mkfs.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 #define _XOPEN_SOURCE 500
  20 #define _GNU_SOURCE
  21
  22 #include "kerncompat.h"
  23
  24 #include <sys/ioctl.h>
  25 #include <sys/mount.h>
  26 #include "ioctl.h"
  27 #include <stdio.h>
  28 #include <stdlib.h>
  29 #include <sys/types.h>
  30 #include <sys/stat.h>
  31 #include <sys/dir.h>
  32 #include <fcntl.h>
  33 #include <unistd.h>
  34 #include <getopt.h>
  35 #include <uuid/uuid.h>
  36 #include <ctype.h>
  37 #include <attr/xattr.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 "version.h"
  46
  47 static u64 index_cnt = 2;
  48
  49 #define DEFAULT_MKFS_FEATURES   (BTRFS_FEATURE_INCOMPAT_EXTENDED_IREF)
  50
  51 #define DEFAULT_MKFS_LEAF_SIZE 16384
  52
  53 struct directory_name_entry {
  54         char *dir_name;
  55         char *path;
  56         ino_t inum;
  57         struct list_head list;
  58 };
  59
  60 static int make_root_dir(struct btrfs_root *root, int mixed)
  61 {
  62         struct btrfs_trans_handle *trans;
  63         struct btrfs_key location;
  64         u64 bytes_used;
  65         u64 chunk_start = 0;
  66         u64 chunk_size = 0;
  67         int ret;
  68
  69         trans = btrfs_start_transaction(root, 1);
  70         bytes_used = btrfs_super_bytes_used(root->fs_info->super_copy);
  71
  72         root->fs_info->system_allocs = 1;
  73         ret = btrfs_make_block_group(trans, root, bytes_used,
  74                                      BTRFS_BLOCK_GROUP_SYSTEM,
  75                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
  76                                      0, BTRFS_MKFS_SYSTEM_GROUP_SIZE);
  77         BUG_ON(ret);
  78
  79         if (mixed) {
  80                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
  81                                         &chunk_start, &chunk_size,
  82                                         BTRFS_BLOCK_GROUP_METADATA |
  83                                         BTRFS_BLOCK_GROUP_DATA);
  84                 if (ret == -ENOSPC) {
  85                         fprintf(stderr,
  86                                 "no space to alloc data/metadata chunk\n");
  87                         goto err;
  88                 }
  89                 BUG_ON(ret);
  90                 ret = btrfs_make_block_group(trans, root, 0,
  91                                              BTRFS_BLOCK_GROUP_METADATA |
  92                                              BTRFS_BLOCK_GROUP_DATA,
  93                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
  94                                              chunk_start, chunk_size);
  95                 BUG_ON(ret);
  96                 printf("Created a data/metadata chunk of size %llu\n", chunk_size);
  97         } else {
  98                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
  99                                         &chunk_start, &chunk_size,
 100                                         BTRFS_BLOCK_GROUP_METADATA);
 101                 if (ret == -ENOSPC) {
 102                         fprintf(stderr, "no space to alloc metadata chunk\n");
 103                         goto err;
 104                 }
 105                 BUG_ON(ret);
 106                 ret = btrfs_make_block_group(trans, root, 0,
 107                                              BTRFS_BLOCK_GROUP_METADATA,
 108                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 109                                              chunk_start, chunk_size);
 110                 BUG_ON(ret);
 111         }
 112
 113         root->fs_info->system_allocs = 0;
 114         btrfs_commit_transaction(trans, root);
 115         trans = btrfs_start_transaction(root, 1);
 116         BUG_ON(!trans);
 117
 118         if (!mixed) {
 119                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 120                                         &chunk_start, &chunk_size,
 121                                         BTRFS_BLOCK_GROUP_DATA);
 122                 if (ret == -ENOSPC) {
 123                         fprintf(stderr, "no space to alloc data chunk\n");
 124                         goto err;
 125                 }
 126                 BUG_ON(ret);
 127                 ret = btrfs_make_block_group(trans, root, 0,
 128                                              BTRFS_BLOCK_GROUP_DATA,
 129                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 130                                              chunk_start, chunk_size);
 131                 BUG_ON(ret);
 132         }
 133
 134         ret = btrfs_make_root_dir(trans, root->fs_info->tree_root,
 135                               BTRFS_ROOT_TREE_DIR_OBJECTID);
 136         if (ret)
 137                 goto err;
 138         ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
 139         if (ret)
 140                 goto err;
 141         memcpy(&location, &root->fs_info->fs_root->root_key, sizeof(location));
 142         location.offset = (u64)-1;
 143         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
 144                         "default", 7,
 145                         btrfs_super_root_dir(root->fs_info->super_copy),
 146                         &location, BTRFS_FT_DIR, 0);
 147         if (ret)
 148                 goto err;
 149
 150         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 151                              "default", 7, location.objectid,
 152                              BTRFS_ROOT_TREE_DIR_OBJECTID, 0);
 153         if (ret)
 154                 goto err;
 155
 156         btrfs_commit_transaction(trans, root);
 157 err:
 158         return ret;
 159 }
 160
 161 static void __recow_root(struct btrfs_trans_handle *trans,
 162                          struct btrfs_root *root)
 163 {
 164         int ret;
 165         struct extent_buffer *tmp;
 166
 167         if (trans->transid != btrfs_root_generation(&root->root_item)) {
 168                 extent_buffer_get(root->node);
 169                 ret = __btrfs_cow_block(trans, root, root->node,
 170                                         NULL, 0, &tmp, 0, 0);
 171                 BUG_ON(ret);
 172                 free_extent_buffer(tmp);
 173         }
 174 }
 175
 176 static void recow_roots(struct btrfs_trans_handle *trans,
 177                        struct btrfs_root *root)
 178 {
 179         struct btrfs_fs_info *info = root->fs_info;
 180
 181         __recow_root(trans, info->fs_root);
 182         __recow_root(trans, info->tree_root);
 183         __recow_root(trans, info->extent_root);
 184         __recow_root(trans, info->chunk_root);
 185         __recow_root(trans, info->dev_root);
 186         __recow_root(trans, info->csum_root);
 187 }
 188
 189 static int create_one_raid_group(struct btrfs_trans_handle *trans,
 190                               struct btrfs_root *root, u64 type)
 191 {
 192         u64 chunk_start;
 193         u64 chunk_size;
 194         int ret;
 195
 196         ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 197                                 &chunk_start, &chunk_size, type);
 198         if (ret == -ENOSPC) {
 199                 fprintf(stderr, "not enough free space\n");
 200                 exit(1);
 201         }
 202         BUG_ON(ret);
 203         ret = btrfs_make_block_group(trans, root->fs_info->extent_root, 0,
 204                                      type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 205                                      chunk_start, chunk_size);
 206         BUG_ON(ret);
 207         return ret;
 208 }
 209
 210 static int create_raid_groups(struct btrfs_trans_handle *trans,
 211                               struct btrfs_root *root, u64 data_profile,
 212                               int data_profile_opt, u64 metadata_profile,
 213                               int metadata_profile_opt, int mixed, int ssd)
 214 {
 215         u64 num_devices = btrfs_super_num_devices(root->fs_info->super_copy);
 216         int ret;
 217
 218         if (metadata_profile) {
 219                 u64 meta_flags = BTRFS_BLOCK_GROUP_METADATA;
 220
 221                 ret = create_one_raid_group(trans, root,
 222                                             BTRFS_BLOCK_GROUP_SYSTEM |
 223                                             metadata_profile);
 224                 BUG_ON(ret);
 225
 226                 if (mixed)
 227                         meta_flags |= BTRFS_BLOCK_GROUP_DATA;
 228
 229                 ret = create_one_raid_group(trans, root, meta_flags |
 230                                             metadata_profile);
 231                 BUG_ON(ret);
 232
 233         }
 234         if (!mixed && num_devices > 1 && data_profile) {
 235                 ret = create_one_raid_group(trans, root,
 236                                             BTRFS_BLOCK_GROUP_DATA |
 237                                             data_profile);
 238                 BUG_ON(ret);
 239         }
 240         recow_roots(trans, root);
 241
 242         return 0;
 243 }
 244
 245 static int create_data_reloc_tree(struct btrfs_trans_handle *trans,
 246                                   struct btrfs_root *root)
 247 {
 248         struct btrfs_key location;
 249         struct btrfs_root_item root_item;
 250         struct extent_buffer *tmp;
 251         u64 objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
 252         int ret;
 253
 254         ret = btrfs_copy_root(trans, root, root->node, &tmp, objectid);
 255         BUG_ON(ret);
 256
 257         memcpy(&root_item, &root->root_item, sizeof(root_item));
 258         btrfs_set_root_bytenr(&root_item, tmp->start);
 259         btrfs_set_root_level(&root_item, btrfs_header_level(tmp));
 260         btrfs_set_root_generation(&root_item, trans->transid);
 261         free_extent_buffer(tmp);
 262
 263         location.objectid = objectid;
 264         location.type = BTRFS_ROOT_ITEM_KEY;
 265         location.offset = 0;
 266         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
 267                                 &location, &root_item);
 268         BUG_ON(ret);
 269         return 0;
 270 }
 271
 272
 273 static void print_usage(void) __attribute__((noreturn));
 274 static void print_usage(void)
 275 {
 276         fprintf(stderr, "usage: mkfs.btrfs [options] dev [ dev ... ]\n");
 277         fprintf(stderr, "options:\n");
 278         fprintf(stderr, "\t -A --alloc-start the offset to start the FS\n");
 279         fprintf(stderr, "\t -b --byte-count total number of bytes in the FS\n");
 280         fprintf(stderr, "\t -d --data data profile, raid0, raid1, raid5, raid6, raid10, dup or single\n");
 281         fprintf(stderr, "\t -f --force force overwrite of existing filesystem\n");
 282         fprintf(stderr, "\t -l --leafsize size of btree leaves\n");
 283         fprintf(stderr, "\t -L --label set a label\n");
 284         fprintf(stderr, "\t -m --metadata metadata profile, values like data profile\n");
 285         fprintf(stderr, "\t -M --mixed mix metadata and data together\n");
 286         fprintf(stderr, "\t -n --nodesize size of btree nodes\n");
 287         fprintf(stderr, "\t -s --sectorsize min block allocation (may not mountable by current kernel)\n");
 288         fprintf(stderr, "\t -r --rootdir the source directory\n");
 289         fprintf(stderr, "\t -K --nodiscard do not perform whole device TRIM\n");
 290         fprintf(stderr, "\t -O --features comma separated list of filesystem features\n");
 291         fprintf(stderr, "\t -V --version print the mkfs.btrfs version and exit\n");
 292         fprintf(stderr, "%s\n", BTRFS_BUILD_VERSION);
 293         exit(1);
 294 }
 295
 296 static void print_version(void) __attribute__((noreturn));
 297 static void print_version(void)
 298 {
 299         fprintf(stderr, "mkfs.btrfs, part of %s\n", BTRFS_BUILD_VERSION);
 300         exit(0);
 301 }
 302
 303 static u64 parse_profile(char *s)
 304 {
 305         if (strcmp(s, "raid0") == 0) {
 306                 return BTRFS_BLOCK_GROUP_RAID0;
 307         } else if (strcmp(s, "raid1") == 0) {
 308                 return BTRFS_BLOCK_GROUP_RAID1;
 309         } else if (strcmp(s, "raid5") == 0) {
 310                 return BTRFS_BLOCK_GROUP_RAID5;
 311         } else if (strcmp(s, "raid6") == 0) {
 312                 return BTRFS_BLOCK_GROUP_RAID6;
 313         } else if (strcmp(s, "raid10") == 0) {
 314                 return BTRFS_BLOCK_GROUP_RAID10;
 315         } else if (strcmp(s, "dup") == 0) {
 316                 return BTRFS_BLOCK_GROUP_DUP;
 317         } else if (strcmp(s, "single") == 0) {
 318                 return 0;
 319         } else {
 320                 fprintf(stderr, "Unknown profile %s\n", s);
 321                 print_usage();
 322         }
 323         /* not reached */
 324         return 0;
 325 }
 326
 327 static char *parse_label(char *input)
 328 {
 329         int len = strlen(input);
 330
 331         if (len >= BTRFS_LABEL_SIZE) {
 332                 fprintf(stderr, "Label %s is too long (max %d)\n", input,
 333                         BTRFS_LABEL_SIZE - 1);
 334                 exit(1);
 335         }
 336         return strdup(input);
 337 }
 338
 339 static struct option long_options[] = {
 340         { "alloc-start", 1, NULL, 'A'},
 341         { "byte-count", 1, NULL, 'b' },
 342         { "force", 0, NULL, 'f' },
 343         { "leafsize", 1, NULL, 'l' },
 344         { "label", 1, NULL, 'L'},
 345         { "metadata", 1, NULL, 'm' },
 346         { "mixed", 0, NULL, 'M' },
 347         { "nodesize", 1, NULL, 'n' },
 348         { "sectorsize", 1, NULL, 's' },
 349         { "data", 1, NULL, 'd' },
 350         { "version", 0, NULL, 'V' },
 351         { "rootdir", 1, NULL, 'r' },
 352         { "nodiscard", 0, NULL, 'K' },
 353         { "features", 0, NULL, 'O' },
 354         { NULL, 0, NULL, 0}
 355 };
 356
 357 static int add_directory_items(struct btrfs_trans_handle *trans,
 358                                struct btrfs_root *root, u64 objectid,
 359                                ino_t parent_inum, const char *name,
 360                                struct stat *st, int *dir_index_cnt)
 361 {
 362         int ret;
 363         int name_len;
 364         struct btrfs_key location;
 365         u8 filetype = 0;
 366
 367         name_len = strlen(name);
 368
 369         location.objectid = objectid;
 370         location.offset = 0;
 371         btrfs_set_key_type(&location, BTRFS_INODE_ITEM_KEY);
 372
 373         if (S_ISDIR(st->st_mode))
 374                 filetype = BTRFS_FT_DIR;
 375         if (S_ISREG(st->st_mode))
 376                 filetype = BTRFS_FT_REG_FILE;
 377         if (S_ISLNK(st->st_mode))
 378                 filetype = BTRFS_FT_SYMLINK;
 379
 380         ret = btrfs_insert_dir_item(trans, root, name, name_len,
 381                                     parent_inum, &location,
 382                                     filetype, index_cnt);
 383
 384         *dir_index_cnt = index_cnt;
 385         index_cnt++;
 386
 387         return ret;
 388 }
 389
 390 static int fill_inode_item(struct btrfs_trans_handle *trans,
 391                            struct btrfs_root *root,
 392                            struct btrfs_inode_item *dst, struct stat *src)
 393 {
 394         u64 blocks = 0;
 395         u64 sectorsize = root->sectorsize;
 396
 397         /*
 398          * btrfs_inode_item has some reserved fields
 399          * and represents on-disk inode entry, so
 400          * zero everything to prevent information leak
 401          */
 402         memset(dst, 0, sizeof (*dst));
 403
 404         btrfs_set_stack_inode_generation(dst, trans->transid);
 405         btrfs_set_stack_inode_size(dst, src->st_size);
 406         btrfs_set_stack_inode_nbytes(dst, 0);
 407         btrfs_set_stack_inode_block_group(dst, 0);
 408         btrfs_set_stack_inode_nlink(dst, src->st_nlink);
 409         btrfs_set_stack_inode_uid(dst, src->st_uid);
 410         btrfs_set_stack_inode_gid(dst, src->st_gid);
 411         btrfs_set_stack_inode_mode(dst, src->st_mode);
 412         btrfs_set_stack_inode_rdev(dst, 0);
 413         btrfs_set_stack_inode_flags(dst, 0);
 414         btrfs_set_stack_timespec_sec(&dst->atime, src->st_atime);
 415         btrfs_set_stack_timespec_nsec(&dst->atime, 0);
 416         btrfs_set_stack_timespec_sec(&dst->ctime, src->st_ctime);
 417         btrfs_set_stack_timespec_nsec(&dst->ctime, 0);
 418         btrfs_set_stack_timespec_sec(&dst->mtime, src->st_mtime);
 419         btrfs_set_stack_timespec_nsec(&dst->mtime, 0);
 420         btrfs_set_stack_timespec_sec(&dst->otime, 0);
 421         btrfs_set_stack_timespec_nsec(&dst->otime, 0);
 422
 423         if (S_ISDIR(src->st_mode)) {
 424                 btrfs_set_stack_inode_size(dst, 0);
 425                 btrfs_set_stack_inode_nlink(dst, 1);
 426         }
 427         if (S_ISREG(src->st_mode)) {
 428                 btrfs_set_stack_inode_size(dst, (u64)src->st_size);
 429                 if (src->st_size <= BTRFS_MAX_INLINE_DATA_SIZE(root))
 430                         btrfs_set_stack_inode_nbytes(dst, src->st_size);
 431                 else {
 432                         blocks = src->st_size / sectorsize;
 433                         if (src->st_size % sectorsize)
 434                                 blocks += 1;
 435                         blocks *= sectorsize;
 436                         btrfs_set_stack_inode_nbytes(dst, blocks);
 437                 }
 438         }
 439         if (S_ISLNK(src->st_mode))
 440                 btrfs_set_stack_inode_nbytes(dst, src->st_size + 1);
 441
 442         return 0;
 443 }
 444
 445 static int directory_select(const struct direct *entry)
 446 {
 447         if ((strncmp(entry->d_name, ".", entry->d_reclen) == 0) ||
 448                 (strncmp(entry->d_name, "..", entry->d_reclen) == 0))
 449                 return 0;
 450         else
 451                 return 1;
 452 }
 453
 454 static void free_namelist(struct direct **files, int count)
 455 {
 456         int i;
 457
 458         if (count < 0)
 459                 return;
 460
 461         for (i = 0; i < count; ++i)
 462                 free(files[i]);
 463         free(files);
 464 }
 465
 466 static u64 calculate_dir_inode_size(char *dirname)
 467 {
 468         int count, i;
 469         struct direct **files, *cur_file;
 470         u64 dir_inode_size = 0;
 471
 472         count = scandir(dirname, &files, directory_select, NULL);
 473
 474         for (i = 0; i < count; i++) {
 475                 cur_file = files[i];
 476                 dir_inode_size += strlen(cur_file->d_name);
 477         }
 478
 479         free_namelist(files, count);
 480
 481         dir_inode_size *= 2;
 482         return dir_inode_size;
 483 }
 484
 485 static int add_inode_items(struct btrfs_trans_handle *trans,
 486                            struct btrfs_root *root,
 487                            struct stat *st, char *name,
 488                            u64 self_objectid, ino_t parent_inum,
 489                            int dir_index_cnt, struct btrfs_inode_item *inode_ret)
 490 {
 491         int ret;
 492         struct btrfs_key inode_key;
 493         struct btrfs_inode_item btrfs_inode;
 494         u64 objectid;
 495         u64 inode_size = 0;
 496         int name_len;
 497
 498         name_len = strlen(name);
 499         fill_inode_item(trans, root, &btrfs_inode, st);
 500         objectid = self_objectid;
 501
 502         if (S_ISDIR(st->st_mode)) {
 503                 inode_size = calculate_dir_inode_size(name);
 504                 btrfs_set_stack_inode_size(&btrfs_inode, inode_size);
 505         }
 506
 507         inode_key.objectid = objectid;
 508         inode_key.offset = 0;
 509         btrfs_set_key_type(&inode_key, BTRFS_INODE_ITEM_KEY);
 510
 511         ret = btrfs_insert_inode(trans, root, objectid, &btrfs_inode);
 512         if (ret)
 513                 goto fail;
 514
 515         ret = btrfs_insert_inode_ref(trans, root, name, name_len,
 516                                      objectid, parent_inum, dir_index_cnt);
 517         if (ret)
 518                 goto fail;
 519
 520         *inode_ret = btrfs_inode;
 521 fail:
 522         return ret;
 523 }
 524
 525 static int add_xattr_item(struct btrfs_trans_handle *trans,
 526                           struct btrfs_root *root, u64 objectid,
 527                           const char *file_name)
 528 {
 529         int ret;
 530         int cur_name_len;
 531         char xattr_list[XATTR_LIST_MAX];
 532         char *cur_name;
 533         char cur_value[XATTR_SIZE_MAX];
 534         char delimiter = '\0';
 535         char *next_location = xattr_list;
 536
 537         ret = llistxattr(file_name, xattr_list, XATTR_LIST_MAX);
 538         if (ret < 0) {
 539                 if(errno == ENOTSUP)
 540                         return 0;
 541                 fprintf(stderr, "get a list of xattr failed for %s\n",
 542                         file_name);
 543                 return ret;
 544         }
 545         if (ret == 0)
 546                 return ret;
 547
 548         cur_name = strtok(xattr_list, &delimiter);
 549         while (cur_name != NULL) {
 550                 cur_name_len = strlen(cur_name);
 551                 next_location += cur_name_len + 1;
 552
 553                 ret = getxattr(file_name, cur_name, cur_value, XATTR_SIZE_MAX);
 554                 if (ret < 0) {
 555                         if(errno == ENOTSUP)
 556                                 return 0;
 557                         fprintf(stderr, "get a xattr value failed for %s attr %s\n",
 558                                 file_name, cur_name);
 559                         return ret;
 560                 }
 561
 562                 ret = btrfs_insert_xattr_item(trans, root, cur_name,
 563                                               cur_name_len, cur_value,
 564                                               ret, objectid);
 565                 if (ret) {
 566                         fprintf(stderr, "insert a xattr item failed for %s\n",
 567                                 file_name);
 568                 }
 569
 570                 cur_name = strtok(next_location, &delimiter);
 571         }
 572
 573         return ret;
 574 }
 575
 576 static int add_symbolic_link(struct btrfs_trans_handle *trans,
 577                              struct btrfs_root *root,
 578                              u64 objectid, const char *path_name)
 579 {
 580         int ret;
 581         u64 sectorsize = root->sectorsize;
 582         char *buf = malloc(sectorsize);
 583
 584         ret = readlink(path_name, buf, sectorsize);
 585         if (ret <= 0) {
 586                 fprintf(stderr, "readlink failed for %s\n", path_name);
 587                 goto fail;
 588         }
 589         if (ret >= sectorsize) {
 590                 fprintf(stderr, "symlink too long for %s", path_name);
 591                 ret = -1;
 592                 goto fail;
 593         }
 594
 595         buf[ret] = '\0'; /* readlink does not do it for us */
 596         ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
 597                                          buf, ret + 1);
 598 fail:
 599         free(buf);
 600         return ret;
 601 }
 602
 603 static int add_file_items(struct btrfs_trans_handle *trans,
 604                           struct btrfs_root *root,
 605                           struct btrfs_inode_item *btrfs_inode, u64 objectid,
 606                           ino_t parent_inum, struct stat *st,
 607                           const char *path_name, int out_fd)
 608 {
 609         int ret = -1;
 610         ssize_t ret_read;
 611         u64 bytes_read = 0;
 612         struct btrfs_key key;
 613         int blocks;
 614         u32 sectorsize = root->sectorsize;
 615         u64 first_block = 0;
 616         u64 file_pos = 0;
 617         u64 cur_bytes;
 618         u64 total_bytes;
 619         struct extent_buffer *eb = NULL;
 620         int fd;
 621
 622         fd = open(path_name, O_RDONLY);
 623         if (fd == -1) {
 624                 fprintf(stderr, "%s open failed\n", path_name);
 625                 return ret;
 626         }
 627
 628         blocks = st->st_size / sectorsize;
 629         if (st->st_size % sectorsize)
 630                 blocks += 1;
 631
 632         if (st->st_size <= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 633                 char *buffer = malloc(st->st_size);
 634                 ret_read = pread64(fd, buffer, st->st_size, bytes_read);
 635                 if (ret_read == -1) {
 636                         fprintf(stderr, "%s read failed\n", path_name);
 637                         free(buffer);
 638                         goto end;
 639                 }
 640
 641                 ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
 642                                                  buffer, st->st_size);
 643                 free(buffer);
 644                 goto end;
 645         }
 646
 647         /* round up our st_size to the FS blocksize */
 648         total_bytes = (u64)blocks * sectorsize;
 649
 650         /*
 651          * do our IO in extent buffers so it can work
 652          * against any raid type
 653          */
 654         eb = malloc(sizeof(*eb) + sectorsize);
 655         if (!eb) {
 656                 ret = -ENOMEM;
 657                 goto end;
 658         }
 659         memset(eb, 0, sizeof(*eb) + sectorsize);
 660
 661 again:
 662
 663         /*
 664          * keep our extent size at 1MB max, this makes it easier to work inside
 665          * the tiny block groups created during mkfs
 666          */
 667         cur_bytes = min(total_bytes, 1024ULL * 1024);
 668         ret = btrfs_reserve_extent(trans, root, cur_bytes, 0, 0, (u64)-1,
 669                                    &key, 1);
 670         if (ret)
 671                 goto end;
 672
 673         first_block = key.objectid;
 674         bytes_read = 0;
 675
 676         while (bytes_read < cur_bytes) {
 677
 678                 memset(eb->data, 0, sectorsize);
 679
 680                 ret_read = pread64(fd, eb->data, sectorsize, file_pos + bytes_read);
 681                 if (ret_read == -1) {
 682                         fprintf(stderr, "%s read failed\n", path_name);
 683                         goto end;
 684                 }
 685
 686                 eb->start = first_block + bytes_read;
 687                 eb->len = sectorsize;
 688
 689                 /*
 690                  * we're doing the csum before we record the extent, but
 691                  * that's ok
 692                  */
 693                 ret = btrfs_csum_file_block(trans, root->fs_info->csum_root,
 694                                             first_block + bytes_read + sectorsize,
 695                                             first_block + bytes_read,
 696                                             eb->data, sectorsize);
 697                 if (ret)
 698                         goto end;
 699
 700                 ret = write_and_map_eb(trans, root, eb);
 701                 if (ret) {
 702                         fprintf(stderr, "output file write failed\n");
 703                         goto end;
 704                 }
 705
 706                 bytes_read += sectorsize;
 707         }
 708
 709         if (bytes_read) {
 710                 ret = btrfs_record_file_extent(trans, root, objectid, btrfs_inode,
 711                                                file_pos, first_block, cur_bytes);
 712                 if (ret)
 713                         goto end;
 714
 715         }
 716
 717         file_pos += cur_bytes;
 718         total_bytes -= cur_bytes;
 719
 720         if (total_bytes)
 721                 goto again;
 722
 723 end:
 724         if (eb)
 725                 free(eb);
 726         close(fd);
 727         return ret;
 728 }
 729
 730 static char *make_path(char *dir, char *name)
 731 {
 732         char *path;
 733
 734         path = malloc(strlen(dir) + strlen(name) + 2);
 735         if (!path)
 736                 return NULL;
 737         strcpy(path, dir);
 738         if (dir[strlen(dir) - 1] != '/')
 739                 strcat(path, "/");
 740         strcat(path, name);
 741         return path;
 742 }
 743
 744 static int traverse_directory(struct btrfs_trans_handle *trans,
 745                               struct btrfs_root *root, char *dir_name,
 746                               struct directory_name_entry *dir_head, int out_fd)
 747 {
 748         int ret = 0;
 749
 750         struct btrfs_inode_item cur_inode;
 751         struct btrfs_inode_item *inode_item;
 752         int count, i, dir_index_cnt;
 753         struct direct **files;
 754         struct stat st;
 755         struct directory_name_entry *dir_entry, *parent_dir_entry;
 756         struct direct *cur_file;
 757         ino_t parent_inum, cur_inum;
 758         ino_t highest_inum = 0;
 759         char *parent_dir_name;
 760         struct btrfs_path path;
 761         struct extent_buffer *leaf;
 762         struct btrfs_key root_dir_key;
 763         u64 root_dir_inode_size = 0;
 764
 765         /* Add list for source directory */
 766         dir_entry = malloc(sizeof(struct directory_name_entry));
 767         dir_entry->dir_name = dir_name;
 768         dir_entry->path = strdup(dir_name);
 769
 770         parent_inum = highest_inum + BTRFS_FIRST_FREE_OBJECTID;
 771         dir_entry->inum = parent_inum;
 772         list_add_tail(&dir_entry->list, &dir_head->list);
 773
 774         btrfs_init_path(&path);
 775
 776         root_dir_key.objectid = btrfs_root_dirid(&root->root_item);
 777         root_dir_key.offset = 0;
 778         btrfs_set_key_type(&root_dir_key, BTRFS_INODE_ITEM_KEY);
 779         ret = btrfs_lookup_inode(trans, root, &path, &root_dir_key, 1);
 780         if (ret) {
 781                 fprintf(stderr, "root dir lookup error\n");
 782                 return -1;
 783         }
 784
 785         leaf = path.nodes[0];
 786         inode_item = btrfs_item_ptr(leaf, path.slots[0],
 787                                     struct btrfs_inode_item);
 788
 789         root_dir_inode_size = calculate_dir_inode_size(dir_name);
 790         btrfs_set_inode_size(leaf, inode_item, root_dir_inode_size);
 791         btrfs_mark_buffer_dirty(leaf);
 792
 793         btrfs_release_path(&path);
 794
 795         do {
 796                 parent_dir_entry = list_entry(dir_head->list.next,
 797                                               struct directory_name_entry,
 798                                               list);
 799                 list_del(&parent_dir_entry->list);
 800
 801                 parent_inum = parent_dir_entry->inum;
 802                 parent_dir_name = parent_dir_entry->dir_name;
 803                 if (chdir(parent_dir_entry->path)) {
 804                         fprintf(stderr, "chdir error for %s\n",
 805                                 parent_dir_name);
 806                         goto fail_no_files;
 807                 }
 808
 809                 count = scandir(parent_dir_entry->path, &files,
 810                                 directory_select, NULL);
 811                 if (count == -1)
 812                 {
 813                         fprintf(stderr, "scandir for %s failed: %s\n",
 814                                 parent_dir_name, strerror (errno));
 815                         goto fail;
 816                 }
 817
 818                 for (i = 0; i < count; i++) {
 819                         cur_file = files[i];
 820
 821                         if (lstat(cur_file->d_name, &st) == -1) {
 822                                 fprintf(stderr, "lstat failed for file %s\n",
 823                                         cur_file->d_name);
 824                                 goto fail;
 825                         }
 826
 827                         cur_inum = ++highest_inum + BTRFS_FIRST_FREE_OBJECTID;
 828                         ret = add_directory_items(trans, root,
 829                                                   cur_inum, parent_inum,
 830                                                   cur_file->d_name,
 831                                                   &st, &dir_index_cnt);
 832                         if (ret) {
 833                                 fprintf(stderr, "add_directory_items failed\n");
 834                                 goto fail;
 835                         }
 836
 837                         ret = add_inode_items(trans, root, &st,
 838                                               cur_file->d_name, cur_inum,
 839                                               parent_inum, dir_index_cnt,
 840                                               &cur_inode);
 841                         if (ret) {
 842                                 fprintf(stderr, "add_inode_items failed\n");
 843                                 goto fail;
 844                         }
 845
 846                         ret = add_xattr_item(trans, root,
 847                                              cur_inum, cur_file->d_name);
 848                         if (ret) {
 849                                 fprintf(stderr, "add_xattr_item failed\n");
 850                                 if(ret != -ENOTSUP)
 851                                         goto fail;
 852                         }
 853
 854                         if (S_ISDIR(st.st_mode)) {
 855                                 dir_entry = malloc(sizeof(struct directory_name_entry));
 856                                 dir_entry->dir_name = cur_file->d_name;
 857                                 dir_entry->path = make_path(parent_dir_entry->path,
 858                                                             cur_file->d_name);
 859                                 dir_entry->inum = cur_inum;
 860                                 list_add_tail(&dir_entry->list, &dir_head->list);
 861                         } else if (S_ISREG(st.st_mode)) {
 862                                 ret = add_file_items(trans, root, &cur_inode,
 863                                                      cur_inum, parent_inum, &st,
 864                                                      cur_file->d_name, out_fd);
 865                                 if (ret) {
 866                                         fprintf(stderr, "add_file_items failed\n");
 867                                         goto fail;
 868                                 }
 869                         } else if (S_ISLNK(st.st_mode)) {
 870                                 ret = add_symbolic_link(trans, root,
 871                                                         cur_inum, cur_file->d_name);
 872                                 if (ret) {
 873                                         fprintf(stderr, "add_symbolic_link failed\n");
 874                                         goto fail;
 875                                 }
 876                         }
 877                 }
 878
 879                 free_namelist(files, count);
 880                 free(parent_dir_entry->path);
 881                 free(parent_dir_entry);
 882
 883                 index_cnt = 2;
 884
 885         } while (!list_empty(&dir_head->list));
 886
 887         return 0;
 888 fail:
 889         free_namelist(files, count);
 890 fail_no_files:
 891         free(parent_dir_entry->path);
 892         free(parent_dir_entry);
 893         return -1;
 894 }
 895
 896 static int open_target(char *output_name)
 897 {
 898         int output_fd;
 899         output_fd = open(output_name, O_CREAT | O_RDWR | O_TRUNC,
 900                          S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH);
 901
 902         return output_fd;
 903 }
 904
 905 static int create_chunks(struct btrfs_trans_handle *trans,
 906                          struct btrfs_root *root, u64 num_of_meta_chunks,
 907                          u64 size_of_data)
 908 {
 909         u64 chunk_start;
 910         u64 chunk_size;
 911         u64 meta_type = BTRFS_BLOCK_GROUP_METADATA;
 912         u64 data_type = BTRFS_BLOCK_GROUP_DATA;
 913         u64 minimum_data_chunk_size = 8 * 1024 * 1024;
 914         u64 i;
 915         int ret;
 916
 917         for (i = 0; i < num_of_meta_chunks; i++) {
 918                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 919                                         &chunk_start, &chunk_size, meta_type);
 920                 BUG_ON(ret);
 921                 ret = btrfs_make_block_group(trans, root->fs_info->extent_root, 0,
 922                                              meta_type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 923                                              chunk_start, chunk_size);
 924                 BUG_ON(ret);
 925                 set_extent_dirty(&root->fs_info->free_space_cache,
 926                                  chunk_start, chunk_start + chunk_size - 1, 0);
 927         }
 928
 929         if (size_of_data < minimum_data_chunk_size)
 930                 size_of_data = minimum_data_chunk_size;
 931
 932         ret = btrfs_alloc_data_chunk(trans, root->fs_info->extent_root,
 933                                      &chunk_start, size_of_data, data_type);
 934         BUG_ON(ret);
 935         ret = btrfs_make_block_group(trans, root->fs_info->extent_root, 0,
 936                                      data_type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 937                                      chunk_start, size_of_data);
 938         BUG_ON(ret);
 939         set_extent_dirty(&root->fs_info->free_space_cache,
 940                          chunk_start, chunk_start + size_of_data - 1, 0);
 941         return ret;
 942 }
 943
 944 static int make_image(char *source_dir, struct btrfs_root *root, int out_fd)
 945 {
 946         int ret;
 947         struct btrfs_trans_handle *trans;
 948
 949         struct stat root_st;
 950
 951         struct directory_name_entry dir_head;
 952
 953         struct directory_name_entry *dir_entry = NULL;
 954
 955         ret = lstat(source_dir, &root_st);
 956         if (ret) {
 957                 fprintf(stderr, "unable to lstat the %s\n", source_dir);
 958                 goto fail;
 959         }
 960
 961         INIT_LIST_HEAD(&dir_head.list);
 962
 963         trans = btrfs_start_transaction(root, 1);
 964         ret = traverse_directory(trans, root, source_dir, &dir_head, out_fd);
 965         if (ret) {
 966                 fprintf(stderr, "unable to traverse_directory\n");
 967                 goto fail;
 968         }
 969         btrfs_commit_transaction(trans, root);
 970
 971         printf("Making image is completed.\n");
 972         return 0;
 973 fail:
 974         while (!list_empty(&dir_head.list)) {
 975                 dir_entry = list_entry(dir_head.list.next,
 976                                        struct directory_name_entry, list);
 977                 list_del(&dir_entry->list);
 978                 free(dir_entry);
 979         }
 980         fprintf(stderr, "Making image is aborted.\n");
 981         return -1;
 982 }
 983
 984 /*
 985  * This ignores symlinks with unreadable targets and subdirs that can't
 986  * be read.  It's a best-effort to give a rough estimate of the size of
 987  * a subdir.  It doesn't guarantee that prepopulating btrfs from this
 988  * tree won't still run out of space.
 989  *
 990  * The rounding up to 4096 is questionable.  Previous code used du -B 4096.
 991  */
 992 static u64 global_total_size;
 993 static int ftw_add_entry_size(const char *fpath, const struct stat *st,
 994                               int type)
 995 {
 996         if (type == FTW_F || type == FTW_D)
 997                 global_total_size += round_up(st->st_size, 4096);
 998
 999         return 0;
1000 }
1001
1002 static u64 size_sourcedir(char *dir_name, u64 sectorsize,
1003                           u64 *num_of_meta_chunks_ret, u64 *size_of_data_ret)
1004 {
1005         u64 dir_size = 0;
1006         u64 total_size = 0;
1007         int ret;
1008         u64 default_chunk_size = 8 * 1024 * 1024;       /* 8MB */
1009         u64 allocated_meta_size = 8 * 1024 * 1024;      /* 8MB */
1010         u64 allocated_total_size = 20 * 1024 * 1024;    /* 20MB */
1011         u64 num_of_meta_chunks = 0;
1012         u64 num_of_data_chunks = 0;
1013         u64 num_of_allocated_meta_chunks =
1014                         allocated_meta_size / default_chunk_size;
1015
1016         global_total_size = 0;
1017         ret = ftw(dir_name, ftw_add_entry_size, 10);
1018         dir_size = global_total_size;
1019         if (ret < 0) {
1020                 fprintf(stderr, "ftw subdir walk of '%s' failed: %s\n",
1021                         dir_name, strerror(errno));
1022                 exit(1);
1023         }
1024
1025         num_of_data_chunks = (dir_size + default_chunk_size - 1) /
1026                 default_chunk_size;
1027
1028         num_of_meta_chunks = (dir_size / 2) / default_chunk_size;
1029         if (((dir_size / 2) % default_chunk_size) != 0)
1030                 num_of_meta_chunks++;
1031         if (num_of_meta_chunks <= num_of_allocated_meta_chunks)
1032                 num_of_meta_chunks = 0;
1033         else
1034                 num_of_meta_chunks -= num_of_allocated_meta_chunks;
1035
1036         total_size = allocated_total_size +
1037                      (num_of_data_chunks * default_chunk_size) +
1038                      (num_of_meta_chunks * default_chunk_size);
1039
1040         *num_of_meta_chunks_ret = num_of_meta_chunks;
1041         *size_of_data_ret = num_of_data_chunks * default_chunk_size;
1042         return total_size;
1043 }
1044
1045 static int zero_output_file(int out_fd, u64 size, u32 sectorsize)
1046 {
1047         int len = sectorsize;
1048         int loop_num = size / sectorsize;
1049         u64 location = 0;
1050         char *buf = malloc(len);
1051         int ret = 0, i;
1052         ssize_t written;
1053
1054         if (!buf)
1055                 return -ENOMEM;
1056         memset(buf, 0, len);
1057         for (i = 0; i < loop_num; i++) {
1058                 written = pwrite64(out_fd, buf, len, location);
1059                 if (written != len)
1060                         ret = -EIO;
1061                 location += sectorsize;
1062         }
1063         free(buf);
1064         return ret;
1065 }
1066
1067 static int check_leaf_or_node_size(u32 size, u32 sectorsize)
1068 {
1069         if (size < sectorsize) {
1070                 fprintf(stderr,
1071                         "Illegal leafsize (or nodesize) %u (smaller than %u)\n",
1072                         size, sectorsize);
1073                 return -1;
1074         } else if (size > BTRFS_MAX_METADATA_BLOCKSIZE) {
1075                 fprintf(stderr,
1076                         "Illegal leafsize (or nodesize) %u (larger than %u)\n",
1077                         size, BTRFS_MAX_METADATA_BLOCKSIZE);
1078                 return -1;
1079         } else if (size & (sectorsize - 1)) {
1080                 fprintf(stderr,
1081                         "Illegal leafsize (or nodesize) %u (not align to %u)\n",
1082                         size, sectorsize);
1083                 return -1;
1084         }
1085         return 0;
1086 }
1087
1088 static int is_ssd(const char *file)
1089 {
1090         blkid_probe probe;
1091         char wholedisk[32];
1092         char sysfs_path[PATH_MAX];
1093         dev_t devno;
1094         int fd;
1095         char rotational;
1096         int ret;
1097
1098         probe = blkid_new_probe_from_filename(file);
1099         if (!probe)
1100                 return 0;
1101
1102         /* Device number of this disk (possibly a partition) */
1103         devno = blkid_probe_get_devno(probe);
1104         if (!devno) {
1105                 blkid_free_probe(probe);
1106                 return 0;
1107         }
1108
1109         /* Get whole disk name (not full path) for this devno */
1110         ret = blkid_devno_to_wholedisk(devno,
1111                         wholedisk, sizeof(wholedisk), NULL);
1112         if (ret) {
1113                 blkid_free_probe(probe);
1114                 return 0;
1115         }
1116
1117         snprintf(sysfs_path, PATH_MAX, "/sys/block/%s/queue/rotational",
1118                  wholedisk);
1119
1120         blkid_free_probe(probe);
1121
1122         fd = open(sysfs_path, O_RDONLY);
1123         if (fd < 0) {
1124                 return 0;
1125         }
1126
1127         if (read(fd, &rotational, sizeof(char)) < sizeof(char)) {
1128                 close(fd);
1129                 return 0;
1130         }
1131         close(fd);
1132
1133         return !atoi((const char *)&rotational);
1134 }
1135
1136 #define BTRFS_FEATURE_LIST_ALL          (1ULL << 63)
1137
1138 static const struct btrfs_fs_feature {
1139         const char *name;
1140         u64 flag;
1141         const char *desc;
1142 } mkfs_features[] = {
1143         { "mixed-bg", BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS,
1144                 "mixed data and metadata block groups" },
1145         { "extref", BTRFS_FEATURE_INCOMPAT_EXTENDED_IREF,
1146                 "increased hardlink limit per file to 65536" },
1147         { "raid56", BTRFS_FEATURE_INCOMPAT_RAID56,
1148                 "raid56 extended format" },
1149         { "skinny-metadata", BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA,
1150                 "reduced-size metadata extent refs" },
1151         /* Keep this one last */
1152         { "list-all", BTRFS_FEATURE_LIST_ALL, NULL }
1153 };
1154
1155 static void list_all_fs_features(void)
1156 {
1157         int i;
1158
1159         fprintf(stderr, "Filesystem features available at mkfs time:\n");
1160         for (i = 0; i < ARRAY_SIZE(mkfs_features) - 1; i++) {
1161                 char *is_default = "";
1162
1163                 if (mkfs_features[i].flag & DEFAULT_MKFS_FEATURES)
1164                         is_default = ", default";
1165                 fprintf(stderr, "%-20s- %s (0x%llx%s)\n",
1166                                 mkfs_features[i].name,
1167                                 mkfs_features[i].desc,
1168                                 mkfs_features[i].flag,
1169                                 is_default);
1170         }
1171 }
1172
1173 static int parse_one_fs_feature(const char *name, u64 *flags)
1174 {
1175         int i;
1176         int found = 0;
1177
1178         for (i = 0; i < ARRAY_SIZE(mkfs_features); i++) {
1179                 if (name[0] == '^' &&
1180                         !strcmp(mkfs_features[i].name, name + 1)) {
1181                         *flags &= ~ mkfs_features[i].flag;
1182                         found = 1;
1183                 } else if (!strcmp(mkfs_features[i].name, name)) {
1184                         *flags |= mkfs_features[i].flag;
1185                         found = 1;
1186                 }
1187         }
1188
1189         return !found;
1190 }
1191
1192 static void process_fs_features(u64 flags)
1193 {
1194         int i;
1195
1196         for (i = 0; i < ARRAY_SIZE(mkfs_features); i++) {
1197                 if (flags & mkfs_features[i].flag) {
1198                         printf("Turning ON incompat feature '%s': %s\n",
1199                                 mkfs_features[i].name,
1200                                 mkfs_features[i].desc);
1201                 }
1202         }
1203 }
1204
1205
1206 /*
1207  * Return NULL if all features were parsed fine, otherwise return the name of
1208  * the first unparsed.
1209  */
1210 static char* parse_fs_features(char *namelist, u64 *flags)
1211 {
1212         char *this_char;
1213         char *save_ptr = NULL; /* Satisfy static checkers */
1214
1215         for (this_char = strtok_r(namelist, ",", &save_ptr);
1216              this_char != NULL;
1217              this_char = strtok_r(NULL, ",", &save_ptr)) {
1218                 if (parse_one_fs_feature(this_char, flags))
1219                         return this_char;
1220         }
1221
1222         return NULL;
1223 }
1224
1225 int main(int ac, char **av)
1226 {
1227         char *file;
1228         struct btrfs_root *root;
1229         struct btrfs_trans_handle *trans;
1230         char *label = NULL;
1231         char *first_file;
1232         u64 block_count = 0;
1233         u64 dev_block_count = 0;
1234         u64 blocks[7];
1235         u64 alloc_start = 0;
1236         u64 metadata_profile = 0;
1237         u64 data_profile = 0;
1238         u32 leafsize = max_t(u32, sysconf(_SC_PAGESIZE), DEFAULT_MKFS_LEAF_SIZE);
1239         u32 sectorsize = 4096;
1240         u32 nodesize = leafsize;
1241         u32 stripesize = 4096;
1242         int zero_end = 1;
1243         int option_index = 0;
1244         int fd;
1245         int ret;
1246         int i;
1247         int mixed = 0;
1248         int leaf_forced = 0;
1249         int data_profile_opt = 0;
1250         int metadata_profile_opt = 0;
1251         int discard = 1;
1252         int ssd = 0;
1253         int force_overwrite = 0;
1254
1255         char *source_dir = NULL;
1256         int source_dir_set = 0;
1257         u64 num_of_meta_chunks = 0;
1258         u64 size_of_data = 0;
1259         u64 source_dir_size = 0;
1260         int dev_cnt = 0;
1261         int saved_optind;
1262         char estr[100];
1263         u64 features = DEFAULT_MKFS_FEATURES;
1264
1265         while(1) {
1266                 int c;
1267                 c = getopt_long(ac, av, "A:b:fl:n:s:m:d:L:O:r:VMK",
1268                                 long_options, &option_index);
1269                 if (c < 0)
1270                         break;
1271                 switch(c) {
1272                         case 'A':
1273                                 alloc_start = parse_size(optarg);
1274                                 break;
1275                         case 'f':
1276                                 force_overwrite = 1;
1277                                 break;
1278                         case 'd':
1279                                 data_profile = parse_profile(optarg);
1280                                 data_profile_opt = 1;
1281                                 break;
1282                         case 'l':
1283                         case 'n':
1284                                 nodesize = parse_size(optarg);
1285                                 leafsize = parse_size(optarg);
1286                                 leaf_forced = 1;
1287                                 break;
1288                         case 'L':
1289                                 label = parse_label(optarg);
1290                                 break;
1291                         case 'm':
1292                                 metadata_profile = parse_profile(optarg);
1293                                 metadata_profile_opt = 1;
1294                                 break;
1295                         case 'M':
1296                                 mixed = 1;
1297                                 break;
1298                         case 'O': {
1299                                 char *orig = strdup(optarg);
1300                                 char *tmp = orig;
1301
1302                                 tmp = parse_fs_features(tmp, &features);
1303                                 if (tmp) {
1304                                         fprintf(stderr,
1305                                                 "Unrecognized filesystem feature '%s'\n",
1306                                                         tmp);
1307                                         free(orig);
1308                                         exit(1);
1309                                 }
1310                                 free(orig);
1311                                 if (features & BTRFS_FEATURE_LIST_ALL) {
1312                                         list_all_fs_features();
1313                                         exit(0);
1314                                 }
1315                                 break;
1316                                 }
1317                         case 's':
1318                                 sectorsize = parse_size(optarg);
1319                                 break;
1320                         case 'b':
1321                                 block_count = parse_size(optarg);
1322                                 if (block_count <= 1024*1024*1024) {
1323                                         printf("SMALL VOLUME: forcing mixed "
1324                                                "metadata/data groups\n");
1325                                         mixed = 1;
1326                                 }
1327                                 zero_end = 0;
1328                                 break;
1329                         case 'V':
1330                                 print_version();
1331                                 break;
1332                         case 'r':
1333                                 source_dir = optarg;
1334                                 source_dir_set = 1;
1335                                 break;
1336                         case 'K':
1337                                 discard = 0;
1338                                 break;
1339                         default:
1340                                 print_usage();
1341                 }
1342         }
1343         sectorsize = max(sectorsize, (u32)sysconf(_SC_PAGESIZE));
1344         if (check_leaf_or_node_size(leafsize, sectorsize))
1345                 exit(1);
1346         if (check_leaf_or_node_size(nodesize, sectorsize))
1347                 exit(1);
1348         saved_optind = optind;
1349         dev_cnt = ac - optind;
1350         if (dev_cnt == 0)
1351                 print_usage();
1352
1353         if (source_dir_set && dev_cnt > 1) {
1354                 fprintf(stderr,
1355                         "The -r option is limited to a single device\n");
1356                 exit(1);
1357         }
1358         while (dev_cnt-- > 0) {
1359                 file = av[optind++];
1360                 if (is_block_device(file))
1361                         if (test_dev_for_mkfs(file, force_overwrite, estr)) {
1362                                 fprintf(stderr, "Error: %s", estr);
1363                                 exit(1);
1364                         }
1365         }
1366
1367         optind = saved_optind;
1368         dev_cnt = ac - optind;
1369
1370         file = av[optind++];
1371         ssd = is_ssd(file);
1372
1373         if (is_vol_small(file)) {
1374                 printf("SMALL VOLUME: forcing mixed metadata/data groups\n");
1375                 mixed = 1;
1376         }
1377
1378         /*
1379         * Set default profiles according to number of added devices.
1380         * For mixed groups defaults are single/single.
1381         */
1382         if (!mixed) {
1383                 if (!metadata_profile_opt) {
1384                         if (dev_cnt == 1 && ssd)
1385                                 printf("Detected a SSD, turning off metadata "
1386                                 "duplication.  Mkfs with -m dup if you want to "
1387                                 "force metadata duplication.\n");
1388
1389                         metadata_profile = (dev_cnt > 1) ?
1390                                         BTRFS_BLOCK_GROUP_RAID1 : (ssd) ?
1391                                         0: BTRFS_BLOCK_GROUP_DUP;
1392                 }
1393                 if (!data_profile_opt) {
1394                         data_profile = (dev_cnt > 1) ?
1395                                 BTRFS_BLOCK_GROUP_RAID0 : 0; /* raid0 or single */
1396                 }
1397         } else {
1398                 u32 best_leafsize = max_t(u32, sysconf(_SC_PAGESIZE), sectorsize);
1399
1400                 if (metadata_profile_opt || data_profile_opt) {
1401                         if (metadata_profile != data_profile) {
1402                                 fprintf(stderr,
1403         "ERROR: With mixed block groups data and metadata profiles must be the same\n");
1404                                 exit(1);
1405                         }
1406                 }
1407
1408                 if (!leaf_forced) {
1409                         leafsize = best_leafsize;
1410                         nodesize = best_leafsize;
1411                         if (check_leaf_or_node_size(leafsize, sectorsize))
1412                                 exit(1);
1413                 }
1414                 if (leafsize != sectorsize) {
1415                         fprintf(stderr, "Error: mixed metadata/data block groups "
1416                                 "require metadata blocksizes equal to the sectorsize\n");
1417                         exit(1);
1418                 }
1419         }
1420
1421         ret = test_num_disk_vs_raid(metadata_profile, data_profile,
1422                         dev_cnt, mixed, estr);
1423         if (ret) {
1424                 fprintf(stderr, "Error: %s\n", estr);
1425                 exit(1);
1426         }
1427
1428         /* if we are here that means all devs are good to btrfsify */
1429         printf("\nWARNING! - %s IS EXPERIMENTAL\n", BTRFS_BUILD_VERSION);
1430         printf("WARNING! - see http://btrfs.wiki.kernel.org before using\n\n");
1431
1432         dev_cnt--;
1433
1434         if (!source_dir_set) {
1435                 /*
1436                  * open without O_EXCL so that the problem should not
1437                  * occur by the following processing.
1438                  * (btrfs_register_one_device() fails if O_EXCL is on)
1439                  */
1440                 fd = open(file, O_RDWR);
1441                 if (fd < 0) {
1442                         fprintf(stderr, "unable to open %s: %s\n", file,
1443                                 strerror(errno));
1444                         exit(1);
1445                 }
1446                 first_file = file;
1447                 ret = btrfs_prepare_device(fd, file, zero_end, &dev_block_count,
1448                                            block_count, &mixed, discard);
1449                 if (block_count && block_count > dev_block_count) {
1450                         fprintf(stderr, "%s is smaller than requested size\n", file);
1451                         exit(1);
1452                 }
1453         } else {
1454                 fd = open_target(file);
1455                 if (fd < 0) {
1456                         fprintf(stderr, "unable to open the %s\n", file);
1457                         exit(1);
1458                 }
1459
1460                 first_file = file;
1461                 source_dir_size = size_sourcedir(source_dir, sectorsize,
1462                                              &num_of_meta_chunks, &size_of_data);
1463                 if(block_count < source_dir_size)
1464                         block_count = source_dir_size;
1465                 ret = zero_output_file(fd, block_count, sectorsize);
1466                 if (ret) {
1467                         fprintf(stderr, "unable to zero the output file\n");
1468                         exit(1);
1469                 }
1470                 /* our "device" is the new image file */
1471                 dev_block_count = block_count;
1472         }
1473
1474         /* To create the first block group and chunk 0 in make_btrfs */
1475         if (dev_block_count < BTRFS_MKFS_SYSTEM_GROUP_SIZE) {
1476                 fprintf(stderr, "device is too small to make filesystem\n");
1477                 exit(1);
1478         }
1479
1480         blocks[0] = BTRFS_SUPER_INFO_OFFSET;
1481         for (i = 1; i < 7; i++) {
1482                 blocks[i] = BTRFS_SUPER_INFO_OFFSET + 1024 * 1024 +
1483                         leafsize * i;
1484         }
1485
1486         /*
1487          * FS features that can be set by other means than -O
1488          * just set the bit here
1489          */
1490         if (mixed)
1491                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS;
1492
1493         if ((data_profile | metadata_profile) &
1494             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
1495                 features |= BTRFS_FEATURE_INCOMPAT_RAID56;
1496         }
1497
1498         process_fs_features(features);
1499
1500         ret = make_btrfs(fd, file, label, blocks, dev_block_count,
1501                          nodesize, leafsize,
1502                          sectorsize, stripesize, features);
1503         if (ret) {
1504                 fprintf(stderr, "error during mkfs: %s\n", strerror(-ret));
1505                 exit(1);
1506         }
1507
1508         root = open_ctree(file, 0, OPEN_CTREE_WRITES);
1509         if (!root) {
1510                 fprintf(stderr, "Open ctree failed\n");
1511                 close(fd);
1512                 exit(1);
1513         }
1514         root->fs_info->alloc_start = alloc_start;
1515
1516         ret = make_root_dir(root, mixed);
1517         if (ret) {
1518                 fprintf(stderr, "failed to setup the root directory\n");
1519                 exit(1);
1520         }
1521
1522         trans = btrfs_start_transaction(root, 1);
1523
1524         if (dev_cnt == 0)
1525                 goto raid_groups;
1526
1527         btrfs_register_one_device(file);
1528
1529         zero_end = 1;
1530         while (dev_cnt-- > 0) {
1531                 int old_mixed = mixed;
1532
1533                 file = av[optind++];
1534
1535                 /*
1536                  * open without O_EXCL so that the problem should not
1537                  * occur by the following processing.
1538                  * (btrfs_register_one_device() fails if O_EXCL is on)
1539                  */
1540                 fd = open(file, O_RDWR);
1541                 if (fd < 0) {
1542                         fprintf(stderr, "unable to open %s: %s\n", file,
1543                                 strerror(errno));
1544                         exit(1);
1545                 }
1546                 ret = btrfs_device_already_in_root(root, fd,
1547                                                    BTRFS_SUPER_INFO_OFFSET);
1548                 if (ret) {
1549                         fprintf(stderr, "skipping duplicate device %s in FS\n",
1550                                 file);
1551                         close(fd);
1552                         continue;
1553                 }
1554                 ret = btrfs_prepare_device(fd, file, zero_end, &dev_block_count,
1555                                            block_count, &mixed, discard);
1556                 mixed = old_mixed;
1557                 BUG_ON(ret);
1558
1559                 ret = btrfs_add_to_fsid(trans, root, fd, file, dev_block_count,
1560                                         sectorsize, sectorsize, sectorsize);
1561                 BUG_ON(ret);
1562                 btrfs_register_one_device(file);
1563         }
1564
1565 raid_groups:
1566         if (!source_dir_set) {
1567                 ret = create_raid_groups(trans, root, data_profile,
1568                                  data_profile_opt, metadata_profile,
1569                                  metadata_profile_opt, mixed, ssd);
1570                 BUG_ON(ret);
1571         }
1572
1573         ret = create_data_reloc_tree(trans, root);
1574         BUG_ON(ret);
1575
1576         printf("fs created label %s on %s\n\tnodesize %u leafsize %u "
1577             "sectorsize %u size %s\n",
1578             label, first_file, nodesize, leafsize, sectorsize,
1579             pretty_size(btrfs_super_total_bytes(root->fs_info->super_copy)));
1580
1581         printf("%s\n", BTRFS_BUILD_VERSION);
1582         btrfs_commit_transaction(trans, root);
1583
1584         if (source_dir_set) {
1585                 trans = btrfs_start_transaction(root, 1);
1586                 ret = create_chunks(trans, root,
1587                                     num_of_meta_chunks, size_of_data);
1588                 BUG_ON(ret);
1589                 btrfs_commit_transaction(trans, root);
1590
1591                 ret = make_image(source_dir, root, fd);
1592                 BUG_ON(ret);
1593         }
1594
1595         ret = close_ctree(root);
1596         BUG_ON(ret);
1597         free(label);
1598         return 0;
1599 }