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