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 struct directory_name_entry {
  50         char *dir_name;
  51         char *path;
  52         ino_t inum;
  53         struct list_head list;
  54 };
  55
  56 static int make_root_dir(struct btrfs_root *root, int mixed)
  57 {
  58         struct btrfs_trans_handle *trans;
  59         struct btrfs_key location;
  60         u64 bytes_used;
  61         u64 chunk_start = 0;
  62         u64 chunk_size = 0;
  63         int ret;
  64
  65         trans = btrfs_start_transaction(root, 1);
  66         bytes_used = btrfs_super_bytes_used(root->fs_info->super_copy);
  67
  68         root->fs_info->system_allocs = 1;
  69         ret = btrfs_make_block_group(trans, root, bytes_used,
  70                                      BTRFS_BLOCK_GROUP_SYSTEM,
  71                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
  72                                      0, BTRFS_MKFS_SYSTEM_GROUP_SIZE);
  73         BUG_ON(ret);
  74
  75         if (mixed) {
  76                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
  77                                         &chunk_start, &chunk_size,
  78                                         BTRFS_BLOCK_GROUP_METADATA |
  79                                         BTRFS_BLOCK_GROUP_DATA);
  80                 if (ret == -ENOSPC) {
  81                         fprintf(stderr,
  82                                 "no space to alloc data/metadata chunk\n");
  83                         goto err;
  84                 }
  85                 BUG_ON(ret);
  86                 ret = btrfs_make_block_group(trans, root, 0,
  87                                              BTRFS_BLOCK_GROUP_METADATA |
  88                                              BTRFS_BLOCK_GROUP_DATA,
  89                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
  90                                              chunk_start, chunk_size);
  91                 BUG_ON(ret);
  92                 printf("Created a data/metadata chunk of size %llu\n", chunk_size);
  93         } else {
  94                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
  95                                         &chunk_start, &chunk_size,
  96                                         BTRFS_BLOCK_GROUP_METADATA);
  97                 if (ret == -ENOSPC) {
  98                         fprintf(stderr, "no space to alloc metadata chunk\n");
  99                         goto err;
 100                 }
 101                 BUG_ON(ret);
 102                 ret = btrfs_make_block_group(trans, root, 0,
 103                                              BTRFS_BLOCK_GROUP_METADATA,
 104                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 105                                              chunk_start, chunk_size);
 106                 BUG_ON(ret);
 107         }
 108
 109         root->fs_info->system_allocs = 0;
 110         btrfs_commit_transaction(trans, root);
 111         trans = btrfs_start_transaction(root, 1);
 112         BUG_ON(!trans);
 113
 114         if (!mixed) {
 115                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 116                                         &chunk_start, &chunk_size,
 117                                         BTRFS_BLOCK_GROUP_DATA);
 118                 if (ret == -ENOSPC) {
 119                         fprintf(stderr, "no space to alloc data chunk\n");
 120                         goto err;
 121                 }
 122                 BUG_ON(ret);
 123                 ret = btrfs_make_block_group(trans, root, 0,
 124                                              BTRFS_BLOCK_GROUP_DATA,
 125                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 126                                              chunk_start, chunk_size);
 127                 BUG_ON(ret);
 128         }
 129
 130         ret = btrfs_make_root_dir(trans, root->fs_info->tree_root,
 131                               BTRFS_ROOT_TREE_DIR_OBJECTID);
 132         if (ret)
 133                 goto err;
 134         ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
 135         if (ret)
 136                 goto err;
 137         memcpy(&location, &root->fs_info->fs_root->root_key, sizeof(location));
 138         location.offset = (u64)-1;
 139         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
 140                         "default", 7,
 141                         btrfs_super_root_dir(root->fs_info->super_copy),
 142                         &location, BTRFS_FT_DIR, 0);
 143         if (ret)
 144                 goto err;
 145
 146         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 147                              "default", 7, location.objectid,
 148                              BTRFS_ROOT_TREE_DIR_OBJECTID, 0);
 149         if (ret)
 150                 goto err;
 151
 152         btrfs_commit_transaction(trans, root);
 153 err:
 154         return ret;
 155 }
 156
 157 static void __recow_root(struct btrfs_trans_handle *trans,
 158                          struct btrfs_root *root)
 159 {
 160         int ret;
 161         struct extent_buffer *tmp;
 162
 163         if (trans->transid != btrfs_root_generation(&root->root_item)) {
 164                 ret = __btrfs_cow_block(trans, root, root->node,
 165                                         NULL, 0, &tmp, 0, 0);
 166                 BUG_ON(ret);
 167                 free_extent_buffer(tmp);
 168         }
 169 }
 170
 171 static void recow_roots(struct btrfs_trans_handle *trans,
 172                        struct btrfs_root *root)
 173 {
 174         struct btrfs_fs_info *info = root->fs_info;
 175
 176         __recow_root(trans, info->fs_root);
 177         __recow_root(trans, info->tree_root);
 178         __recow_root(trans, info->extent_root);
 179         __recow_root(trans, info->chunk_root);
 180         __recow_root(trans, info->dev_root);
 181         __recow_root(trans, info->csum_root);
 182 }
 183
 184 static int create_one_raid_group(struct btrfs_trans_handle *trans,
 185                               struct btrfs_root *root, u64 type)
 186 {
 187         u64 chunk_start;
 188         u64 chunk_size;
 189         int ret;
 190
 191         ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 192                                 &chunk_start, &chunk_size, type);
 193         if (ret == -ENOSPC) {
 194                 fprintf(stderr, "not enough free space\n");
 195                 exit(1);
 196         }
 197         BUG_ON(ret);
 198         ret = btrfs_make_block_group(trans, root->fs_info->extent_root, 0,
 199                                      type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 200                                      chunk_start, chunk_size);
 201         BUG_ON(ret);
 202         return ret;
 203 }
 204
 205 static int create_raid_groups(struct btrfs_trans_handle *trans,
 206                               struct btrfs_root *root, u64 data_profile,
 207                               int data_profile_opt, u64 metadata_profile,
 208                               int metadata_profile_opt, int mixed, int ssd)
 209 {
 210         u64 num_devices = btrfs_super_num_devices(root->fs_info->super_copy);
 211         int ret;
 212
 213         if (metadata_profile) {
 214                 u64 meta_flags = BTRFS_BLOCK_GROUP_METADATA;
 215
 216                 ret = create_one_raid_group(trans, root,
 217                                             BTRFS_BLOCK_GROUP_SYSTEM |
 218                                             metadata_profile);
 219                 BUG_ON(ret);
 220
 221                 if (mixed)
 222                         meta_flags |= BTRFS_BLOCK_GROUP_DATA;
 223
 224                 ret = create_one_raid_group(trans, root, meta_flags |
 225                                             metadata_profile);
 226                 BUG_ON(ret);
 227
 228         }
 229         if (!mixed && num_devices > 1 && data_profile) {
 230                 ret = create_one_raid_group(trans, root,
 231                                             BTRFS_BLOCK_GROUP_DATA |
 232                                             data_profile);
 233                 BUG_ON(ret);
 234         }
 235         recow_roots(trans, root);
 236
 237         return 0;
 238 }
 239
 240 static int create_data_reloc_tree(struct btrfs_trans_handle *trans,
 241                                   struct btrfs_root *root)
 242 {
 243         struct btrfs_key location;
 244         struct btrfs_root_item root_item;
 245         struct extent_buffer *tmp;
 246         u64 objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
 247         int ret;
 248
 249         ret = btrfs_copy_root(trans, root, root->node, &tmp, objectid);
 250         BUG_ON(ret);
 251
 252         memcpy(&root_item, &root->root_item, sizeof(root_item));
 253         btrfs_set_root_bytenr(&root_item, tmp->start);
 254         btrfs_set_root_level(&root_item, btrfs_header_level(tmp));
 255         btrfs_set_root_generation(&root_item, trans->transid);
 256         free_extent_buffer(tmp);
 257
 258         location.objectid = objectid;
 259         location.type = BTRFS_ROOT_ITEM_KEY;
 260         location.offset = 0;
 261         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
 262                                 &location, &root_item);
 263         BUG_ON(ret);
 264         return 0;
 265 }
 266
 267 static void print_usage(void)
 268 {
 269         fprintf(stderr, "usage: mkfs.btrfs [options] dev [ dev ... ]\n");
 270         fprintf(stderr, "options:\n");
 271         fprintf(stderr, "\t -A --alloc-start the offset to start the FS\n");
 272         fprintf(stderr, "\t -b --byte-count total number of bytes in the FS\n");
 273         fprintf(stderr, "\t -d --data data profile, raid0, raid1, raid5, raid6, raid10, dup or single\n");
 274         fprintf(stderr, "\t -f --force force overwrite of existing filesystem\n");
 275         fprintf(stderr, "\t -l --leafsize size of btree leaves\n");
 276         fprintf(stderr, "\t -L --label set a label\n");
 277         fprintf(stderr, "\t -m --metadata metadata profile, values like data profile\n");
 278         fprintf(stderr, "\t -M --mixed mix metadata and data together\n");
 279         fprintf(stderr, "\t -n --nodesize size of btree nodes\n");
 280         fprintf(stderr, "\t -s --sectorsize min block allocation (may not mountable by current kernel)\n");
 281         fprintf(stderr, "\t -r --rootdir the source directory\n");
 282         fprintf(stderr, "\t -K --nodiscard do not perform whole device TRIM\n");
 283         fprintf(stderr, "\t -O --features comma separated list of filesystem features\n");
 284         fprintf(stderr, "\t -V --version print the mkfs.btrfs version and exit\n");
 285         fprintf(stderr, "%s\n", BTRFS_BUILD_VERSION);
 286         exit(1);
 287 }
 288
 289 static void print_version(void)
 290 {
 291         fprintf(stderr, "mkfs.btrfs, part of %s\n", BTRFS_BUILD_VERSION);
 292         exit(0);
 293 }
 294
 295 static u64 parse_profile(char *s)
 296 {
 297         if (strcmp(s, "raid0") == 0) {
 298                 return BTRFS_BLOCK_GROUP_RAID0;
 299         } else if (strcmp(s, "raid1") == 0) {
 300                 return BTRFS_BLOCK_GROUP_RAID1;
 301         } else if (strcmp(s, "raid5") == 0) {
 302                 return BTRFS_BLOCK_GROUP_RAID5;
 303         } else if (strcmp(s, "raid6") == 0) {
 304                 return BTRFS_BLOCK_GROUP_RAID6;
 305         } else if (strcmp(s, "raid10") == 0) {
 306                 return BTRFS_BLOCK_GROUP_RAID10;
 307         } else if (strcmp(s, "dup") == 0) {
 308                 return BTRFS_BLOCK_GROUP_DUP;
 309         } else if (strcmp(s, "single") == 0) {
 310                 return 0;
 311         } else {
 312                 fprintf(stderr, "Unknown profile %s\n", s);
 313                 print_usage();
 314         }
 315         /* not reached */
 316         return 0;
 317 }
 318
 319 static char *parse_label(char *input)
 320 {
 321         int len = strlen(input);
 322
 323         if (len >= BTRFS_LABEL_SIZE) {
 324                 fprintf(stderr, "Label %s is too long (max %d)\n", input,
 325                         BTRFS_LABEL_SIZE - 1);
 326                 exit(1);
 327         }
 328         return strdup(input);
 329 }
 330
 331 static struct option long_options[] = {
 332         { "alloc-start", 1, NULL, 'A'},
 333         { "byte-count", 1, NULL, 'b' },
 334         { "force", 0, NULL, 'f' },
 335         { "leafsize", 1, NULL, 'l' },
 336         { "label", 1, NULL, 'L'},
 337         { "metadata", 1, NULL, 'm' },
 338         { "mixed", 0, NULL, 'M' },
 339         { "nodesize", 1, NULL, 'n' },
 340         { "sectorsize", 1, NULL, 's' },
 341         { "data", 1, NULL, 'd' },
 342         { "version", 0, NULL, 'V' },
 343         { "rootdir", 1, NULL, 'r' },
 344         { "nodiscard", 0, NULL, 'K' },
 345         { "features", 0, NULL, 'O' },
 346         { NULL, 0, NULL, 0}
 347 };
 348
 349 static int add_directory_items(struct btrfs_trans_handle *trans,
 350                                struct btrfs_root *root, u64 objectid,
 351                                ino_t parent_inum, const char *name,
 352                                struct stat *st, int *dir_index_cnt)
 353 {
 354         int ret;
 355         int name_len;
 356         struct btrfs_key location;
 357         u8 filetype = 0;
 358
 359         name_len = strlen(name);
 360
 361         location.objectid = objectid;
 362         location.offset = 0;
 363         btrfs_set_key_type(&location, BTRFS_INODE_ITEM_KEY);
 364
 365         if (S_ISDIR(st->st_mode))
 366                 filetype = BTRFS_FT_DIR;
 367         if (S_ISREG(st->st_mode))
 368                 filetype = BTRFS_FT_REG_FILE;
 369         if (S_ISLNK(st->st_mode))
 370                 filetype = BTRFS_FT_SYMLINK;
 371
 372         ret = btrfs_insert_dir_item(trans, root, name, name_len,
 373                                     parent_inum, &location,
 374                                     filetype, index_cnt);
 375
 376         *dir_index_cnt = index_cnt;
 377         index_cnt++;
 378
 379         return ret;
 380 }
 381
 382 static int fill_inode_item(struct btrfs_trans_handle *trans,
 383                            struct btrfs_root *root,
 384                            struct btrfs_inode_item *dst, struct stat *src)
 385 {
 386         u64 blocks = 0;
 387         u64 sectorsize = root->sectorsize;
 388
 389         /*
 390          * btrfs_inode_item has some reserved fields
 391          * and represents on-disk inode entry, so
 392          * zero everything to prevent information leak
 393          */
 394         memset(dst, 0, sizeof (*dst));
 395
 396         btrfs_set_stack_inode_generation(dst, trans->transid);
 397         btrfs_set_stack_inode_size(dst, src->st_size);
 398         btrfs_set_stack_inode_nbytes(dst, 0);
 399         btrfs_set_stack_inode_block_group(dst, 0);
 400         btrfs_set_stack_inode_nlink(dst, src->st_nlink);
 401         btrfs_set_stack_inode_uid(dst, src->st_uid);
 402         btrfs_set_stack_inode_gid(dst, src->st_gid);
 403         btrfs_set_stack_inode_mode(dst, src->st_mode);
 404         btrfs_set_stack_inode_rdev(dst, 0);
 405         btrfs_set_stack_inode_flags(dst, 0);
 406         btrfs_set_stack_timespec_sec(&dst->atime, src->st_atime);
 407         btrfs_set_stack_timespec_nsec(&dst->atime, 0);
 408         btrfs_set_stack_timespec_sec(&dst->ctime, src->st_ctime);
 409         btrfs_set_stack_timespec_nsec(&dst->ctime, 0);
 410         btrfs_set_stack_timespec_sec(&dst->mtime, src->st_mtime);
 411         btrfs_set_stack_timespec_nsec(&dst->mtime, 0);
 412         btrfs_set_stack_timespec_sec(&dst->otime, 0);
 413         btrfs_set_stack_timespec_nsec(&dst->otime, 0);
 414
 415         if (S_ISDIR(src->st_mode)) {
 416                 btrfs_set_stack_inode_size(dst, 0);
 417                 btrfs_set_stack_inode_nlink(dst, 1);
 418         }
 419         if (S_ISREG(src->st_mode)) {
 420                 btrfs_set_stack_inode_size(dst, (u64)src->st_size);
 421                 if (src->st_size <= BTRFS_MAX_INLINE_DATA_SIZE(root))
 422                         btrfs_set_stack_inode_nbytes(dst, src->st_size);
 423                 else {
 424                         blocks = src->st_size / sectorsize;
 425                         if (src->st_size % sectorsize)
 426                                 blocks += 1;
 427                         blocks *= sectorsize;
 428                         btrfs_set_stack_inode_nbytes(dst, blocks);
 429                 }
 430         }
 431         if (S_ISLNK(src->st_mode))
 432                 btrfs_set_stack_inode_nbytes(dst, src->st_size + 1);
 433
 434         return 0;
 435 }
 436
 437 static int directory_select(const struct direct *entry)
 438 {
 439         if ((strncmp(entry->d_name, ".", entry->d_reclen) == 0) ||
 440                 (strncmp(entry->d_name, "..", entry->d_reclen) == 0))
 441                 return 0;
 442         else
 443                 return 1;
 444 }
 445
 446 static void free_namelist(struct direct **files, int count)
 447 {
 448         int i;
 449
 450         if (count < 0)
 451                 return;
 452
 453         for (i = 0; i < count; ++i)
 454                 free(files[i]);
 455         free(files);
 456 }
 457
 458 static u64 calculate_dir_inode_size(char *dirname)
 459 {
 460         int count, i;
 461         struct direct **files, *cur_file;
 462         u64 dir_inode_size = 0;
 463
 464         count = scandir(dirname, &files, directory_select, NULL);
 465
 466         for (i = 0; i < count; i++) {
 467                 cur_file = files[i];
 468                 dir_inode_size += strlen(cur_file->d_name);
 469         }
 470
 471         free_namelist(files, count);
 472
 473         dir_inode_size *= 2;
 474         return dir_inode_size;
 475 }
 476
 477 static int add_inode_items(struct btrfs_trans_handle *trans,
 478                            struct btrfs_root *root,
 479                            struct stat *st, char *name,
 480                            u64 self_objectid, ino_t parent_inum,
 481                            int dir_index_cnt, struct btrfs_inode_item *inode_ret)
 482 {
 483         int ret;
 484         struct btrfs_key inode_key;
 485         struct btrfs_inode_item btrfs_inode;
 486         u64 objectid;
 487         u64 inode_size = 0;
 488         int name_len;
 489
 490         name_len = strlen(name);
 491         fill_inode_item(trans, root, &btrfs_inode, st);
 492         objectid = self_objectid;
 493
 494         if (S_ISDIR(st->st_mode)) {
 495                 inode_size = calculate_dir_inode_size(name);
 496                 btrfs_set_stack_inode_size(&btrfs_inode, inode_size);
 497         }
 498
 499         inode_key.objectid = objectid;
 500         inode_key.offset = 0;
 501         btrfs_set_key_type(&inode_key, BTRFS_INODE_ITEM_KEY);
 502
 503         ret = btrfs_insert_inode(trans, root, objectid, &btrfs_inode);
 504         if (ret)
 505                 goto fail;
 506
 507         ret = btrfs_insert_inode_ref(trans, root, name, name_len,
 508                                      objectid, parent_inum, dir_index_cnt);
 509         if (ret)
 510                 goto fail;
 511
 512         *inode_ret = btrfs_inode;
 513 fail:
 514         return ret;
 515 }
 516
 517 static int add_xattr_item(struct btrfs_trans_handle *trans,
 518                           struct btrfs_root *root, u64 objectid,
 519                           const char *file_name)
 520 {
 521         int ret;
 522         int cur_name_len;
 523         char xattr_list[XATTR_LIST_MAX];
 524         char *cur_name;
 525         char cur_value[XATTR_SIZE_MAX];
 526         char delimiter = '\0';
 527         char *next_location = xattr_list;
 528
 529         ret = llistxattr(file_name, xattr_list, XATTR_LIST_MAX);
 530         if (ret < 0) {
 531                 if(errno == ENOTSUP)
 532                         return 0;
 533                 fprintf(stderr, "get a list of xattr failed for %s\n",
 534                         file_name);
 535                 return ret;
 536         }
 537         if (ret == 0)
 538                 return ret;
 539
 540         cur_name = strtok(xattr_list, &delimiter);
 541         while (cur_name != NULL) {
 542                 cur_name_len = strlen(cur_name);
 543                 next_location += cur_name_len + 1;
 544
 545                 ret = getxattr(file_name, cur_name, cur_value, XATTR_SIZE_MAX);
 546                 if (ret < 0) {
 547                         if(errno == ENOTSUP)
 548                                 return 0;
 549                         fprintf(stderr, "get a xattr value failed for %s attr %s\n",
 550                                 file_name, cur_name);
 551                         return ret;
 552                 }
 553
 554                 ret = btrfs_insert_xattr_item(trans, root, cur_name,
 555                                               cur_name_len, cur_value,
 556                                               ret, objectid);
 557                 if (ret) {
 558                         fprintf(stderr, "insert a xattr item failed for %s\n",
 559                                 file_name);
 560                 }
 561
 562                 cur_name = strtok(next_location, &delimiter);
 563         }
 564
 565         return ret;
 566 }
 567 static int custom_alloc_extent(struct btrfs_root *root, u64 num_bytes,
 568                                u64 hint_byte, struct btrfs_key *ins)
 569 {
 570         u64 start;
 571         u64 end;
 572         u64 last = hint_byte;
 573         int ret;
 574         int wrapped = 0;
 575         struct btrfs_block_group_cache *cache;
 576
 577         while (1) {
 578                 ret = find_first_extent_bit(&root->fs_info->free_space_cache,
 579                                             last, &start, &end, EXTENT_DIRTY);
 580                 if (ret) {
 581                         if (wrapped++ == 0) {
 582                                 last = 0;
 583                                 continue;
 584                         } else {
 585                                 goto fail;
 586                         }
 587                 }
 588
 589                 start = max(last, start);
 590                 last = end + 1;
 591                 if (last - start < num_bytes)
 592                         continue;
 593
 594                 last = start + num_bytes;
 595                 if (test_range_bit(&root->fs_info->pinned_extents,
 596                                    start, last - 1, EXTENT_DIRTY, 0))
 597                         continue;
 598
 599                 cache = btrfs_lookup_block_group(root->fs_info, start);
 600                 BUG_ON(!cache);
 601                 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM ||
 602                     last > cache->key.objectid + cache->key.offset) {
 603                         last = cache->key.objectid + cache->key.offset;
 604                         continue;
 605                 }
 606
 607                 if (cache->flags & (BTRFS_BLOCK_GROUP_SYSTEM |
 608                             BTRFS_BLOCK_GROUP_METADATA)) {
 609                         last = cache->key.objectid + cache->key.offset;
 610                         continue;
 611                 }
 612
 613                 clear_extent_dirty(&root->fs_info->free_space_cache,
 614                                    start, start + num_bytes - 1, 0);
 615
 616                 ins->objectid = start;
 617                 ins->offset = num_bytes;
 618                 ins->type = BTRFS_EXTENT_ITEM_KEY;
 619                 return 0;
 620         }
 621 fail:
 622         fprintf(stderr, "not enough free space\n");
 623         return -ENOSPC;
 624 }
 625
 626 static int record_file_extent(struct btrfs_trans_handle *trans,
 627                               struct btrfs_root *root, u64 objectid,
 628                               struct btrfs_inode_item *inode,
 629                               u64 file_pos, u64 disk_bytenr,
 630                               u64 num_bytes)
 631 {
 632         int ret;
 633         struct btrfs_fs_info *info = root->fs_info;
 634         struct btrfs_root *extent_root = info->extent_root;
 635         struct extent_buffer *leaf;
 636         struct btrfs_file_extent_item *fi;
 637         struct btrfs_key ins_key;
 638         struct btrfs_path path;
 639         struct btrfs_extent_item *ei;
 640
 641         btrfs_init_path(&path);
 642
 643         ins_key.objectid = objectid;
 644         ins_key.offset = 0;
 645         btrfs_set_key_type(&ins_key, BTRFS_EXTENT_DATA_KEY);
 646         ret = btrfs_insert_empty_item(trans, root, &path, &ins_key,
 647                                       sizeof(*fi));
 648         if (ret)
 649                 goto fail;
 650         leaf = path.nodes[0];
 651         fi = btrfs_item_ptr(leaf, path.slots[0],
 652                             struct btrfs_file_extent_item);
 653         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
 654         btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
 655         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
 656         btrfs_set_file_extent_disk_num_bytes(leaf, fi, num_bytes);
 657         btrfs_set_file_extent_offset(leaf, fi, 0);
 658         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
 659         btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
 660         btrfs_set_file_extent_compression(leaf, fi, 0);
 661         btrfs_set_file_extent_encryption(leaf, fi, 0);
 662         btrfs_set_file_extent_other_encoding(leaf, fi, 0);
 663         btrfs_mark_buffer_dirty(leaf);
 664
 665         btrfs_release_path(&path);
 666
 667         ins_key.objectid = disk_bytenr;
 668         ins_key.offset = num_bytes;
 669         ins_key.type = BTRFS_EXTENT_ITEM_KEY;
 670
 671         ret = btrfs_insert_empty_item(trans, extent_root, &path,
 672                                 &ins_key, sizeof(*ei));
 673         if (ret == 0) {
 674                 leaf = path.nodes[0];
 675                 ei = btrfs_item_ptr(leaf, path.slots[0],
 676                                     struct btrfs_extent_item);
 677
 678                 btrfs_set_extent_refs(leaf, ei, 0);
 679                 btrfs_set_extent_generation(leaf, ei, trans->transid);
 680                 btrfs_set_extent_flags(leaf, ei, BTRFS_EXTENT_FLAG_DATA);
 681
 682                 btrfs_mark_buffer_dirty(leaf);
 683                 ret = btrfs_update_block_group(trans, root, disk_bytenr,
 684                                                num_bytes, 1, 0);
 685                 if (ret)
 686                         goto fail;
 687         } else if (ret != -EEXIST) {
 688                 goto fail;
 689         }
 690
 691         ret = btrfs_inc_extent_ref(trans, root, disk_bytenr, num_bytes, 0,
 692                                    root->root_key.objectid,
 693                                    objectid, 0);
 694 fail:
 695         btrfs_release_path(&path);
 696         return ret;
 697 }
 698
 699 static int add_symbolic_link(struct btrfs_trans_handle *trans,
 700                              struct btrfs_root *root,
 701                              u64 objectid, const char *path_name)
 702 {
 703         int ret;
 704         u64 sectorsize = root->sectorsize;
 705         char *buf = malloc(sectorsize);
 706
 707         ret = readlink(path_name, buf, sectorsize);
 708         if (ret <= 0) {
 709                 fprintf(stderr, "readlink failed for %s\n", path_name);
 710                 goto fail;
 711         }
 712         if (ret >= sectorsize) {
 713                 fprintf(stderr, "symlink too long for %s", path_name);
 714                 ret = -1;
 715                 goto fail;
 716         }
 717
 718         buf[ret] = '\0'; /* readlink does not do it for us */
 719         ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
 720                                          buf, ret + 1);
 721 fail:
 722         free(buf);
 723         return ret;
 724 }
 725
 726 static int add_file_items(struct btrfs_trans_handle *trans,
 727                           struct btrfs_root *root,
 728                           struct btrfs_inode_item *btrfs_inode, u64 objectid,
 729                           ino_t parent_inum, struct stat *st,
 730                           const char *path_name, int out_fd)
 731 {
 732         int ret = -1;
 733         ssize_t ret_read;
 734         u64 bytes_read = 0;
 735         char *buffer = NULL;
 736         struct btrfs_key key;
 737         int blocks;
 738         u32 sectorsize = root->sectorsize;
 739         u64 first_block = 0;
 740         u64 num_blocks = 0;
 741         int fd;
 742
 743         fd = open(path_name, O_RDONLY);
 744         if (fd == -1) {
 745                 fprintf(stderr, "%s open failed\n", path_name);
 746                 return ret;
 747         }
 748
 749         blocks = st->st_size / sectorsize;
 750         if (st->st_size % sectorsize)
 751                 blocks += 1;
 752
 753         if (st->st_size <= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 754                 buffer = malloc(st->st_size);
 755                 ret_read = pread64(fd, buffer, st->st_size, bytes_read);
 756                 if (ret_read == -1) {
 757                         fprintf(stderr, "%s read failed\n", path_name);
 758                         goto end;
 759                 }
 760
 761                 ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
 762                                                  buffer, st->st_size);
 763                 goto end;
 764         }
 765
 766         ret = custom_alloc_extent(root, blocks * sectorsize, 0, &key);
 767         if (ret)
 768                 goto end;
 769
 770         first_block = key.objectid;
 771         bytes_read = 0;
 772         buffer = malloc(sectorsize);
 773
 774         do {
 775                 memset(buffer, 0, sectorsize);
 776                 ret_read = pread64(fd, buffer, sectorsize, bytes_read);
 777                 if (ret_read == -1) {
 778                         fprintf(stderr, "%s read failed\n", path_name);
 779                         goto end;
 780                 }
 781
 782                 ret = pwrite64(out_fd, buffer, sectorsize,
 783                                first_block + bytes_read);
 784                 if (ret != sectorsize) {
 785                         fprintf(stderr, "output file write failed\n");
 786                         goto end;
 787                 }
 788
 789                 /* checksum for file data */
 790                 ret = btrfs_csum_file_block(trans, root->fs_info->csum_root,
 791                                 first_block + (blocks * sectorsize),
 792                                 first_block + bytes_read,
 793                                 buffer, sectorsize);
 794                 if (ret) {
 795                         fprintf(stderr, "%s checksum failed\n", path_name);
 796                         goto end;
 797                 }
 798
 799                 bytes_read += ret_read;
 800                 num_blocks++;
 801         } while (ret_read == sectorsize);
 802
 803         if (num_blocks > 0) {
 804                 ret = record_file_extent(trans, root, objectid, btrfs_inode,
 805                                          first_block, first_block,
 806                                          blocks * sectorsize);
 807                 if (ret)
 808                         goto end;
 809         }
 810
 811 end:
 812         if (buffer)
 813                 free(buffer);
 814         close(fd);
 815         return ret;
 816 }
 817
 818 static char *make_path(char *dir, char *name)
 819 {
 820         char *path;
 821
 822         path = malloc(strlen(dir) + strlen(name) + 2);
 823         if (!path)
 824                 return NULL;
 825         strcpy(path, dir);
 826         if (dir[strlen(dir) - 1] != '/')
 827                 strcat(path, "/");
 828         strcat(path, name);
 829         return path;
 830 }
 831
 832 static int traverse_directory(struct btrfs_trans_handle *trans,
 833                               struct btrfs_root *root, char *dir_name,
 834                               struct directory_name_entry *dir_head, int out_fd)
 835 {
 836         int ret = 0;
 837
 838         struct btrfs_inode_item cur_inode;
 839         struct btrfs_inode_item *inode_item;
 840         int count, i, dir_index_cnt;
 841         struct direct **files;
 842         struct stat st;
 843         struct directory_name_entry *dir_entry, *parent_dir_entry;
 844         struct direct *cur_file;
 845         ino_t parent_inum, cur_inum;
 846         ino_t highest_inum = 0;
 847         char *parent_dir_name;
 848         struct btrfs_path path;
 849         struct extent_buffer *leaf;
 850         struct btrfs_key root_dir_key;
 851         u64 root_dir_inode_size = 0;
 852
 853         /* Add list for source directory */
 854         dir_entry = malloc(sizeof(struct directory_name_entry));
 855         dir_entry->dir_name = dir_name;
 856         dir_entry->path = strdup(dir_name);
 857
 858         parent_inum = highest_inum + BTRFS_FIRST_FREE_OBJECTID;
 859         dir_entry->inum = parent_inum;
 860         list_add_tail(&dir_entry->list, &dir_head->list);
 861
 862         btrfs_init_path(&path);
 863
 864         root_dir_key.objectid = btrfs_root_dirid(&root->root_item);
 865         root_dir_key.offset = 0;
 866         btrfs_set_key_type(&root_dir_key, BTRFS_INODE_ITEM_KEY);
 867         ret = btrfs_lookup_inode(trans, root, &path, &root_dir_key, 1);
 868         if (ret) {
 869                 fprintf(stderr, "root dir lookup error\n");
 870                 return -1;
 871         }
 872
 873         leaf = path.nodes[0];
 874         inode_item = btrfs_item_ptr(leaf, path.slots[0],
 875                                     struct btrfs_inode_item);
 876
 877         root_dir_inode_size = calculate_dir_inode_size(dir_name);
 878         btrfs_set_inode_size(leaf, inode_item, root_dir_inode_size);
 879         btrfs_mark_buffer_dirty(leaf);
 880
 881         btrfs_release_path(&path);
 882
 883         do {
 884                 parent_dir_entry = list_entry(dir_head->list.next,
 885                                               struct directory_name_entry,
 886                                               list);
 887                 list_del(&parent_dir_entry->list);
 888
 889                 parent_inum = parent_dir_entry->inum;
 890                 parent_dir_name = parent_dir_entry->dir_name;
 891                 if (chdir(parent_dir_entry->path)) {
 892                         fprintf(stderr, "chdir error for %s\n",
 893                                 parent_dir_name);
 894                         goto fail_no_files;
 895                 }
 896
 897                 count = scandir(parent_dir_entry->path, &files,
 898                                 directory_select, NULL);
 899                 if (count == -1)
 900                 {
 901                         fprintf(stderr, "scandir for %s failed: %s\n",
 902                                 parent_dir_name, strerror (errno));
 903                         goto fail;
 904                 }
 905
 906                 for (i = 0; i < count; i++) {
 907                         cur_file = files[i];
 908
 909                         if (lstat(cur_file->d_name, &st) == -1) {
 910                                 fprintf(stderr, "lstat failed for file %s\n",
 911                                         cur_file->d_name);
 912                                 goto fail;
 913                         }
 914
 915                         cur_inum = ++highest_inum + BTRFS_FIRST_FREE_OBJECTID;
 916                         ret = add_directory_items(trans, root,
 917                                                   cur_inum, parent_inum,
 918                                                   cur_file->d_name,
 919                                                   &st, &dir_index_cnt);
 920                         if (ret) {
 921                                 fprintf(stderr, "add_directory_items failed\n");
 922                                 goto fail;
 923                         }
 924
 925                         ret = add_inode_items(trans, root, &st,
 926                                               cur_file->d_name, cur_inum,
 927                                               parent_inum, dir_index_cnt,
 928                                               &cur_inode);
 929                         if (ret) {
 930                                 fprintf(stderr, "add_inode_items failed\n");
 931                                 goto fail;
 932                         }
 933
 934                         ret = add_xattr_item(trans, root,
 935                                              cur_inum, cur_file->d_name);
 936                         if (ret) {
 937                                 fprintf(stderr, "add_xattr_item failed\n");
 938                                 if(ret != -ENOTSUP)
 939                                         goto fail;
 940                         }
 941
 942                         if (S_ISDIR(st.st_mode)) {
 943                                 dir_entry = malloc(sizeof(struct directory_name_entry));
 944                                 dir_entry->dir_name = cur_file->d_name;
 945                                 dir_entry->path = make_path(parent_dir_entry->path,
 946                                                             cur_file->d_name);
 947                                 dir_entry->inum = cur_inum;
 948                                 list_add_tail(&dir_entry->list, &dir_head->list);
 949                         } else if (S_ISREG(st.st_mode)) {
 950                                 ret = add_file_items(trans, root, &cur_inode,
 951                                                      cur_inum, parent_inum, &st,
 952                                                      cur_file->d_name, out_fd);
 953                                 if (ret) {
 954                                         fprintf(stderr, "add_file_items failed\n");
 955                                         goto fail;
 956                                 }
 957                         } else if (S_ISLNK(st.st_mode)) {
 958                                 ret = add_symbolic_link(trans, root,
 959                                                         cur_inum, cur_file->d_name);
 960                                 if (ret) {
 961                                         fprintf(stderr, "add_symbolic_link failed\n");
 962                                         goto fail;
 963                                 }
 964                         }
 965                 }
 966
 967                 free_namelist(files, count);
 968                 free(parent_dir_entry->path);
 969                 free(parent_dir_entry);
 970
 971                 index_cnt = 2;
 972
 973         } while (!list_empty(&dir_head->list));
 974
 975         return 0;
 976 fail:
 977         free_namelist(files, count);
 978 fail_no_files:
 979         free(parent_dir_entry->path);
 980         free(parent_dir_entry);
 981         return -1;
 982 }
 983
 984 static int open_target(char *output_name)
 985 {
 986         int output_fd;
 987         output_fd = open(output_name, O_CREAT | O_RDWR | O_TRUNC,
 988                          S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH);
 989
 990         return output_fd;
 991 }
 992
 993 static int create_chunks(struct btrfs_trans_handle *trans,
 994                          struct btrfs_root *root, u64 num_of_meta_chunks,
 995                          u64 size_of_data)
 996 {
 997         u64 chunk_start;
 998         u64 chunk_size;
 999         u64 meta_type = BTRFS_BLOCK_GROUP_METADATA;
1000         u64 data_type = BTRFS_BLOCK_GROUP_DATA;
1001         u64 minimum_data_chunk_size = 8 * 1024 * 1024;
1002         u64 i;
1003         int ret;
1004
1005         for (i = 0; i < num_of_meta_chunks; i++) {
1006                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
1007                                         &chunk_start, &chunk_size, meta_type);
1008                 BUG_ON(ret);
1009                 ret = btrfs_make_block_group(trans, root->fs_info->extent_root, 0,
1010                                              meta_type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1011                                              chunk_start, chunk_size);
1012                 BUG_ON(ret);
1013                 set_extent_dirty(&root->fs_info->free_space_cache,
1014                                  chunk_start, chunk_start + chunk_size - 1, 0);
1015         }
1016
1017         if (size_of_data < minimum_data_chunk_size)
1018                 size_of_data = minimum_data_chunk_size;
1019         ret = btrfs_alloc_data_chunk(trans, root->fs_info->extent_root,
1020                                      &chunk_start, size_of_data, data_type);
1021         BUG_ON(ret);
1022         ret = btrfs_make_block_group(trans, root->fs_info->extent_root, 0,
1023                                      data_type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1024                                      chunk_start, size_of_data);
1025         BUG_ON(ret);
1026         set_extent_dirty(&root->fs_info->free_space_cache,
1027                          chunk_start, chunk_start + size_of_data - 1, 0);
1028         return ret;
1029 }
1030
1031 static int make_image(char *source_dir, struct btrfs_root *root, int out_fd)
1032 {
1033         int ret;
1034         struct btrfs_trans_handle *trans;
1035
1036         struct stat root_st;
1037
1038         struct directory_name_entry dir_head;
1039
1040         struct directory_name_entry *dir_entry = NULL;
1041
1042         ret = lstat(source_dir, &root_st);
1043         if (ret) {
1044                 fprintf(stderr, "unable to lstat the %s\n", source_dir);
1045                 goto fail;
1046         }
1047
1048         INIT_LIST_HEAD(&dir_head.list);
1049
1050         trans = btrfs_start_transaction(root, 1);
1051         ret = traverse_directory(trans, root, source_dir, &dir_head, out_fd);
1052         if (ret) {
1053                 fprintf(stderr, "unable to traverse_directory\n");
1054                 goto fail;
1055         }
1056         btrfs_commit_transaction(trans, root);
1057
1058         printf("Making image is completed.\n");
1059         return 0;
1060 fail:
1061         while (!list_empty(&dir_head.list)) {
1062                 dir_entry = list_entry(dir_head.list.next,
1063                                        struct directory_name_entry, list);
1064                 list_del(&dir_entry->list);
1065                 free(dir_entry);
1066         }
1067         fprintf(stderr, "Making image is aborted.\n");
1068         return -1;
1069 }
1070
1071 /*
1072  * This ignores symlinks with unreadable targets and subdirs that can't
1073  * be read.  It's a best-effort to give a rough estimate of the size of
1074  * a subdir.  It doesn't guarantee that prepopulating btrfs from this
1075  * tree won't still run out of space.
1076  *
1077  * The rounding up to 4096 is questionable.  Previous code used du -B 4096.
1078  */
1079 static u64 global_total_size;
1080 static int ftw_add_entry_size(const char *fpath, const struct stat *st,
1081                               int type)
1082 {
1083         if (type == FTW_F || type == FTW_D)
1084                 global_total_size += round_up(st->st_size, 4096);
1085
1086         return 0;
1087 }
1088
1089 static u64 size_sourcedir(char *dir_name, u64 sectorsize,
1090                           u64 *num_of_meta_chunks_ret, u64 *size_of_data_ret)
1091 {
1092         u64 dir_size = 0;
1093         u64 total_size = 0;
1094         int ret;
1095         u64 default_chunk_size = 8 * 1024 * 1024;       /* 8MB */
1096         u64 allocated_meta_size = 8 * 1024 * 1024;      /* 8MB */
1097         u64 allocated_total_size = 20 * 1024 * 1024;    /* 20MB */
1098         u64 num_of_meta_chunks = 0;
1099         u64 num_of_allocated_meta_chunks =
1100                         allocated_meta_size / default_chunk_size;
1101
1102         global_total_size = 0;
1103         ret = ftw(dir_name, ftw_add_entry_size, 10);
1104         dir_size = global_total_size;
1105         if (ret < 0) {
1106                 fprintf(stderr, "ftw subdir walk of '%s' failed: %s\n",
1107                         dir_name, strerror(errno));
1108                 exit(1);
1109         }
1110
1111         num_of_meta_chunks = (dir_size / 2) / default_chunk_size;
1112         if (((dir_size / 2) % default_chunk_size) != 0)
1113                 num_of_meta_chunks++;
1114         if (num_of_meta_chunks <= num_of_allocated_meta_chunks)
1115                 num_of_meta_chunks = 0;
1116         else
1117                 num_of_meta_chunks -= num_of_allocated_meta_chunks;
1118
1119         total_size = allocated_total_size + dir_size +
1120                      (num_of_meta_chunks * default_chunk_size);
1121
1122         *num_of_meta_chunks_ret = num_of_meta_chunks;
1123
1124         return total_size;
1125 }
1126
1127 static int zero_output_file(int out_fd, u64 size, u32 sectorsize)
1128 {
1129         int len = sectorsize;
1130         int loop_num = size / sectorsize;
1131         u64 location = 0;
1132         char *buf = malloc(len);
1133         int ret = 0, i;
1134         ssize_t written;
1135
1136         if (!buf)
1137                 return -ENOMEM;
1138         memset(buf, 0, len);
1139         for (i = 0; i < loop_num; i++) {
1140                 written = pwrite64(out_fd, buf, len, location);
1141                 if (written != len)
1142                         ret = -EIO;
1143                 location += sectorsize;
1144         }
1145         free(buf);
1146         return ret;
1147 }
1148
1149 static int check_leaf_or_node_size(u32 size, u32 sectorsize)
1150 {
1151         if (size < sectorsize) {
1152                 fprintf(stderr,
1153                         "Illegal leafsize (or nodesize) %u (smaller than %u)\n",
1154                         size, sectorsize);
1155                 return -1;
1156         } else if (size > BTRFS_MAX_METADATA_BLOCKSIZE) {
1157                 fprintf(stderr,
1158                         "Illegal leafsize (or nodesize) %u (larger than %u)\n",
1159                         size, BTRFS_MAX_METADATA_BLOCKSIZE);
1160                 return -1;
1161         } else if (size & (sectorsize - 1)) {
1162                 fprintf(stderr,
1163                         "Illegal leafsize (or nodesize) %u (not align to %u)\n",
1164                         size, sectorsize);
1165                 return -1;
1166         }
1167         return 0;
1168 }
1169
1170 static int is_ssd(const char *file)
1171 {
1172         blkid_probe probe;
1173         char wholedisk[32];
1174         char sysfs_path[PATH_MAX];
1175         dev_t devno;
1176         int fd;
1177         char rotational;
1178         int ret;
1179
1180         probe = blkid_new_probe_from_filename(file);
1181         if (!probe)
1182                 return 0;
1183
1184         /* Device number of this disk (possibly a partition) */
1185         devno = blkid_probe_get_devno(probe);
1186         if (!devno) {
1187                 blkid_free_probe(probe);
1188                 return 0;
1189         }
1190
1191         /* Get whole disk name (not full path) for this devno */
1192         ret = blkid_devno_to_wholedisk(devno,
1193                         wholedisk, sizeof(wholedisk), NULL);
1194         if (ret) {
1195                 blkid_free_probe(probe);
1196                 return 0;
1197         }
1198
1199         snprintf(sysfs_path, PATH_MAX, "/sys/block/%s/queue/rotational",
1200                  wholedisk);
1201
1202         blkid_free_probe(probe);
1203
1204         fd = open(sysfs_path, O_RDONLY);
1205         if (fd < 0) {
1206                 return 0;
1207         }
1208
1209         if (read(fd, &rotational, sizeof(char)) < sizeof(char)) {
1210                 close(fd);
1211                 return 0;
1212         }
1213         close(fd);
1214
1215         return !atoi((const char *)&rotational);
1216 }
1217
1218 #define BTRFS_FEATURE_LIST_ALL          (1ULL << 63)
1219
1220 static const struct btrfs_fs_feature {
1221         const char *name;
1222         u64 flag;
1223         const char *desc;
1224 } mkfs_features[] = {
1225         { "mixed-bg", BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS,
1226                 "mixed data and metadata block groups" },
1227         { "extref", BTRFS_FEATURE_INCOMPAT_EXTENDED_IREF,
1228                 "increased hardlink limit per file to 65536" },
1229         { "raid56", BTRFS_FEATURE_INCOMPAT_RAID56,
1230                 "raid56 extended format" },
1231         { "skinny-metadata", BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA,
1232                 "reduced-size metadata extent refs" },
1233         /* Keep this one last */
1234         { "list-all", BTRFS_FEATURE_LIST_ALL, NULL }
1235 };
1236
1237 static void list_all_fs_features(void)
1238 {
1239         int i;
1240
1241         fprintf(stderr, "Filesystem features available at mkfs time:\n");
1242         for (i = 0; i < ARRAY_SIZE(mkfs_features) - 1; i++) {
1243                 fprintf(stderr, "%-20s- %s (0x%llx)\n",
1244                                 mkfs_features[i].name,
1245                                 mkfs_features[i].desc,
1246                                 mkfs_features[i].flag);
1247         }
1248 }
1249
1250 static int parse_one_fs_feature(const char *name, u64 *flags)
1251 {
1252         int i;
1253         int found = 0;
1254
1255         for (i = 0; i < ARRAY_SIZE(mkfs_features); i++) {
1256                 if (!strcmp(mkfs_features[i].name, name)) {
1257                         *flags |= mkfs_features[i].flag;
1258                         found = 1;
1259                 }
1260         }
1261
1262         return !found;
1263 }
1264
1265 static void process_fs_features(u64 flags)
1266 {
1267         int i;
1268
1269         for (i = 0; i < ARRAY_SIZE(mkfs_features); i++) {
1270                 if (flags & mkfs_features[i].flag) {
1271                         fprintf(stderr,
1272                                 "Turning ON incompat feature '%s': %s\n",
1273                                 mkfs_features[i].name,
1274                                 mkfs_features[i].desc);
1275                 }
1276         }
1277 }
1278
1279
1280 /*
1281  * Return NULL if all features were parsed fine, otherwise return the name of
1282  * the first unparsed.
1283  */
1284 static char* parse_fs_features(char *namelist, u64 *flags)
1285 {
1286         char *this_char;
1287         char *save_ptr = NULL; /* Satisfy static checkers */
1288
1289         for (this_char = strtok_r(namelist, ",", &save_ptr);
1290              this_char != NULL;
1291              this_char = strtok_r(NULL, ",", &save_ptr)) {
1292                 if (parse_one_fs_feature(this_char, flags))
1293                         return this_char;
1294         }
1295
1296         return NULL;
1297 }
1298
1299 int main(int ac, char **av)
1300 {
1301         char *file;
1302         struct btrfs_root *root;
1303         struct btrfs_trans_handle *trans;
1304         char *label = NULL;
1305         char *first_file;
1306         u64 block_count = 0;
1307         u64 dev_block_count = 0;
1308         u64 blocks[7];
1309         u64 alloc_start = 0;
1310         u64 metadata_profile = 0;
1311         u64 data_profile = 0;
1312         u32 leafsize = sysconf(_SC_PAGESIZE);
1313         u32 sectorsize = 4096;
1314         u32 nodesize = leafsize;
1315         u32 stripesize = 4096;
1316         int zero_end = 1;
1317         int option_index = 0;
1318         int fd;
1319         int ret;
1320         int i;
1321         int mixed = 0;
1322         int data_profile_opt = 0;
1323         int metadata_profile_opt = 0;
1324         int nodiscard = 0;
1325         int ssd = 0;
1326         int force_overwrite = 0;
1327
1328         char *source_dir = NULL;
1329         int source_dir_set = 0;
1330         u64 num_of_meta_chunks = 0;
1331         u64 size_of_data = 0;
1332         u64 source_dir_size = 0;
1333         int dev_cnt = 0;
1334         int saved_optind;
1335         char estr[100];
1336         u64 features = 0;
1337
1338         while(1) {
1339                 int c;
1340                 c = getopt_long(ac, av, "A:b:fl:n:s:m:d:L:O:r:VMK",
1341                                 long_options, &option_index);
1342                 if (c < 0)
1343                         break;
1344                 switch(c) {
1345                         case 'A':
1346                                 alloc_start = parse_size(optarg);
1347                                 break;
1348                         case 'f':
1349                                 force_overwrite = 1;
1350                                 break;
1351                         case 'd':
1352                                 data_profile = parse_profile(optarg);
1353                                 data_profile_opt = 1;
1354                                 break;
1355                         case 'l':
1356                         case 'n':
1357                                 nodesize = parse_size(optarg);
1358                                 leafsize = parse_size(optarg);
1359                                 break;
1360                         case 'L':
1361                                 label = parse_label(optarg);
1362                                 break;
1363                         case 'm':
1364                                 metadata_profile = parse_profile(optarg);
1365                                 metadata_profile_opt = 1;
1366                                 break;
1367                         case 'M':
1368                                 mixed = 1;
1369                                 break;
1370                         case 'O': {
1371                                 char *orig = strdup(optarg);
1372                                 char *tmp = orig;
1373
1374                                 tmp = parse_fs_features(tmp, &features);
1375                                 if (tmp) {
1376                                         fprintf(stderr,
1377                                                 "Unrecognized filesystem feature '%s'\n",
1378                                                         tmp);
1379                                         free(orig);
1380                                         exit(1);
1381                                 }
1382                                 free(orig);
1383                                 if (features & BTRFS_FEATURE_LIST_ALL) {
1384                                         list_all_fs_features();
1385                                         exit(0);
1386                                 }
1387                                 break;
1388                                 }
1389                         case 's':
1390                                 sectorsize = parse_size(optarg);
1391                                 break;
1392                         case 'b':
1393                                 block_count = parse_size(optarg);
1394                                 if (block_count <= 1024*1024*1024) {
1395                                         printf("SMALL VOLUME: forcing mixed "
1396                                                "metadata/data groups\n");
1397                                         mixed = 1;
1398                                 }
1399                                 zero_end = 0;
1400                                 break;
1401                         case 'V':
1402                                 print_version();
1403                                 break;
1404                         case 'r':
1405                                 source_dir = optarg;
1406                                 source_dir_set = 1;
1407                                 break;
1408                         case 'K':
1409                                 nodiscard=1;
1410                                 break;
1411                         default:
1412                                 print_usage();
1413                 }
1414         }
1415         sectorsize = max(sectorsize, (u32)sysconf(_SC_PAGESIZE));
1416         if (check_leaf_or_node_size(leafsize, sectorsize))
1417                 exit(1);
1418         if (check_leaf_or_node_size(nodesize, sectorsize))
1419                 exit(1);
1420         saved_optind = optind;
1421         dev_cnt = ac - optind;
1422         if (dev_cnt == 0)
1423                 print_usage();
1424
1425         if (source_dir_set && dev_cnt > 1) {
1426                 fprintf(stderr,
1427                         "The -r option is limited to a single device\n");
1428                 exit(1);
1429         }
1430         while (dev_cnt-- > 0) {
1431                 file = av[optind++];
1432                 if (is_block_device(file))
1433                         if (test_dev_for_mkfs(file, force_overwrite, estr)) {
1434                                 fprintf(stderr, "Error: %s", estr);
1435                                 exit(1);
1436                         }
1437         }
1438
1439         optind = saved_optind;
1440         dev_cnt = ac - optind;
1441
1442         file = av[optind++];
1443         ssd = is_ssd(file);
1444
1445         if (is_vol_small(file)) {
1446                 printf("SMALL VOLUME: forcing mixed metadata/data groups\n");
1447                 mixed = 1;
1448                 if (metadata_profile != data_profile) {
1449                         if (metadata_profile_opt || data_profile_opt) {
1450                                 fprintf(stderr,
1451         "With mixed block groups data and metadata profiles must be the same\n");
1452                                 exit(1);
1453                         }
1454                 }
1455         }
1456         /*
1457         * Set default profiles according to number of added devices.
1458         * For mixed groups defaults are single/single.
1459         */
1460         if (!mixed) {
1461                 if (!metadata_profile_opt) {
1462                         if (dev_cnt == 1 && ssd)
1463                                 printf("Detected a SSD, turning off metadata "
1464                                 "duplication.  Mkfs with -m dup if you want to "
1465                                 "force metadata duplication.\n");
1466
1467                         metadata_profile = (dev_cnt > 1) ?
1468                                         BTRFS_BLOCK_GROUP_RAID1 : (ssd) ?
1469                                         0: BTRFS_BLOCK_GROUP_DUP;
1470                 }
1471                 if (!data_profile_opt) {
1472                         data_profile = (dev_cnt > 1) ?
1473                                 BTRFS_BLOCK_GROUP_RAID0 : 0; /* raid0 or single */
1474                 }
1475         } else {
1476                 metadata_profile = 0;
1477                 data_profile = 0;
1478         }
1479
1480         ret = test_num_disk_vs_raid(metadata_profile, data_profile,
1481                         dev_cnt, mixed, estr);
1482         if (ret) {
1483                 fprintf(stderr, "Error: %s\n", estr);
1484                 exit(1);
1485         }
1486
1487         /* if we are here that means all devs are good to btrfsify */
1488         printf("\nWARNING! - %s IS EXPERIMENTAL\n", BTRFS_BUILD_VERSION);
1489         printf("WARNING! - see http://btrfs.wiki.kernel.org before using\n\n");
1490
1491         dev_cnt--;
1492
1493         if (!source_dir_set) {
1494                 /*
1495                  * open without O_EXCL so that the problem should not
1496                  * occur by the following processing.
1497                  * (btrfs_register_one_device() fails if O_EXCL is on)
1498                  */
1499                 fd = open(file, O_RDWR);
1500                 if (fd < 0) {
1501                         fprintf(stderr, "unable to open %s: %s\n", file,
1502                                 strerror(errno));
1503                         exit(1);
1504                 }
1505                 first_file = file;
1506                 ret = btrfs_prepare_device(fd, file, zero_end, &dev_block_count,
1507                                            block_count, &mixed, nodiscard);
1508                 if (block_count && block_count > dev_block_count) {
1509                         fprintf(stderr, "%s is smaller than requested size\n", file);
1510                         exit(1);
1511                 }
1512         } else {
1513                 fd = open_target(file);
1514                 if (fd < 0) {
1515                         fprintf(stderr, "unable to open the %s\n", file);
1516                         exit(1);
1517                 }
1518
1519                 first_file = file;
1520                 source_dir_size = size_sourcedir(source_dir, sectorsize,
1521                                              &num_of_meta_chunks, &size_of_data);
1522                 if(block_count < source_dir_size)
1523                         block_count = source_dir_size;
1524                 ret = zero_output_file(fd, block_count, sectorsize);
1525                 if (ret) {
1526                         fprintf(stderr, "unable to zero the output file\n");
1527                         exit(1);
1528                 }
1529                 /* our "device" is the new image file */
1530                 dev_block_count = block_count;
1531         }
1532
1533         /* To create the first block group and chunk 0 in make_btrfs */
1534         if (dev_block_count < BTRFS_MKFS_SYSTEM_GROUP_SIZE) {
1535                 fprintf(stderr, "device is too small to make filesystem\n");
1536                 exit(1);
1537         }
1538
1539         blocks[0] = BTRFS_SUPER_INFO_OFFSET;
1540         for (i = 1; i < 7; i++) {
1541                 blocks[i] = BTRFS_SUPER_INFO_OFFSET + 1024 * 1024 +
1542                         leafsize * i;
1543         }
1544
1545         /*
1546          * FS features that can be set by other means than -O
1547          * just set the bit here
1548          */
1549         if (mixed)
1550                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS;
1551
1552         if ((data_profile | metadata_profile) &
1553             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
1554                 features |= BTRFS_FEATURE_INCOMPAT_RAID56;
1555         }
1556
1557         process_fs_features(features);
1558
1559         ret = make_btrfs(fd, file, label, blocks, dev_block_count,
1560                          nodesize, leafsize,
1561                          sectorsize, stripesize, features);
1562         if (ret) {
1563                 fprintf(stderr, "error during mkfs: %s\n", strerror(-ret));
1564                 exit(1);
1565         }
1566
1567         root = open_ctree(file, 0, O_RDWR);
1568         if (!root) {
1569                 fprintf(stderr, "Open ctree failed\n");
1570                 close(fd);
1571                 exit(1);
1572         }
1573         root->fs_info->alloc_start = alloc_start;
1574
1575         ret = make_root_dir(root, mixed);
1576         if (ret) {
1577                 fprintf(stderr, "failed to setup the root directory\n");
1578                 exit(1);
1579         }
1580
1581         trans = btrfs_start_transaction(root, 1);
1582
1583         if (dev_cnt == 0)
1584                 goto raid_groups;
1585
1586         btrfs_register_one_device(file);
1587
1588         zero_end = 1;
1589         while (dev_cnt-- > 0) {
1590                 int old_mixed = mixed;
1591
1592                 file = av[optind++];
1593
1594                 /*
1595                  * open without O_EXCL so that the problem should not
1596                  * occur by the following processing.
1597                  * (btrfs_register_one_device() fails if O_EXCL is on)
1598                  */
1599                 fd = open(file, O_RDWR);
1600                 if (fd < 0) {
1601                         fprintf(stderr, "unable to open %s: %s\n", file,
1602                                 strerror(errno));
1603                         exit(1);
1604                 }
1605                 ret = btrfs_device_already_in_root(root, fd,
1606                                                    BTRFS_SUPER_INFO_OFFSET);
1607                 if (ret) {
1608                         fprintf(stderr, "skipping duplicate device %s in FS\n",
1609                                 file);
1610                         close(fd);
1611                         continue;
1612                 }
1613                 ret = btrfs_prepare_device(fd, file, zero_end, &dev_block_count,
1614                                            block_count, &mixed, nodiscard);
1615                 mixed = old_mixed;
1616                 BUG_ON(ret);
1617
1618                 ret = btrfs_add_to_fsid(trans, root, fd, file, dev_block_count,
1619                                         sectorsize, sectorsize, sectorsize);
1620                 BUG_ON(ret);
1621                 btrfs_register_one_device(file);
1622         }
1623
1624 raid_groups:
1625         if (!source_dir_set) {
1626                 ret = create_raid_groups(trans, root, data_profile,
1627                                  data_profile_opt, metadata_profile,
1628                                  metadata_profile_opt, mixed, ssd);
1629                 BUG_ON(ret);
1630         }
1631
1632         ret = create_data_reloc_tree(trans, root);
1633         BUG_ON(ret);
1634
1635         printf("fs created label %s on %s\n\tnodesize %u leafsize %u "
1636             "sectorsize %u size %s\n",
1637             label, first_file, nodesize, leafsize, sectorsize,
1638             pretty_size(btrfs_super_total_bytes(root->fs_info->super_copy)));
1639
1640         printf("%s\n", BTRFS_BUILD_VERSION);
1641         btrfs_commit_transaction(trans, root);
1642
1643         if (source_dir_set) {
1644                 trans = btrfs_start_transaction(root, 1);
1645                 ret = create_chunks(trans, root,
1646                                     num_of_meta_chunks, size_of_data);
1647                 BUG_ON(ret);
1648                 btrfs_commit_transaction(trans, root);
1649
1650                 ret = make_image(source_dir, root, fd);
1651                 BUG_ON(ret);
1652         }
1653
1654         ret = close_ctree(root);
1655         BUG_ON(ret);
1656         free(label);
1657         return 0;
1658 }