         printf("\t-b|--byte-count SIZE    set filesystem size to SIZE (on the first device)\n");
 368         printf("\t-r|--rootdir DIR        copy files from DIR to the image root directory\n");
 369         printf("\t-K|--nodiscard          do not perform whole device TRIM\n");
 370         printf("\t-f|--force              force overwrite of existing filesystem\n");
 371         printf("  general:\n");
 372         printf("\t-q|--quiet              no messages except errors\n");
 373         printf("\t-V|--version            print the mkfs.btrfs version and exit\n");
 374         printf("\t--help                  print this help and exit\n");
 375         printf("  deprecated:\n");
 376         printf("\t-A|--alloc-start START  the offset to start the filesystem\n");
 377         printf("\t-l|--leafsize SIZE      deprecated, alias for nodesize\n");
 378         exit(ret);
 379 }
 380
 381 static u64 parse_profile(const char *s)
 382 {
 383         if (strcasecmp(s, "raid0") == 0) {
 384                 return BTRFS_BLOCK_GROUP_RAID0;
 385         } else if (strcasecmp(s, "raid1") == 0) {
 386                 return BTRFS_BLOCK_GROUP_RAID1;
 387         } else if (strcasecmp(s, "raid5") == 0) {
 388                 return BTRFS_BLOCK_GROUP_RAID5;
 389         } else if (strcasecmp(s, "raid6") == 0) {
 390                 return BTRFS_BLOCK_GROUP_RAID6;
 391         } else if (strcasecmp(s, "raid10") == 0) {
 392                 return BTRFS_BLOCK_GROUP_RAID10;
 393         } else if (strcasecmp(s, "dup") == 0) {
 394                 return BTRFS_BLOCK_GROUP_DUP;
 395         } else if (strcasecmp(s, "single") == 0) {
 396                 return 0;
 397         } else {
 398                 error("unknown profile %s", s);
 399                 exit(1);
 400         }
 401         /* not reached */
 402         return 0;
 403 }
 404
 405 static char *parse_label(const char *input)
 406 {
 407         int len = strlen(input);
 408
 409         if (len >= BTRFS_LABEL_SIZE) {
 410                 error("label %s is too long (max %d)", input,
 411                         BTRFS_LABEL_SIZE - 1);
 412                 exit(1);
 413         }
 414         return strdup(input);
 415 }
 416
 417 static int add_directory_items(struct btrfs_trans_handle *trans,
 418                                struct btrfs_root *root, u64 objectid,
 419                                ino_t parent_inum, const char *name,
 420                                struct stat *st, int *dir_index_cnt)
 421 {
 422         int ret;
 423         int name_len;
 424         struct btrfs_key location;
 425         u8 filetype = 0;
 426
 427         name_len = strlen(name);
 428
 429         location.objectid = objectid;
 430         location.offset = 0;
 431         location.type = BTRFS_INODE_ITEM_KEY;
 432
 433         if (S_ISDIR(st->st_mode))
 434                 filetype = BTRFS_FT_DIR;
 435         if (S_ISREG(st->st_mode))
 436                 filetype = BTRFS_FT_REG_FILE;
 437         if (S_ISLNK(st->st_mode))
 438                 filetype = BTRFS_FT_SYMLINK;
 439
 440         ret = btrfs_insert_dir_item(trans, root, name, name_len,
 441                                     parent_inum, &location,
 442                                     filetype, index_cnt);
 443         if (ret)
 444                 return ret;
 445         ret = btrfs_insert_inode_ref(trans, root, name, name_len,
 446                                      objectid, parent_inum, index_cnt);
 447         *dir_index_cnt = index_cnt;
 448         index_cnt++;
 449
 450         return ret;
 451 }
 452
 453 static int fill_inode_item(struct btrfs_trans_handle *trans,
 454                            struct btrfs_root *root,
 455                            struct btrfs_inode_item *dst, struct stat *src)
 456 {
 457         u64 blocks = 0;
 458         u64 sectorsize = root->fs_info->sectorsize;
 459
 460         /*
 461          * btrfs_inode_item has some reserved fields
 462          * and represents on-disk inode entry, so
 463          * zero everything to prevent information leak
 464          */
 465         memset(dst, 0, sizeof (*dst));
 466
 467         btrfs_set_stack_inode_generation(dst, trans->transid);
 468         btrfs_set_stack_inode_size(dst, src->st_size);
 469         btrfs_set_stack_inode_nbytes(dst, 0);
 470         btrfs_set_stack_inode_block_group(dst, 0);
 471         btrfs_set_stack_inode_nlink(dst, src->st_nlink);
 472         btrfs_set_stack_inode_uid(dst, src->st_uid);
 473         btrfs_set_stack_inode_gid(dst, src->st_gid);
 474         btrfs_set_stack_inode_mode(dst, src->st_mode);
 475         btrfs_set_stack_inode_rdev(dst, 0);
 476         btrfs_set_stack_inode_flags(dst, 0);
 477         btrfs_set_stack_timespec_sec(&dst->atime, src->st_atime);
 478         btrfs_set_stack_timespec_nsec(&dst->atime, 0);
 479         btrfs_set_stack_timespec_sec(&dst->ctime, src->st_ctime);
 480         btrfs_set_stack_timespec_nsec(&dst->ctime, 0);
 481         btrfs_set_stack_timespec_sec(&dst->mtime, src->st_mtime);
 482         btrfs_set_stack_timespec_nsec(&dst->mtime, 0);
 483         btrfs_set_stack_timespec_sec(&dst->otime, 0);
 484         btrfs_set_stack_timespec_nsec(&dst->otime, 0);
 485
 486         if (S_ISDIR(src->st_mode)) {
 487                 btrfs_set_stack_inode_size(dst, 0);
 488                 btrfs_set_stack_inode_nlink(dst, 1);
 489         }
 490         if (S_ISREG(src->st_mode)) {
 491                 btrfs_set_stack_inode_size(dst, (u64)src->st_size);
 492                 if (src->st_size <= BTRFS_MAX_INLINE_DATA_SIZE(root))
 493                         btrfs_set_stack_inode_nbytes(dst, src->st_size);
 494                 else {
 495                         blocks = src->st_size / sectorsize;
 496                         if (src->st_size % sectorsize)
 497                                 blocks += 1;
 498                         blocks *= sectorsize;
 499                         btrfs_set_stack_inode_nbytes(dst, blocks);
 500                 }
 501         }
 502         if (S_ISLNK(src->st_mode))
 503                 btrfs_set_stack_inode_nbytes(dst, src->st_size + 1);
 504
 505         return 0;
 506 }
 507
 508 static int directory_select(const struct direct *entry)
 509 {
 510         if (entry->d_name[0] == '.' &&
 511                 (entry->d_name[1] == 0 ||
 512                  (entry->d_name[1] == '.' && entry->d_name[2] == 0)))
 513                 return 0;
 514         return 1;
 515 }
 516
 517 static void free_namelist(struct direct **files, int count)
 518 {
 519         int i;
 520
 521         if (count < 0)
 522                 return;
 523
 524         for (i = 0; i < count; ++i)
 525                 free(files[i]);
 526         free(files);
 527 }
 528
 529 static u64 calculate_dir_inode_size(const char *dirname)
 530 {
 531         int count, i;
 532         struct direct **files, *cur_file;
 533         u64 dir_inode_size = 0;
 534
 535         count = scandir(dirname, &files, directory_select, NULL);
 536
 537         for (i = 0; i < count; i++) {
 538                 cur_file = files[i];
 539                 dir_inode_size += strlen(cur_file->d_name);
 540         }
 541
 542         free_namelist(files, count);
 543
 544         dir_inode_size *= 2;
 545         return dir_inode_size;
 546 }
 547
 548 static int add_inode_items(struct btrfs_trans_handle *trans,
 549                            struct btrfs_root *root,
 550                            struct stat *st, const char *name,
 551                            u64 self_objectid,
 552                            struct btrfs_inode_item *inode_ret)
 553 {
 554         int ret;
 555         struct btrfs_inode_item btrfs_inode;
 556         u64 objectid;
 557         u64 inode_size = 0;
 558
 559         fill_inode_item(trans, root, &btrfs_inode, st);
 560         objectid = self_objectid;
 561
 562         if (S_ISDIR(st->st_mode)) {
 563                 inode_size = calculate_dir_inode_size(name);
 564                 btrfs_set_stack_inode_size(&btrfs_inode, inode_size);
 565         }
 566
 567         ret = btrfs_insert_inode(trans, root, objectid, &btrfs_inode);
 568
 569         *inode_ret = btrfs_inode;
 570         return ret;
 571 }
 572
 573 static int add_xattr_item(struct btrfs_trans_handle *trans,
 574                           struct btrfs_root *root, u64 objectid,
 575                           const char *file_name)
 576 {
 577         int ret;
 578         int cur_name_len;
 579         char xattr_list[XATTR_LIST_MAX];
 580         char *cur_name;
 581         char cur_value[XATTR_SIZE_MAX];
 582         char delimiter = '\0';
 583         char *next_location = xattr_list;
 584
 585         ret = llistxattr(file_name, xattr_list, XATTR_LIST_MAX);
 586         if (ret < 0) {
 587                 if(errno == ENOTSUP)
 588                         return 0;
 589                 error("getting a list of xattr failed for %s: %s", file_name,
 590                                 strerror(errno));
 591                 return ret;
 592         }
 593         if (ret == 0)
 594                 return ret;
 595
 596         cur_name = strtok(xattr_list, &delimiter);
 597         while (cur_name != NULL) {
 598                 cur_name_len = strlen(cur_name);
 599                 next_location += cur_name_len + 1;
 600
 601                 ret = getxattr(file_name, cur_name, cur_value, XATTR_SIZE_MAX);
 602                 if (ret < 0) {
 603                         if(errno == ENOTSUP)
 604                                 return 0;
 605                         error("gettig a xattr value failed for %s attr %s: %s",
 606                                 file_name, cur_name, strerror(errno));
 607                         return ret;
 608                 }
 609
 610                 ret = btrfs_insert_xattr_item(trans, root, cur_name,
 611                                               cur_name_len, cur_value,
 612                                               ret, objectid);
 613                 if (ret) {
 614                         error("inserting a xattr item failed for %s: %s",
 615                                         file_name, strerror(-ret));
 616                 }
 617
 618                 cur_name = strtok(next_location, &delimiter);
 619         }
 620
 621         return ret;
 622 }
 623
 624 static int add_symbolic_link(struct btrfs_trans_handle *trans,
 625                              struct btrfs_root *root,
 626                              u64 objectid, const char *path_name)
 627 {
 628         int ret;
 629         char buf[PATH_MAX];
 630
 631         ret = readlink(path_name, buf, sizeof(buf));
 632         if (ret <= 0) {
 633                 error("readlink failed for %s: %s", path_name, strerror(errno));
 634                 goto fail;
 635         }
 636         if (ret >= sizeof(buf)) {
 637                 error("symlink too long for %s", path_name);
 638                 ret = -1;
 639                 goto fail;
 640         }
 641
 642         buf[ret] = '\0'; /* readlink does not do it for us */
 643         ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
 644                                          buf, ret + 1);
 645 fail:
 646         return ret;
 647 }
 648
 649 static int add_file_items(struct btrfs_trans_handle *trans,
 650                           struct btrfs_root *root,
 651                           struct btrfs_inode_item *btrfs_inode, u64 objectid,
 652                           struct stat *st, const char *path_name)
 653 {
 654         int ret = -1;
 655         ssize_t ret_read;
 656         u64 bytes_read = 0;
 657         struct btrfs_key key;
 658         int blocks;
 659         u32 sectorsize = root->fs_info->sectorsize;
 660         u64 first_block = 0;
 661         u64 file_pos = 0;
 662         u64 cur_bytes;
 663         u64 total_bytes;
 664         struct extent_buffer *eb = NULL;
 665         int fd;
 666
 667         if (st->st_size == 0)
 668                 return 0;
 669
 670         fd = open(path_name, O_RDONLY);
 671         if (fd == -1) {
 672                 error("cannot open %s: %s", path_name, strerror(errno));
 673                 return ret;
 674         }
 675
 676         blocks = st->st_size / sectorsize;
 677         if (st->st_size % sectorsize)
 678                 blocks += 1;
 679
 680         if (st->st_size <= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 681                 char *buffer = malloc(st->st_size);
 682
 683                 if (!buffer) {
 684                         ret = -ENOMEM;
 685                         goto end;
 686                 }
 687
 688                 ret_read = pread64(fd, buffer, st->st_size, bytes_read);
 689                 if (ret_read == -1) {
 690                         error("cannot read %s at offset %llu length %llu: %s",
 691                                 path_name, (unsigned long long)bytes_read,
 692                                 (unsigned long long)st->st_size,
 693                                 strerror(errno));
 694                         free(buffer);
 695                         goto end;
 696                 }
 697
 698                 ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
 699                                                  buffer, st->st_size);
 700                 free(buffer);
 701                 goto end;
 702         }
 703
 704         /* round up our st_size to the FS blocksize */
 705         total_bytes = (u64)blocks * sectorsize;
 706
 707         /*
 708          * do our IO in extent buffers so it can work
 709          * against any raid type
 710          */
 711         eb = calloc(1, sizeof(*eb) + sectorsize);
 712         if (!eb) {
 713                 ret = -ENOMEM;
 714                 goto end;
 715         }
 716
 717 again:
 718
 719         /*
 720          * keep our extent size at 1MB max, this makes it easier to work inside
 721          * the tiny block groups created during mkfs
 722          */
 723         cur_bytes = min(total_bytes, (u64)SZ_1M);
 724         ret = btrfs_reserve_extent(trans, root, cur_bytes, 0, 0, (u64)-1,
 725                                    &key, 1);
 726         if (ret)
 727                 goto end;
 728
 729         first_block = key.objectid;
 730         bytes_read = 0;
 731
 732         while (bytes_read < cur_bytes) {
 733
 734                 memset(eb->data, 0, sectorsize);
 735
 736                 ret_read = pread64(fd, eb->data, sectorsize, file_pos + bytes_read);
 737                 if (ret_read == -1) {
 738                         error("cannot read %s at offset %llu length %llu: %s",
 739                                 path_name,
 740                                 (unsigned long long)file_pos + bytes_read,
 741                                 (unsigned long long)sectorsize,
 742                                 strerror(errno));
 743                         goto end;
 744                 }
 745
 746                 eb->start = first_block + bytes_read;
 747                 eb->len = sectorsize;
 748
 749                 /*
 750                  * we're doing the csum before we record the extent, but
 751                  * that's ok
 752                  */
 753                 ret = btrfs_csum_file_block(trans, root->fs_info->csum_root,
 754                                             first_block + bytes_read + sectorsize,
 755                                             first_block + bytes_read,
 756                                             eb->data, sectorsize);
 757                 if (ret)
 758                         goto end;
 759
 760                 ret = write_and_map_eb(root->fs_info, eb);
 761                 if (ret) {
 762                         error("failed to write %s", path_name);
 763                         goto end;
 764                 }
 765
 766                 bytes_read += sectorsize;
 767         }
 768
 769         if (bytes_read) {
 770                 ret = btrfs_record_file_extent(trans, root, objectid, btrfs_inode,
 771                                                file_pos, first_block, cur_bytes);
 772                 if (ret)
 773                         goto end;
 774
 775         }
 776
 777         file_pos += cur_bytes;
 778         total_bytes -= cur_bytes;
 779
 780         if (total_bytes)
 781                 goto again;
 782
 783 end:
 784         free(eb);
 785         close(fd);
 786         return ret;
 787 }
 788
 789 static char *make_path(const char *dir, const char *name)
 790 {
 791         char *path;
 792
 793         path = malloc(strlen(dir) + strlen(name) + 2);
 794         if (!path)
 795                 return NULL;
 796         strcpy(path, dir);
 797         if (dir[strlen(dir) - 1] != '/')
 798                 strcat(path, "/");
 799         strcat(path, name);
 800         return path;
 801 }
 802
 803 static int traverse_directory(struct btrfs_trans_handle *trans,
 804                               struct btrfs_root *root, const char *dir_name,
 805                               struct directory_name_entry *dir_head)
 806 {
 807         int ret = 0;
 808
 809         struct btrfs_inode_item cur_inode;
 810         struct btrfs_inode_item *inode_item;
 811         int count, i, dir_index_cnt;
 812         struct direct **files;
 813         struct stat st;
 814         struct directory_name_entry *dir_entry, *parent_dir_entry;
 815         struct direct *cur_file;
 816         ino_t parent_inum, cur_inum;
 817         ino_t highest_inum = 0;
 818         const char *parent_dir_name;
 819         char real_path[PATH_MAX];
 820         struct btrfs_path path;
 821         struct extent_buffer *leaf;
 822         struct btrfs_key root_dir_key;
 823         u64 root_dir_inode_size = 0;
 824
 825         /* Add list for source directory */
 826         dir_entry = malloc(sizeof(struct directory_name_entry));
 827         if (!dir_entry)
 828                 return -ENOMEM;
 829         dir_entry->dir_name = dir_name;
 830         dir_entry->path = realpath(dir_name, real_path);
 831         if (!dir_entry->path) {
 832                 error("realpath  failed for %s: %s", dir_name, strerror(errno));
 833                 ret = -1;
 834                 goto fail_no_dir;
 835         }
 836
 837         parent_inum = highest_inum + BTRFS_FIRST_FREE_OBJECTID;
 838         dir_entry->inum = parent_inum;
 839         list_add_tail(&dir_entry->list, &dir_head->list);
 840
 841         btrfs_init_path(&path);
 842
 843         root_dir_key.objectid = btrfs_root_dirid(&root->root_item);
 844         root_dir_key.offset = 0;
 845         root_dir_key.type = BTRFS_INODE_ITEM_KEY;
 846         ret = btrfs_lookup_inode(trans, root, &path, &root_dir_key, 1);
 847         if (ret) {
 848                 error("failed to lookup root dir: %d", ret);
 849                 goto fail_no_dir;
 850         }
 851
 852         leaf = path.nodes[0];
 853         inode_item = btrfs_item_ptr(leaf, path.slots[0],
 854                                     struct btrfs_inode_item);
 855
 856         root_dir_inode_size = calculate_dir_inode_size(dir_name);
 857         btrfs_set_inode_size(leaf, inode_item, root_dir_inode_size);
 858         btrfs_mark_buffer_dirty(leaf);
 859
 860         btrfs_release_path(&path);
 861
 862         do {
 863                 parent_dir_entry = list_entry(dir_head->list.next,
 864                                               struct directory_name_entry,
 865                                               list);
 866                 list_del(&parent_dir_entry->list);
 867
 868                 parent_inum = parent_dir_entry->inum;
 869                 parent_dir_name = parent_dir_entry->dir_name;
 870                 if (chdir(parent_dir_entry->path)) {
 871                         error("chdir failed for %s: %s",
 872                                 parent_dir_name, strerror(errno));
 873                         ret = -1;
 874                         goto fail_no_files;
 875                 }
 876
 877                 count = scandir(parent_dir_entry->path, &files,
 878                                 directory_select, NULL);
 879                 if (count == -1)
 880                 {
 881                         error("scandir failed for %s: %s",
 882                                 parent_dir_name, strerror (errno));
 883                         ret = -1;
 884                         goto fail;
 885                 }
 886
 887                 for (i = 0; i < count; i++) {
 888                         cur_file = files[i];
 889
 890                         if (lstat(cur_file->d_name, &st) == -1) {
 891                                 error("lstat failed for %s: %s",
 892                                         cur_file->d_name, strerror(errno));
 893                                 ret = -1;
 894                                 goto fail;
 895                         }
 896
 897                         cur_inum = st.st_ino;
 898                         ret = add_directory_items(trans, root,
 899                                                   cur_inum, parent_inum,
 900                                                   cur_file->d_name,
 901                                                   &st, &dir_index_cnt);
 902                         if (ret) {
 903                                 error("unable to add directory items for %s: %d",
 904                                         cur_file->d_name, ret);
 905                                 goto fail;
 906                         }
 907
 908                         ret = add_inode_items(trans, root, &st,
 909                                               cur_file->d_name, cur_inum,
 910                                               &cur_inode);
 911                         if (ret == -EEXIST) {
 912                                 if (st.st_nlink <= 1) {
 913                                         error(
 914                         "item %s already exists but has wrong st_nlink %lu <= 1",
 915                                                 cur_file->d_name,
 916                                                 (unsigned long)st.st_nlink);
 917                                         goto fail;
 918                                 }
 919                                 continue;
 920                         }
 921                         if (ret) {
 922                                 error("unable to add inode items for %s: %d",
 923                                         cur_file->d_name, ret);
 924                                 goto fail;
 925                         }
 926
 927                         ret = add_xattr_item(trans, root,
 928                                              cur_inum, cur_file->d_name);
 929                         if (ret) {
 930                                 error("unable to add xattr items for %s: %d",
 931                                         cur_file->d_name, ret);
 932                                 if(ret != -ENOTSUP)
 933                                         goto fail;
 934                         }
 935
 936                         if (S_ISDIR(st.st_mode)) {
 937                                 dir_entry = malloc(sizeof(struct directory_name_entry));
 938                                 if (!dir_entry) {
 939                                         ret = -ENOMEM;
 940                                         goto fail;
 941                                 }
 942                                 dir_entry->dir_name = cur_file->d_name;
 943                                 dir_entry->path = make_path(parent_dir_entry->path,
 944                                                             cur_file->d_name);
 945                                 dir_entry->inum = cur_inum;
 946                                 list_add_tail(&dir_entry->list, &dir_head->list);
 947                         } else if (S_ISREG(st.st_mode)) {
 948                                 ret = add_file_items(trans, root, &cur_inode,
 949                                                      cur_inum, &st,
 950                                                      cur_file->d_name);
 951                                 if (ret) {
 952                                         error("unable to add file items for %s: %d",
 953                                                 cur_file->d_name, ret);
 954                                         goto fail;
 955                                 }
 956                         } else if (S_ISLNK(st.st_mode)) {
 957                                 ret = add_symbolic_link(trans, root,
 958                                                         cur_inum, cur_file->d_name);
 959                                 if (ret) {
 960                                         error("unable to add symlink for %s: %d",
 961                                                 cur_file->d_name, ret);
 962                                         goto fail;
 963                                 }
 964                         }
 965                 }
 966
 967                 free_namelist(files, count);
 968                 free(parent_dir_entry);
 969
 970                 index_cnt = 2;
 971
 972         } while (!list_empty(&dir_head->list));
 973
 974 out:
 975         return !!ret;
 976 fail:
 977         free_namelist(files, count);
 978 fail_no_files:
 979         free(parent_dir_entry);
 980         goto out;
 981 fail_no_dir:
 982         free(dir_entry);
 983         goto out;
 984 }
 985
 986 static int create_chunks(struct btrfs_trans_handle *trans,
 987                          struct btrfs_root *root, u64 num_of_meta_chunks,
 988                          u64 size_of_data,
 989                          struct mkfs_allocation *allocation)
 990 {
 991         struct btrfs_fs_info *fs_info = root->fs_info;
 992         u64 chunk_start;
 993         u64 chunk_size;
 994         u64 meta_type = BTRFS_BLOCK_GROUP_METADATA;
 995         u64 data_type = BTRFS_BLOCK_GROUP_DATA;
 996         u64 minimum_data_chunk_size = SZ_8M;
 997         u64 i;
 998         int ret;
 999
1000         for (i = 0; i < num_of_meta_chunks; i++) {
1001                 ret = btrfs_alloc_chunk(trans, fs_info,
1002                                         &chunk_start, &chunk_size, meta_type);
1003                 if (ret)
1004                         return ret;
1005                 ret = btrfs_make_block_group(trans, fs_info, 0,
1006                                              meta_type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1007                                              chunk_start, chunk_size);
1008                 allocation->metadata += chunk_size;
1009                 if (ret)
1010                         return ret;
1011                 set_extent_dirty(&root->fs_info->free_space_cache,
1012                                  chunk_start, chunk_start + chunk_size - 1);
1013         }
1014
1015         if (size_of_data < minimum_data_chunk_size)
1016                 size_of_data = minimum_data_chunk_size;
1017
1018         ret = btrfs_alloc_data_chunk(trans, fs_info,
1019                                      &chunk_start, size_of_data, data_type, 0);
1020         if (ret)
1021                 return ret;
1022         ret = btrfs_make_block_group(trans, fs_info, 0,
1023                                      data_type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1024                                      chunk_start, size_of_data);
1025         allocation->data += size_of_data;
1026         if (ret)
1027                 return ret;
1028         set_extent_dirty(&root->fs_info->free_space_cache,
1029                          chunk_start, chunk_start + size_of_data - 1);
1030         return ret;
1031 }
1032
1033 static int make_image(const char *source_dir, struct btrfs_root *root)
1034 {
1035         int ret;
1036         struct btrfs_trans_handle *trans;
1037         struct stat root_st;
1038         struct directory_name_entry dir_head;
1039         struct directory_name_entry *dir_entry = NULL;
1040
1041         ret = lstat(source_dir, &root_st);
1042         if (ret) {
1043                 error("unable to lstat %s: %s", source_dir, strerror(errno));
1044                 ret = -errno;
1045                 goto out;
1046         }
1047
1048         INIT_LIST_HEAD(&dir_head.list);
1049
1050         trans = btrfs_start_transaction(root, 1);
1051         BUG_ON(IS_ERR(trans));
1052         ret = traverse_directory(trans, root, source_dir, &dir_head);
1053         if (ret) {
1054                 error("unable to traverse directory %s: %d", source_dir, ret);
1055                 goto fail;
1056         }
1057         ret = btrfs_commit_transaction(trans, root);
1058         if (ret) {
1059                 error("transaction commit failed: %d", ret);
1060                 goto out;
1061         }
1062
1063         if (verbose)
1064                 printf("Making image is completed.\n");
1065         return 0;
1066 fail:
1067         while (!list_empty(&dir_head.list)) {
1068                 dir_entry = list_entry(dir_head.list.next,
1069                                        struct directory_name_entry, list);
1070                 list_del(&dir_entry->list);
1071                 free(dir_entry);
1072         }
1073 out:
1074         return ret;
1075 }
1076
1077 /*
1078  * This ignores symlinks with unreadable targets and subdirs that can't
1079  * be read.  It's a best-effort to give a rough estimate of the size of
1080  * a subdir.  It doesn't guarantee that prepopulating btrfs from this
1081  * tree won't still run out of space.
1082  */
1083 static u64 global_total_size;
1084 static u64 fs_block_size;
1085 static int ftw_add_entry_size(const char *fpath, const struct stat *st,
1086                               int type)
1087 {
1088         if (type == FTW_F || type == FTW_D)
1089                 global_total_size += round_up(st->st_size, fs_block_size);
1090
1091         return 0;
1092 }
1093
1094 static u64 size_sourcedir(const char *dir_name, u64 sectorsize,
1095                           u64 *num_of_meta_chunks_ret, u64 *size_of_data_ret)
1096 {
1097         u64 dir_size = 0;
1098         u64 total_size = 0;
1099         int ret;
1100         u64 default_chunk_size = SZ_8M;
1101         u64 allocated_meta_size = SZ_8M;
1102         u64 allocated_total_size = 20 * SZ_1M;  /* 20MB */
1103         u64 num_of_meta_chunks = 0;
1104         u64 num_of_data_chunks = 0;
1105         u64 num_of_allocated_meta_chunks =
1106                         allocated_meta_size / default_chunk_size;
1107
1108         global_total_size = 0;
1109         fs_block_size = sectorsize;
1110         ret = ftw(dir_name, ftw_add_entry_size, 10);
1111         dir_size = global_total_size;
1112         if (ret < 0) {
1113                 error("ftw subdir walk of %s failed: %s", dir_name,
1114                         strerror(errno));
1115                 exit(1);
1116         }
1117
1118         num_of_data_chunks = (dir_size + default_chunk_size - 1) /
1119                 default_chunk_size;
1120
1121         num_of_meta_chunks = (dir_size / 2) / default_chunk_size;
1122         if (((dir_size / 2) % default_chunk_size) != 0)
1123                 num_of_meta_chunks++;
1124         if (num_of_meta_chunks <= num_of_allocated_meta_chunks)
1125                 num_of_meta_chunks = 0;
1126         else
1127                 num_of_meta_chunks -= num_of_allocated_meta_chunks;
1128
1129         total_size = allocated_total_size +
1130                      (num_of_data_chunks * default_chunk_size) +
1131                      (num_of_meta_chunks * default_chunk_size);
1132
1133         *num_of_meta_chunks_ret = num_of_meta_chunks;
1134         *size_of_data_ret = num_of_data_chunks * default_chunk_size;
1135         return total_size;
1136 }
1137
1138 static int zero_output_file(int out_fd, u64 size)
1139 {
1140         int loop_num;
1141         u64 location = 0;
1142         char buf[4096];
1143         int ret = 0, i;
1144         ssize_t written;
1145
1146         memset(buf, 0, 4096);
1147         loop_num = size / 4096;
1148         for (i = 0; i < loop_num; i++) {
1149                 written = pwrite64(out_fd, buf, 4096, location);
1150                 if (written != 4096)
1151                         ret = -EIO;
1152                 location += 4096;
1153         }
1154         return ret;
1155 }
1156
1157 static int is_ssd(const char *file)
1158 {
1159         blkid_probe probe;
1160         char wholedisk[PATH_MAX];
1161         char sysfs_path[PATH_MAX];
1162         dev_t devno;
1163         int fd;
1164         char rotational;
1165         int ret;
1166
1167         probe = blkid_new_probe_from_filename(file);
1168         if (!probe)
1169                 return 0;
1170
1171         /* Device number of this disk (possibly a partition) */
1172         devno = blkid_probe_get_devno(probe);
1173         if (!devno) {
1174                 blkid_free_probe(probe);
1175                 return 0;
1176         }
1177
1178         /* Get whole disk name (not full path) for this devno */
1179         ret = blkid_devno_to_wholedisk(devno,
1180                         wholedisk, sizeof(wholedisk), NULL);
1181         if (ret) {
1182                 blkid_free_probe(probe);
1183                 return 0;
1184         }
1185
1186         snprintf(sysfs_path, PATH_MAX, "/sys/block/%s/queue/rotational",
1187                  wholedisk);
1188
1189         blkid_free_probe(probe);
1190
1191         fd = open(sysfs_path, O_RDONLY);
1192         if (fd < 0) {
1193                 return 0;
1194         }
1195
1196         if (read(fd, &rotational, 1) < 1) {
1197                 close(fd);
1198                 return 0;
1199         }
1200         close(fd);
1201
1202         return rotational == '0';
1203 }
1204
1205 static int _cmp_device_by_id(void *priv, struct list_head *a,
1206                              struct list_head *b)
1207 {
1208         return list_entry(a, struct btrfs_device, dev_list)->devid -
1209                list_entry(b, struct btrfs_device, dev_list)->devid;
1210 }
1211
1212 static void list_all_devices(struct btrfs_root *root)
1213 {
1214         struct btrfs_fs_devices *fs_devices;
1215         struct btrfs_device *device;
1216         int number_of_devices = 0;
1217         u64 total_block_count = 0;
1218
1219         fs_devices = root->fs_info->fs_devices;
1220
1221         list_for_each_entry(device, &fs_devices->devices, dev_list)
1222                 number_of_devices++;
1223
1224         list_sort(NULL, &fs_devices->devices, _cmp_device_by_id);
1225
1226         printf("Number of devices:  %d\n", number_of_devices);
1227         /* printf("Total devices size: %10s\n", */
1228                 /* pretty_size(total_block_count)); */
1229         printf("Devices:\n");
1230         printf("   ID        SIZE  PATH\n");
1231         list_for_each_entry(device, &fs_devices->devices, dev_list) {
1232                 printf("  %3llu  %10s  %s\n",
1233                         device->devid,
1234                         pretty_size(device->total_bytes),
1235                         device->name);
1236                 total_block_count += device->total_bytes;
1237         }
1238
1239         printf("\n");
1240 }
1241
1242 static int is_temp_block_group(struct extent_buffer *node,
1243                                struct btrfs_block_group_item *bgi,
1244                                u64 data_profile, u64 meta_profile,
1245                                u64 sys_profile)
1246 {
1247         u64 flag = btrfs_disk_block_group_flags(node, bgi);
1248         u64 flag_type = flag & BTRFS_BLOCK_GROUP_TYPE_MASK;
1249         u64 flag_profile = flag & BTRFS_BLOCK_GROUP_PROFILE_MASK;
1250         u64 used = btrfs_disk_block_group_used(node, bgi);
1251
1252         /*
1253          * Chunks meets all the following conditions is a temp chunk
1254          * 1) Empty chunk
1255          * Temp chunk is always empty.
1256          *
1257          * 2) profile mismatch with mkfs profile.
1258          * Temp chunk is always in SINGLE
1259          *
1260          * 3) Size differs with mkfs_alloc
1261          * Special case for SINGLE/SINGLE btrfs.
1262          * In that case, temp data chunk and real data chunk are always empty.
1263          * So we need to use mkfs_alloc to be sure which chunk is the newly
1264          * allocated.
1265          *
1266          * Normally, new chunk size is equal to mkfs one (One chunk)
1267          * If it has multiple chunks, we just refuse to delete any one.
1268          * As they are all single, so no real problem will happen.
1269          * So only use condition 1) and 2) to judge them.
1270          */
1271         if (used != 0)
1272                 return 0;
1273         switch (flag_type) {
1274         case BTRFS_BLOCK_GROUP_DATA:
1275         case BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA:
1276                 data_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1277                 if (flag_profile != data_profile)
1278                         return 1;
1279                 break;
1280         case BTRFS_BLOCK_GROUP_METADATA:
1281                 meta_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1282                 if (flag_profile != meta_profile)
1283                         return 1;
1284                 break;
1285         case BTRFS_BLOCK_GROUP_SYSTEM:
1286                 sys_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1287                 if (flag_profile != sys_profile)
1288                         return 1;
1289                 break;
1290         }
1291         return 0;
1292 }
1293
1294 /* Note: if current is a block group, it will skip it anyway */
1295 static int next_block_group(struct btrfs_root *root,
1296                             struct btrfs_path *path)
1297 {
1298         struct btrfs_key key;
1299         int ret = 0;
1300
1301         while (1) {
1302                 ret = btrfs_next_item(root, path);
1303                 if (ret)
1304                         goto out;
1305
1306                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1307                 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
1308                         goto out;
1309         }
1310 out:
1311         return ret;
1312 }
1313
1314 /* This function will cleanup  */
1315 static int cleanup_temp_chunks(struct btrfs_fs_info *fs_info,
1316                                struct mkfs_allocation *alloc,
1317                                u64 data_profile, u64 meta_profile,
1318                                u64 sys_profile)
1319 {
1320         struct btrfs_trans_handle *trans = NULL;
1321         struct btrfs_block_group_item *bgi;
1322         struct btrfs_root *root = fs_info->extent_root;
1323         struct btrfs_key key;
1324         struct btrfs_key found_key;
1325         struct btrfs_path path;
1326         int ret = 0;
1327
1328         btrfs_init_path(&path);
1329         trans = btrfs_start_transaction(root, 1);
1330         BUG_ON(IS_ERR(trans));
1331
1332         key.objectid = 0;
1333         key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
1334         key.offset = 0;
1335
1336         while (1) {
1337                 /*
1338                  * as the rest of the loop may modify the tree, we need to
1339                  * start a new search each time.
1340                  */
1341                 ret = btrfs_search_slot(trans, root, &key, &path, 0, 0);
1342                 if (ret < 0)
1343                         goto out;
1344
1345                 btrfs_item_key_to_cpu(path.nodes[0], &found_key,
1346                                       path.slots[0]);
1347                 if (found_key.objectid < key.objectid)
1348                         goto out;
1349                 if (found_key.type != BTRFS_BLOCK_GROUP_ITEM_KEY) {
1350                         ret = next_block_group(root, &path);
1351                         if (ret < 0)
1352                                 goto out;
1353                         if (ret > 0) {
1354                                 ret = 0;
1355                                 goto out;
1356                         }
1357                         btrfs_item_key_to_cpu(path.nodes[0], &found_key,
1358                                               path.slots[0]);
1359                 }
1360
1361                 bgi = btrfs_item_ptr(path.nodes[0], path.slots[0],
1362                                      struct btrfs_block_group_item);
1363                 if (is_temp_block_group(path.nodes[0], bgi,
1364                                         data_profile, meta_profile,
1365                                         sys_profile)) {
1366                         u64 flags = btrfs_disk_block_group_flags(path.nodes[0],
1367                                                              bgi);
1368
1369                         ret = btrfs_free_block_group(trans, fs_info,
1370                                         found_key.objectid, found_key.offset);
1371                         if (ret < 0)
1372                                 goto out;
1373
1374                         if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
1375                             BTRFS_BLOCK_GROUP_DATA)
1376                                 alloc->data -= found_key.offset;
1377                         else if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
1378                                  BTRFS_BLOCK_GROUP_METADATA)
1379                                 alloc->metadata -= found_key.offset;
1380                         else if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
1381                                  BTRFS_BLOCK_GROUP_SYSTEM)
1382                                 alloc->system -= found_key.offset;
1383                         else if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
1384                                  (BTRFS_BLOCK_GROUP_METADATA |
1385                                   BTRFS_BLOCK_GROUP_DATA))
1386                                 alloc->mixed -= found_key.offset;
1387                 }
1388                 btrfs_release_path(&path);
1389                 key.objectid = found_key.objectid + found_key.offset;
1390         }
1391 out:
1392         if (trans)
1393                 btrfs_commit_transaction(trans, root);
1394         btrfs_release_path(&path);
1395         return ret;
1396 }
1397
1398 int main(int argc, char **argv)
1399 {
1400         char *file;
1401         struct btrfs_root *root;
1402         struct btrfs_fs_info *fs_info;
1403         struct btrfs_trans_handle *trans;
1404         char *label = NULL;
1405         u64 block_count = 0;
1406         u64 dev_block_count = 0;
1407         u64 alloc_start = 0;
1408         u64 metadata_profile = 0;
1409         u64 data_profile = 0;
1410         u32 nodesize = max_t(u32, sysconf(_SC_PAGESIZE),
1411                         BTRFS_MKFS_DEFAULT_NODE_SIZE);
1412         u32 sectorsize = 4096;
1413         u32 stripesize = 4096;
1414         int zero_end = 1;
1415         int fd = -1;
1416         int ret;
1417         int i;
1418         int mixed = 0;
1419         int nodesize_forced = 0;
1420         int data_profile_opt = 0;
1421         int metadata_profile_opt = 0;
1422         int discard = 1;
1423         int ssd = 0;
1424         int force_overwrite = 0;
1425         char *source_dir = NULL;
1426         int source_dir_set = 0;
1427         u64 num_of_meta_chunks = 0;
1428         u64 size_of_data = 0;
1429         u64 source_dir_size = 0;
1430         int dev_cnt = 0;
1431         int saved_optind;
1432         char fs_uuid[BTRFS_UUID_UNPARSED_SIZE] = { 0 };
1433         u64 features = BTRFS_MKFS_DEFAULT_FEATURES;
1434         struct mkfs_allocation allocation = { 0 };
1435         struct btrfs_mkfs_config mkfs_cfg;
1436
1437         while(1) {
1438                 int c;
1439                 static const struct option long_options[] = {
1440                         { "alloc-start", required_argument, NULL, 'A'},
1441                         { "byte-count", required_argument, NULL, 'b' },
1442                         { "force", no_argument, NULL, 'f' },
1443                         { "leafsize", required_argument, NULL, 'l' },
1444                         { "label", required_argument, NULL, 'L'},
1445                         { "metadata", required_argument, NULL, 'm' },
1446                         { "mixed", no_argument, NULL, 'M' },
1447                         { "nodesize", required_argument, NULL, 'n' },
1448                         { "sectorsize", required_argument, NULL, 's' },
1449                         { "data", required_argument, NULL, 'd' },
1450                         { "version", no_argument, NULL, 'V' },
1451                         { "rootdir", required_argument, NULL, 'r' },
1452                         { "nodiscard", no_argument, NULL, 'K' },
1453                         { "features", required_argument, NULL, 'O' },
1454                         { "uuid", required_argument, NULL, 'U' },
1455                         { "quiet", 0, NULL, 'q' },
1456                         { "help", no_argument, NULL, GETOPT_VAL_HELP },
1457                         { NULL, 0, NULL, 0}
1458                 };
1459
1460                 c = getopt_long(argc, argv, "A:b:fl:n:s:m:d:L:O:r:U:VMKq",
1461                                 long_options, NULL);
1462                 if (c < 0)
1463                         break;
1464                 switch(c) {
1465                         case 'A':
1466                                 alloc_start = parse_size(optarg);
1467                                 break;
1468                         case 'f':
1469                                 force_overwrite = 1;
1470                                 break;
1471                         case 'd':
1472                                 data_profile = parse_profile(optarg);
1473                                 data_profile_opt = 1;
1474                                 break;
1475                         case 'l':
1476                                 warning("--leafsize is deprecated, use --nodesize");
1477                                 /* fall through */
1478                         case 'n':
1479                                 nodesize = parse_size(optarg);
1480                                 nodesize_forced = 1;
1481                                 break;
1482                         case 'L':
1483                                 label = parse_label(optarg);
1484                                 break;
1485                         case 'm':
1486                                 metadata_profile = parse_profile(optarg);
1487                                 metadata_profile_opt = 1;
1488                                 break;
1489                         case 'M':
1490                                 mixed = 1;
1491                                 break;
1492                         case 'O': {
1493                                 char *orig = strdup(optarg);
1494                                 char *tmp = orig;
1495
1496                                 tmp = btrfs_parse_fs_features(tmp, &features);
1497                                 if (tmp) {
1498                                         error("unrecognized filesystem feature '%s'",
1499                                                         tmp);
1500                                         free(orig);
1501                                         goto error;
1502                                 }
1503                                 free(orig);
1504                                 if (features & BTRFS_FEATURE_LIST_ALL) {
1505                                         btrfs_list_all_fs_features(0);
1506                                         goto success;
1507                                 }
1508                                 break;
1509                                 }
1510                         case 's':
1511                                 sectorsize = parse_size(optarg);
1512                                 break;
1513                         case 'b':
1514                                 block_count = parse_size(optarg);
1515                                 zero_end = 0;
1516                                 break;
1517                         case 'V':
1518                                 printf("mkfs.btrfs, part of %s\n",
1519                                                 PACKAGE_STRING);
1520                                 goto success;
1521                         case 'r':
1522                                 source_dir = optarg;
1523                                 source_dir_set = 1;
1524                                 break;
1525                         case 'U':
1526                                 strncpy(fs_uuid, optarg,
1527                                         BTRFS_UUID_UNPARSED_SIZE - 1);
1528                                 break;
1529                         case 'K':
1530                                 discard = 0;
1531                                 break;
1532                         case 'q':
1533                                 verbose = 0;
1534                                 break;
1535                         case GETOPT_VAL_HELP:
1536                         default:
1537                                 print_usage(c != GETOPT_VAL_HELP);
1538                 }
1539         }
1540
1541         if (verbose) {
1542                 printf("%s\n", PACKAGE_STRING);
1543                 printf("See %s for more information.\n\n", PACKAGE_URL);
1544         }
1545
1546         sectorsize = max(sectorsize, (u32)sysconf(_SC_PAGESIZE));
1547         stripesize = sectorsize;
1548         saved_optind = optind;
1549         dev_cnt = argc - optind;
1550         if (dev_cnt == 0)
1551                 print_usage(1);
1552
1553         if (source_dir_set && dev_cnt > 1) {
1554                 error("the option -r is limited to a single device");
1555                 goto error;
1556         }
1557
1558         if (*fs_uuid) {
1559                 uuid_t dummy_uuid;
1560
1561                 if (uuid_parse(fs_uuid, dummy_uuid) != 0) {
1562                         error("could not parse UUID: %s", fs_uuid);
1563                         goto error;
1564                 }
1565                 if (!test_uuid_unique(fs_uuid)) {
1566                         error("non-unique UUID: %s", fs_uuid);
1567                         goto error;
1568                 }
1569         }
1570
1571         while (dev_cnt-- > 0) {
1572                 file = argv[optind++];
1573                 if (is_block_device(file) == 1)
1574                         if (test_dev_for_mkfs(file, force_overwrite))
1575                                 goto error;
1576         }
1577
1578         optind = saved_optind;
1579         dev_cnt = argc - optind;
1580
1581         file = argv[optind++];
1582         ssd = is_ssd(file);
1583
1584         /*
1585         * Set default profiles according to number of added devices.
1586         * For mixed groups defaults are single/single.
1587         */
1588         if (!mixed) {
1589                 if (!metadata_profile_opt) {
1590                         if (dev_cnt == 1 && ssd && verbose)
1591                                 printf("Detected a SSD, turning off metadata "
1592                                 "duplication.  Mkfs with -m dup if you want to "
1593                                 "force metadata duplication.\n");
1594
1595                         metadata_profile = (dev_cnt > 1) ?
1596                                         BTRFS_BLOCK_GROUP_RAID1 : (ssd) ?
1597                                         0: BTRFS_BLOCK_GROUP_DUP;
1598                 }
1599                 if (!data_profile_opt) {
1600                         data_profile = (dev_cnt > 1) ?
1601                                 BTRFS_BLOCK_GROUP_RAID0 : 0; /* raid0 or single */
1602                 }
1603         } else {
1604                 u32 best_nodesize = max_t(u32, sysconf(_SC_PAGESIZE), sectorsize);
1605
1606                 if (metadata_profile_opt || data_profile_opt) {
1607                         if (metadata_profile != data_profile) {
1608                                 error(
1609         "with mixed block groups data and metadata profiles must be the same");
1610                                 goto error;
1611                         }
1612                 }
1613
1614                 if (!nodesize_forced)
1615                         nodesize = best_nodesize;
1616         }
1617
1618         /*
1619          * FS features that can be set by other means than -O
1620          * just set the bit here
1621          */
1622         if (mixed)
1623                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS;
1624
1625         if ((data_profile | metadata_profile) &
1626             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
1627                 features |= BTRFS_FEATURE_INCOMPAT_RAID56;
1628         }
1629
1630         if (btrfs_check_nodesize(nodesize, sectorsize,
1631                                  features))
1632                 goto error;
1633
1634         if (sectorsize < sizeof(struct btrfs_super_block)) {
1635                 error("sectorsize smaller than superblock: %u < %zu",
1636                                 sectorsize, sizeof(struct btrfs_super_block));
1637                 goto error;
1638         }
1639
1640         /* Check device/block_count after the nodesize is determined */
1641         if (block_count && block_count < btrfs_min_dev_size(nodesize)) {
1642                 error("size %llu is too small to make a usable filesystem",
1643                         block_count);
1644                 error("minimum size for btrfs filesystem is %llu",
1645                         btrfs_min_dev_size(nodesize));
1646                 goto error;
1647         }
1648         for (i = saved_optind; i < saved_optind + dev_cnt; i++) {
1649                 char *path;
1650
1651                 path = argv[i];
1652                 ret = test_minimum_size(path, nodesize);
1653                 if (ret < 0) {
1654                         error("failed to check size for %s: %s",
1655                                 path, strerror(-ret));
1656                         goto error;
1657                 }
1658                 if (ret > 0) {
1659                         error("'%s' is too small to make a usable filesystem",
1660                                 path);
1661                         error("minimum size for each btrfs device is %llu",
1662                                 btrfs_min_dev_size(nodesize));
1663                         goto error;
1664                 }
1665         }
1666         ret = test_num_disk_vs_raid(metadata_profile, data_profile,
1667                         dev_cnt, mixed, ssd);
1668         if (ret)
1669                 goto error;
1670
1671         dev_cnt--;
1672
1673         if (!source_dir_set) {
1674                 /*
1675                  * open without O_EXCL so that the problem should not
1676                  * occur by the following processing.
1677                  * (btrfs_register_one_device() fails if O_EXCL is on)
1678                  */
1679                 fd = open(file, O_RDWR);
1680                 if (fd < 0) {
1681                         error("unable to open %s: %s", file, strerror(errno));
1682                         goto error;
1683                 }
1684                 ret = btrfs_prepare_device(fd, file, &dev_block_count,
1685                                 block_count,
1686                                 (zero_end ? PREP_DEVICE_ZERO_END : 0) |
1687                                 (discard ? PREP_DEVICE_DISCARD : 0) |
1688                                 (verbose ? PREP_DEVICE_VERBOSE : 0));
1689                 if (ret) {
1690                         goto error;
1691                 }
1692                 if (block_count && block_count > dev_block_count) {
1693                         error("%s is smaller than requested size, expected %llu, found %llu",
1694                                         file,
1695                                         (unsigned long long)block_count,
1696                                         (unsigned long long)dev_block_count);
1697                         goto error;
1698                 }
1699         } else {
1700                 fd = open(file, O_CREAT | O_RDWR,
1701                                 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH);
1702                 if (fd < 0) {
1703                         error("unable to open %s: %s", file, strerror(errno));
1704                         goto error;
1705                 }
1706
1707                 source_dir_size = size_sourcedir(source_dir, sectorsize,
1708                                              &num_of_meta_chunks, &size_of_data);
1709                 if(block_count < source_dir_size)
1710                         block_count = source_dir_size;
1711                 ret = zero_output_file(fd, block_count);
1712                 if (ret) {
1713                         error("unable to zero the output file");
1714                         goto error;
1715                 }
1716                 /* our "device" is the new image file */
1717                 dev_block_count = block_count;
1718         }
1719
1720         /* To create the first block group and chunk 0 in make_btrfs */
1721         if (dev_block_count < BTRFS_MKFS_SYSTEM_GROUP_SIZE) {
1722                 error("device is too small to make filesystem, must be at least %llu",
1723                                 (unsigned long long)BTRFS_MKFS_SYSTEM_GROUP_SIZE);
1724                 goto error;
1725         }
1726
1727         if (group_profile_max_safe_loss(metadata_profile) <
1728                 group_profile_max_safe_loss(data_profile)){
1729                 warning("metadata has lower redundancy than data!\n");
1730         }
1731
1732         mkfs_cfg.label = label;
1733         memcpy(mkfs_cfg.fs_uuid, fs_uuid, sizeof(mkfs_cfg.fs_uuid));
1734         mkfs_cfg.num_bytes = dev_block_count;
1735         mkfs_cfg.nodesize = nodesize;
1736         mkfs_cfg.sectorsize = sectorsize;
1737         mkfs_cfg.stripesize = stripesize;
1738         mkfs_cfg.features = features;
1739
1740         ret = make_btrfs(fd, &mkfs_cfg);
1741         if (ret) {
1742                 error("error during mkfs: %s", strerror(-ret));
1743                 goto error;
1744         }
1745
1746         fs_info = open_ctree_fs_info(file, 0, 0, 0,
1747                         OPEN_CTREE_WRITES | OPEN_CTREE_FS_PARTIAL);
1748         if (!fs_info) {
1749                 error("open ctree failed");
1750                 goto error;
1751         }
1752         close(fd);
1753         fd = -1;
1754         root = fs_info->fs_root;
1755         fs_info->alloc_start = alloc_start;
1756
1757         ret = create_metadata_block_groups(root, mixed, &allocation);
1758         if (ret) {
1759                 error("failed to create default block groups: %d", ret);
1760                 goto error;
1761         }
1762
1763         trans = btrfs_start_transaction(root, 1);
1764         if (IS_ERR(trans)) {
1765                 error("failed to start transaction");
1766                 goto error;
1767         }
1768
1769         ret = create_data_block_groups(trans, root, mixed, &allocation);
1770         if (ret) {
1771                 error("failed to create default data block groups: %d", ret);
1772                 goto error;
1773         }
1774
1775         ret = make_root_dir(trans, root);
1776         if (ret) {
1777                 error("failed to setup the root directory: %d", ret);
1778                 goto error;
1779         }
1780
1781         ret = btrfs_commit_transaction(trans, root);
1782         if (ret) {
1783                 error("unable to commit transaction: %d", ret);
1784                 goto out;
1785         }
1786
1787         trans = btrfs_start_transaction(root, 1);
1788         if (IS_ERR(trans)) {
1789                 error("failed to start transaction");
1790                 goto error;
1791         }
1792
1793         if (dev_cnt == 0)
1794                 goto raid_groups;
1795
1796         while (dev_cnt-- > 0) {
1797                 file = argv[optind++];
1798
1799                 /*
1800                  * open without O_EXCL so that the problem should not
1801                  * occur by the following processing.
1802                  * (btrfs_register_one_device() fails if O_EXCL is on)
1803                  */
1804                 fd = open(file, O_RDWR);
1805                 if (fd < 0) {
1806                         error("unable to open %s: %s", file, strerror(errno));
1807                         goto error;
1808                 }
1809                 ret = btrfs_device_already_in_root(root, fd,
1810                                                    BTRFS_SUPER_INFO_OFFSET);
1811                 if (ret) {
1812                         error("skipping duplicate device %s in the filesystem",
1813                                 file);
1814                         close(fd);
1815                         continue;
1816                 }
1817                 ret = btrfs_prepare_device(fd, file, &dev_block_count,
1818                                 block_count,
1819                                 (verbose ? PREP_DEVICE_VERBOSE : 0) |
1820                                 (zero_end ? PREP_DEVICE_ZERO_END : 0) |
1821                                 (discard ? PREP_DEVICE_DISCARD : 0));
1822                 if (ret) {
1823                         goto error;
1824                 }
1825
1826                 ret = btrfs_add_to_fsid(trans, root, fd, file, dev_block_count,
1827                                         sectorsize, sectorsize, sectorsize);
1828                 if (ret) {
1829                         error("unable to add %s to filesystem: %d", file, ret);
1830                         goto out;
1831                 }
1832                 if (verbose >= 2) {
1833                         struct btrfs_device *device;
1834
1835                         device = container_of(fs_info->fs_devices->devices.next,
1836                                         struct btrfs_device, dev_list);
1837                         printf("adding device %s id %llu\n", file,
1838                                 (unsigned long long)device->devid);
1839                 }
1840         }
1841
1842 raid_groups:
1843         if (!source_dir_set) {
1844                 ret = create_raid_groups(trans, root, data_profile,
1845                                  metadata_profile, mixed, &allocation);
1846                 if (ret) {
1847                         error("unable to create raid groups: %d", ret);
1848                         goto out;
1849                 }
1850         }
1851
1852         ret = create_data_reloc_tree(trans, root);
1853         if (ret) {
1854                 error("unable to create data reloc tree: %d", ret);
1855                 goto out;
1856         }
1857
1858         ret = btrfs_commit_transaction(trans, root);
1859         if (ret) {
1860                 error("unable to commit transaction: %d", ret);
1861                 goto out;
1862         }
1863
1864         if (source_dir_set) {
1865                 trans = btrfs_start_transaction(root, 1);
1866                 BUG_ON(IS_ERR(trans));
1867                 ret = create_chunks(trans, root,
1868                                     num_of_meta_chunks, size_of_data,
1869                                     &allocation);
1870                 if (ret) {
1871                         error("unable to create chunks: %d", ret);
1872                         goto out;
1873                 }
1874                 ret = btrfs_commit_transaction(trans, root);
1875                 if (ret) {
1876                         error("transaction commit failed: %d", ret);
1877                         goto out;
1878                 }
1879
1880                 ret = make_image(source_dir, root);
1881                 if (ret) {
1882                         error("error wihle filling filesystem: %d", ret);
1883                         goto out;
1884                 }
1885         }
1886         ret = cleanup_temp_chunks(fs_info, &allocation, data_profile,
1887                                   metadata_profile, metadata_profile);
1888         if (ret < 0) {
1889                 error("failed to cleanup temporary chunks: %d", ret);
1890                 goto out;
1891         }
1892
1893         if (verbose) {
1894                 char features_buf[64];
1895
1896                 printf("Label:              %s\n", label);
1897                 printf("UUID:               %s\n", mkfs_cfg.fs_uuid);
1898                 printf("Node size:          %u\n", nodesize);
1899                 printf("Sector size:        %u\n", sectorsize);
1900                 printf("Filesystem size:    %s\n",
1901                         pretty_size(btrfs_super_total_bytes(fs_info->super_copy)));
1902                 printf("Block group profiles:\n");
1903                 if (allocation.data)
1904                         printf("  Data:             %-8s %16s\n",
1905                                 btrfs_group_profile_str(data_profile),
1906                                 pretty_size(allocation.data));
1907                 if (allocation.metadata)
1908                         printf("  Metadata:         %-8s %16s\n",
1909                                 btrfs_group_profile_str(metadata_profile),
1910                                 pretty_size(allocation.metadata));
1911                 if (allocation.mixed)
1912                         printf("  Data+Metadata:    %-8s %16s\n",
1913                                 btrfs_group_profile_str(data_profile),
1914                                 pretty_size(allocation.mixed));
1915                 printf("  System:           %-8s %16s\n",
1916                         btrfs_group_profile_str(metadata_profile),
1917                         pretty_size(allocation.system));
1918                 printf("SSD detected:       %s\n", ssd ? "yes" : "no");
1919                 btrfs_parse_features_to_string(features_buf, features);
1920                 printf("Incompat features:  %s", features_buf);
1921                 printf("\n");
1922
1923                 list_all_devices(root);
1924         }
1925
1926         /*
1927          * The filesystem is now fully set up, commit the remaining changes and
1928          * fix the signature as the last step before closing the devices.
1929          */
1930         fs_info->finalize_on_close = 1;
1931 out:
1932         ret = close_ctree(root);
1933
1934         if (!ret) {
1935                 optind = saved_optind;
1936                 dev_cnt = argc - optind;
1937                 while (dev_cnt-- > 0) {
1938                         file = argv[optind++];
1939                         if (is_block_device(file) == 1)
1940                                 btrfs_register_one_device(file);
1941                 }
1942         }
1943
1944         btrfs_close_all_devices();
1945         free(label);
1946
1947         return !!ret;
1948 error:
1949         if (fd > 0)
1950                 close(fd);
1951
1952         free(label);
1953         exit(1);
1954 success:
1955         exit(0);
1956 }