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
 268 static void print_usage(void) __attribute__((noreturn));
 269 static void print_usage(void)
 270 {
 271         fprintf(stderr, "usage: mkfs.btrfs [options] dev [ dev ... ]\n");
 272         fprintf(stderr, "options:\n");
 273         fprintf(stderr, "\t -A --alloc-start the offset to start the FS\n");
 274         fprintf(stderr, "\t -b --byte-count total number of bytes in the FS\n");
 275         fprintf(stderr, "\t -d --data data profile, raid0, raid1, raid5, raid6, raid10, dup or single\n");
 276         fprintf(stderr, "\t -f --force force overwrite of existing filesystem\n");
 277         fprintf(stderr, "\t -l --leafsize size of btree leaves\n");
 278         fprintf(stderr, "\t -L --label set a label\n");
 279         fprintf(stderr, "\t -m --metadata metadata profile, values like data profile\n");
 280         fprintf(stderr, "\t -M --mixed mix metadata and data together\n");
 281         fprintf(stderr, "\t -n --nodesize size of btree nodes\n");
 282         fprintf(stderr, "\t -s --sectorsize min block allocation (may not mountable by current kernel)\n");
 283         fprintf(stderr, "\t -r --rootdir the source directory\n");
 284         fprintf(stderr, "\t -K --nodiscard do not perform whole device TRIM\n");
 285         fprintf(stderr, "\t -O --features comma separated list of filesystem features\n");
 286         fprintf(stderr, "\t -V --version print the mkfs.btrfs version and exit\n");
 287         fprintf(stderr, "%s\n", BTRFS_BUILD_VERSION);
 288         exit(1);
 289 }
 290
 291 static void print_version(void) __attribute__((noreturn));
 292 static void print_version(void)
 293 {
 294         fprintf(stderr, "mkfs.btrfs, part of %s\n", BTRFS_BUILD_VERSION);
 295         exit(0);
 296 }
 297
 298 static u64 parse_profile(char *s)
 299 {
 300         if (strcmp(s, "raid0") == 0) {
 301                 return BTRFS_BLOCK_GROUP_RAID0;
 302         } else if (strcmp(s, "raid1") == 0) {
 303                 return BTRFS_BLOCK_GROUP_RAID1;
 304         } else if (strcmp(s, "raid5") == 0) {
 305                 return BTRFS_BLOCK_GROUP_RAID5;
 306         } else if (strcmp(s, "raid6") == 0) {
 307                 return BTRFS_BLOCK_GROUP_RAID6;
 308         } else if (strcmp(s, "raid10") == 0) {
 309                 return BTRFS_BLOCK_GROUP_RAID10;
 310         } else if (strcmp(s, "dup") == 0) {
 311                 return BTRFS_BLOCK_GROUP_DUP;
 312         } else if (strcmp(s, "single") == 0) {
 313                 return 0;
 314         } else {
 315                 fprintf(stderr, "Unknown profile %s\n", s);
 316                 print_usage();
 317         }
 318         /* not reached */
 319         return 0;
 320 }
 321
 322 static char *parse_label(char *input)
 323 {
 324         int len = strlen(input);
 325
 326         if (len >= BTRFS_LABEL_SIZE) {
 327                 fprintf(stderr, "Label %s is too long (max %d)\n", input,
 328                         BTRFS_LABEL_SIZE - 1);
 329                 exit(1);
 330         }
 331         return strdup(input);
 332 }
 333
 334 static struct option long_options[] = {
 335         { "alloc-start", 1, NULL, 'A'},
 336         { "byte-count", 1, NULL, 'b' },
 337         { "force", 0, NULL, 'f' },
 338         { "leafsize", 1, NULL, 'l' },
 339         { "label", 1, NULL, 'L'},
 340         { "metadata", 1, NULL, 'm' },
 341         { "mixed", 0, NULL, 'M' },
 342         { "nodesize", 1, NULL, 'n' },
 343         { "sectorsize", 1, NULL, 's' },
 344         { "data", 1, NULL, 'd' },
 345         { "version", 0, NULL, 'V' },
 346         { "rootdir", 1, NULL, 'r' },
 347         { "nodiscard", 0, NULL, 'K' },
 348         { "features", 0, NULL, 'O' },
 349         { NULL, 0, NULL, 0}
 350 };
 351
 352 static int add_directory_items(struct btrfs_trans_handle *trans,
 353                                struct btrfs_root *root, u64 objectid,
 354                                ino_t parent_inum, const char *name,
 355                                struct stat *st, int *dir_index_cnt)
 356 {
 357         int ret;
 358         int name_len;
 359         struct btrfs_key location;
 360         u8 filetype = 0;
 361
 362         name_len = strlen(name);
 363
 364         location.objectid = objectid;
 365         location.offset = 0;
 366         btrfs_set_key_type(&location, BTRFS_INODE_ITEM_KEY);
 367
 368         if (S_ISDIR(st->st_mode))
 369                 filetype = BTRFS_FT_DIR;
 370         if (S_ISREG(st->st_mode))
 371                 filetype = BTRFS_FT_REG_FILE;
 372         if (S_ISLNK(st->st_mode))
 373                 filetype = BTRFS_FT_SYMLINK;
 374
 375         ret = btrfs_insert_dir_item(trans, root, name, name_len,
 376                                     parent_inum, &location,
 377                                     filetype, index_cnt);
 378
 379         *dir_index_cnt = index_cnt;
 380         index_cnt++;
 381
 382         return ret;
 383 }
 384
 385 static int fill_inode_item(struct btrfs_trans_handle *trans,
 386                            struct btrfs_root *root,
 387                            struct btrfs_inode_item *dst, struct stat *src)
 388 {
 389         u64 blocks = 0;
 390         u64 sectorsize = root->sectorsize;
 391
 392         /*
 393          * btrfs_inode_item has some reserved fields
 394          * and represents on-disk inode entry, so
 395          * zero everything to prevent information leak
 396          */
 397         memset(dst, 0, sizeof (*dst));
 398
 399         btrfs_set_stack_inode_generation(dst, trans->transid);
 400         btrfs_set_stack_inode_size(dst, src->st_size);
 401         btrfs_set_stack_inode_nbytes(dst, 0);
 402         btrfs_set_stack_inode_block_group(dst, 0);
 403         btrfs_set_stack_inode_nlink(dst, src->st_nlink);
 404         btrfs_set_stack_inode_uid(dst, src->st_uid);
 405         btrfs_set_stack_inode_gid(dst, src->st_gid);
 406         btrfs_set_stack_inode_mode(dst, src->st_mode);
 407         btrfs_set_stack_inode_rdev(dst, 0);
 408         btrfs_set_stack_inode_flags(dst, 0);
 409         btrfs_set_stack_timespec_sec(&dst->atime, src->st_atime);
 410         btrfs_set_stack_timespec_nsec(&dst->atime, 0);
 411         btrfs_set_stack_timespec_sec(&dst->ctime, src->st_ctime);
 412         btrfs_set_stack_timespec_nsec(&dst->ctime, 0);
 413         btrfs_set_stack_timespec_sec(&dst->mtime, src->st_mtime);
 414         btrfs_set_stack_timespec_nsec(&dst->mtime, 0);
 415         btrfs_set_stack_timespec_sec(&dst->otime, 0);
 416         btrfs_set_stack_timespec_nsec(&dst->otime, 0);
 417
 418         if (S_ISDIR(src->st_mode)) {
 419                 btrfs_set_stack_inode_size(dst, 0);
 420                 btrfs_set_stack_inode_nlink(dst, 1);
 421         }
 422         if (S_ISREG(src->st_mode)) {
 423                 btrfs_set_stack_inode_size(dst, (u64)src->st_size);
 424                 if (src->st_size <= BTRFS_MAX_INLINE_DATA_SIZE(root))
 425                         btrfs_set_stack_inode_nbytes(dst, src->st_size);
 426                 else {
 427                         blocks = src->st_size / sectorsize;
 428                         if (src->st_size % sectorsize)
 429                                 blocks += 1;
 430                         blocks *= sectorsize;
 431                         btrfs_set_stack_inode_nbytes(dst, blocks);
 432                 }
 433         }
 434         if (S_ISLNK(src->st_mode))
 435                 btrfs_set_stack_inode_nbytes(dst, src->st_size + 1);
 436
 437         return 0;
 438 }
 439
 440 static int directory_select(const struct direct *entry)
 441 {
 442         if ((strncmp(entry->d_name, ".", entry->d_reclen) == 0) ||
 443                 (strncmp(entry->d_name, "..", entry->d_reclen) == 0))
 444                 return 0;
 445         else
 446                 return 1;
 447 }
 448
 449 static void free_namelist(struct direct **files, int count)
 450 {
 451         int i;
 452
 453         if (count < 0)
 454                 return;
 455
 456         for (i = 0; i < count; ++i)
 457                 free(files[i]);
 458         free(files);
 459 }
 460
 461 static u64 calculate_dir_inode_size(char *dirname)
 462 {
 463         int count, i;
 464         struct direct **files, *cur_file;
 465         u64 dir_inode_size = 0;
 466
 467         count = scandir(dirname, &files, directory_select, NULL);
 468
 469         for (i = 0; i < count; i++) {
 470                 cur_file = files[i];
 471                 dir_inode_size += strlen(cur_file->d_name);
 472         }
 473
 474         free_namelist(files, count);
 475
 476         dir_inode_size *= 2;
 477         return dir_inode_size;
 478 }
 479
 480 static int add_inode_items(struct btrfs_trans_handle *trans,
 481                            struct btrfs_root *root,
 482                            struct stat *st, char *name,
 483                            u64 self_objectid, ino_t parent_inum,
 484                            int dir_index_cnt, struct btrfs_inode_item *inode_ret)
 485 {
 486         int ret;
 487         struct btrfs_key inode_key;
 488         struct btrfs_inode_item btrfs_inode;
 489         u64 objectid;
 490         u64 inode_size = 0;
 491         int name_len;
 492
 493         name_len = strlen(name);
 494         fill_inode_item(trans, root, &btrfs_inode, st);
 495         objectid = self_objectid;
 496
 497         if (S_ISDIR(st->st_mode)) {
 498                 inode_size = calculate_dir_inode_size(name);
 499                 btrfs_set_stack_inode_size(&btrfs_inode, inode_size);
 500         }
 501
 502         inode_key.objectid = objectid;
 503         inode_key.offset = 0;
 504         btrfs_set_key_type(&inode_key, BTRFS_INODE_ITEM_KEY);
 505
 506         ret = btrfs_insert_inode(trans, root, objectid, &btrfs_inode);
 507         if (ret)
 508                 goto fail;
 509
 510         ret = btrfs_insert_inode_ref(trans, root, name, name_len,
 511                                      objectid, parent_inum, dir_index_cnt);
 512         if (ret)
 513                 goto fail;
 514
 515         *inode_ret = btrfs_inode;
 516 fail:
 517         return ret;
 518 }
 519
 520 static int add_xattr_item(struct btrfs_trans_handle *trans,
 521                           struct btrfs_root *root, u64 objectid,
 522                           const char *file_name)
 523 {
 524         int ret;
 525         int cur_name_len;
 526         char xattr_list[XATTR_LIST_MAX];
 527         char *cur_name;
 528         char cur_value[XATTR_SIZE_MAX];
 529         char delimiter = '\0';
 530         char *next_location = xattr_list;
 531
 532         ret = llistxattr(file_name, xattr_list, XATTR_LIST_MAX);
 533         if (ret < 0) {
 534                 if(errno == ENOTSUP)
 535                         return 0;
 536                 fprintf(stderr, "get a list of xattr failed for %s\n",
 537                         file_name);
 538                 return ret;
 539         }
 540         if (ret == 0)
 541                 return ret;
 542
 543         cur_name = strtok(xattr_list, &delimiter);
 544         while (cur_name != NULL) {
 545                 cur_name_len = strlen(cur_name);
 546                 next_location += cur_name_len + 1;
 547
 548                 ret = getxattr(file_name, cur_name, cur_value, XATTR_SIZE_MAX);
 549                 if (ret < 0) {
 550                         if(errno == ENOTSUP)
 551                                 return 0;
 552                         fprintf(stderr, "get a xattr value failed for %s attr %s\n",
 553                                 file_name, cur_name);
 554                         return ret;
 555                 }
 556
 557                 ret = btrfs_insert_xattr_item(trans, root, cur_name,
 558                                               cur_name_len, cur_value,
 559                                               ret, objectid);
 560                 if (ret) {
 561                         fprintf(stderr, "insert a xattr item failed for %s\n",
 562                                 file_name);
 563                 }
 564
 565                 cur_name = strtok(next_location, &delimiter);
 566         }
 567
 568         return ret;
 569 }
 570 static int custom_alloc_extent(struct btrfs_root *root, u64 num_bytes,
 571                                u64 hint_byte, struct btrfs_key *ins)
 572 {
 573         u64 start;
 574         u64 end;
 575         u64 last = hint_byte;
 576         int ret;
 577         int wrapped = 0;
 578         struct btrfs_block_group_cache *cache;
 579
 580         while (1) {
 581                 ret = find_first_extent_bit(&root->fs_info->free_space_cache,
 582                                             last, &start, &end, EXTENT_DIRTY);
 583                 if (ret) {
 584                         if (wrapped++ == 0) {
 585                                 last = 0;
 586                                 continue;
 587                         } else {
 588                                 goto fail;
 589                         }
 590                 }
 591
 592                 start = max(last, start);
 593                 last = end + 1;
 594                 if (last - start < num_bytes)
 595                         continue;
 596
 597                 last = start + num_bytes;
 598                 if (test_range_bit(&root->fs_info->pinned_extents,
 599                                    start, last - 1, EXTENT_DIRTY, 0))
 600                         continue;
 601
 602                 cache = btrfs_lookup_block_group(root->fs_info, start);
 603                 BUG_ON(!cache);
 604                 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM ||
 605                     last > cache->key.objectid + cache->key.offset) {
 606                         last = cache->key.objectid + cache->key.offset;
 607                         continue;
 608                 }
 609
 610                 if (cache->flags & (BTRFS_BLOCK_GROUP_SYSTEM |
 611                             BTRFS_BLOCK_GROUP_METADATA)) {
 612                         last = cache->key.objectid + cache->key.offset;
 613                         continue;
 614                 }
 615
 616                 clear_extent_dirty(&root->fs_info->free_space_cache,
 617                                    start, start + num_bytes - 1, 0);
 618
 619                 ins->objectid = start;
 620                 ins->offset = num_bytes;
 621                 ins->type = BTRFS_EXTENT_ITEM_KEY;
 622                 return 0;
 623         }
 624 fail:
 625         fprintf(stderr, "not enough free space\n");
 626         return -ENOSPC;
 627 }
 628
 629 static int record_file_extent(struct btrfs_trans_handle *trans,
 630                               struct btrfs_root *root, u64 objectid,
 631                               struct btrfs_inode_item *inode,
 632                               u64 file_pos, u64 disk_bytenr,
 633                               u64 num_bytes)
 634 {
 635         int ret;
 636         struct btrfs_fs_info *info = root->fs_info;
 637         struct btrfs_root *extent_root = info->extent_root;
 638         struct extent_buffer *leaf;
 639         struct btrfs_file_extent_item *fi;
 640         struct btrfs_key ins_key;
 641         struct btrfs_path path;
 642         struct btrfs_extent_item *ei;
 643
 644         btrfs_init_path(&path);
 645
 646         ins_key.objectid = objectid;
 647         ins_key.offset = 0;
 648         btrfs_set_key_type(&ins_key, BTRFS_EXTENT_DATA_KEY);
 649         ret = btrfs_insert_empty_item(trans, root, &path, &ins_key,
 650                                       sizeof(*fi));
 651         if (ret)
 652                 goto fail;
 653         leaf = path.nodes[0];
 654         fi = btrfs_item_ptr(leaf, path.slots[0],
 655                             struct btrfs_file_extent_item);
 656         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
 657         btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
 658         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
 659         btrfs_set_file_extent_disk_num_bytes(leaf, fi, num_bytes);
 660         btrfs_set_file_extent_offset(leaf, fi, 0);
 661         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
 662         btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
 663         btrfs_set_file_extent_compression(leaf, fi, 0);
 664         btrfs_set_file_extent_encryption(leaf, fi, 0);
 665         btrfs_set_file_extent_other_encoding(leaf, fi, 0);
 666         btrfs_mark_buffer_dirty(leaf);
 667
 668         btrfs_release_path(&path);
 669
 670         ins_key.objectid = disk_bytenr;
 671         ins_key.offset = num_bytes;
 672         ins_key.type = BTRFS_EXTENT_ITEM_KEY;
 673
 674         ret = btrfs_insert_empty_item(trans, extent_root, &path,
 675                                 &ins_key, sizeof(*ei));
 676         if (ret == 0) {
 677                 leaf = path.nodes[0];
 678                 ei = btrfs_item_ptr(leaf, path.slots[0],
 679                                     struct btrfs_extent_item);
 680
 681                 btrfs_set_extent_refs(leaf, ei, 0);
 682                 btrfs_set_extent_generation(leaf, ei, trans->transid);
 683                 btrfs_set_extent_flags(leaf, ei, BTRFS_EXTENT_FLAG_DATA);
 684
 685                 btrfs_mark_buffer_dirty(leaf);
 686                 ret = btrfs_update_block_group(trans, root, disk_bytenr,
 687                                                num_bytes, 1, 0);
 688                 if (ret)
 689                         goto fail;
 690         } else if (ret != -EEXIST) {
 691                 goto fail;
 692         }
 693
 694         ret = btrfs_inc_extent_ref(trans, root, disk_bytenr, num_bytes, 0,
 695                                    root->root_key.objectid,
 696                                    objectid, 0);
 697 fail:
 698         btrfs_release_path(&path);
 699         return ret;
 700 }
 701
 702 static int add_symbolic_link(struct btrfs_trans_handle *trans,
 703                              struct btrfs_root *root,
 704                              u64 objectid, const char *path_name)
 705 {
 706         int ret;
 707         u64 sectorsize = root->sectorsize;
 708         char *buf = malloc(sectorsize);
 709
 710         ret = readlink(path_name, buf, sectorsize);
 711         if (ret <= 0) {
 712                 fprintf(stderr, "readlink failed for %s\n", path_name);
 713                 goto fail;
 714         }
 715         if (ret >= sectorsize) {
 716                 fprintf(stderr, "symlink too long for %s", path_name);
 717                 ret = -1;
 718                 goto fail;
 719         }
 720
 721         buf[ret] = '\0'; /* readlink does not do it for us */
 722         ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
 723                                          buf, ret + 1);
 724 fail:
 725         free(buf);
 726         return ret;
 727 }
 728
 729 static int add_file_items(struct btrfs_trans_handle *trans,
 730                           struct btrfs_root *root,
 731                           struct btrfs_inode_item *btrfs_inode, u64 objectid,
 732                           ino_t parent_inum, struct stat *st,
 733                           const char *path_name, int out_fd)
 734 {
 735         int ret = -1;
 736         ssize_t ret_read;
 737         u64 bytes_read = 0;
 738         char *buffer = NULL;
 739         struct btrfs_key key;
 740         int blocks;
 741         u32 sectorsize = root->sectorsize;
 742         u64 first_block = 0;
 743         u64 num_blocks = 0;
 744         int fd;
 745
 746         fd = open(path_name, O_RDONLY);
 747         if (fd == -1) {
 748                 fprintf(stderr, "%s open failed\n", path_name);
 749                 return ret;
 750         }
 751
 752         blocks = st->st_size / sectorsize;
 753         if (st->st_size % sectorsize)
 754                 blocks += 1;
 755
 756         if (st->st_size <= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 757                 buffer = malloc(st->st_size);
 758                 ret_read = pread64(fd, buffer, st->st_size, bytes_read);
 759                 if (ret_read == -1) {
 760                         fprintf(stderr, "%s read failed\n", path_name);
 761                         goto end;
 762                 }
 763
 764                 ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
 765                                                  buffer, st->st_size);
 766                 goto end;
 767         }
 768
 769         ret = custom_alloc_extent(root, blocks * sectorsize, 0, &key);
 770         if (ret)
 771                 goto end;
 772
 773         first_block = key.objectid;
 774         bytes_read = 0;
 775         buffer = malloc(sectorsize);
 776
 777         do {
 778                 memset(buffer, 0, sectorsize);
 779                 ret_read = pread64(fd, buffer, sectorsize, bytes_read);
 780                 if (ret_read == -1) {
 781                         fprintf(stderr, "%s read failed\n", path_name);
 782                         goto end;
 783                 }
 784
 785                 ret = pwrite64(out_fd, buffer, sectorsize,
 786                                first_block + bytes_read);
 787                 if (ret != sectorsize) {
 788                         fprintf(stderr, "output file write failed\n");
 789                         goto end;
 790                 }
 791
 792                 /* checksum for file data */
 793                 ret = btrfs_csum_file_block(trans, root->fs_info->csum_root,
 794                                 first_block + (blocks * sectorsize),
 795                                 first_block + bytes_read,
 796                                 buffer, sectorsize);
 797                 if (ret) {
 798                         fprintf(stderr, "%s checksum failed\n", path_name);
 799                         goto end;
 800                 }
 801
 802                 bytes_read += ret_read;
 803                 num_blocks++;
 804         } while (ret_read == sectorsize);
 805
 806         if (num_blocks > 0) {
 807                 ret = record_file_extent(trans, root, objectid, btrfs_inode,
 808                                          first_block, first_block,
 809                                          blocks * sectorsize);
 810                 if (ret)
 811                         goto end;
 812         }
 813
 814 end:
 815         if (buffer)
 816                 free(buffer);
 817         close(fd);
 818         return ret;
 819 }
 820
 821 static char *make_path(char *dir, char *name)
 822 {
 823         char *path;
 824
 825         path = malloc(strlen(dir) + strlen(name) + 2);
 826         if (!path)
 827                 return NULL;
 828         strcpy(path, dir);
 829         if (dir[strlen(dir) - 1] != '/')
 830                 strcat(path, "/");
 831         strcat(path, name);
 832         return path;
 833 }
 834
 835 static int traverse_directory(struct btrfs_trans_handle *trans,
 836                               struct btrfs_root *root, char *dir_name,
 837                               struct directory_name_entry *dir_head, int out_fd)
 838 {
 839         int ret = 0;
 840
 841         struct btrfs_inode_item cur_inode;
 842         struct btrfs_inode_item *inode_item;
 843         int count, i, dir_index_cnt;
 844         struct direct **files;
 845         struct stat st;
 846         struct directory_name_entry *dir_entry, *parent_dir_entry;
 847         struct direct *cur_file;
 848         ino_t parent_inum, cur_inum;
 849         ino_t highest_inum = 0;
 850         char *parent_dir_name;
 851         struct btrfs_path path;
 852         struct extent_buffer *leaf;
 853         struct btrfs_key root_dir_key;
 854         u64 root_dir_inode_size = 0;
 855
 856         /* Add list for source directory */
 857         dir_entry = malloc(sizeof(struct directory_name_entry));
 858         dir_entry->dir_name = dir_name;
 859         dir_entry->path = strdup(dir_name);
 860
 861         parent_inum = highest_inum + BTRFS_FIRST_FREE_OBJECTID;
 862         dir_entry->inum = parent_inum;
 863         list_add_tail(&dir_entry->list, &dir_head->list);
 864
 865         btrfs_init_path(&path);
 866
 867         root_dir_key.objectid = btrfs_root_dirid(&root->root_item);
 868         root_dir_key.offset = 0;
 869         btrfs_set_key_type(&root_dir_key, BTRFS_INODE_ITEM_KEY);
 870         ret = btrfs_lookup_inode(trans, root, &path, &root_dir_key, 1);
 871         if (ret) {
 872                 fprintf(stderr, "root dir lookup error\n");
 873                 return -1;
 874         }
 875
 876         leaf = path.nodes[0];
 877         inode_item = btrfs_item_ptr(leaf, path.slots[0],
 878                                     struct btrfs_inode_item);
 879
 880         root_dir_inode_size = calculate_dir_inode_size(dir_name);
 881         btrfs_set_inode_size(leaf, inode_item, root_dir_inode_size);
 882         btrfs_mark_buffer_dirty(leaf);
 883
 884         btrfs_release_path(&path);
 885
 886         do {
 887                 parent_dir_entry = list_entry(dir_head->list.next,
 888                                               struct directory_name_entry,
 889                                               list);
 890                 list_del(&parent_dir_entry->list);
 891
 892                 parent_inum = parent_dir_entry->inum;
 893                 parent_dir_name = parent_dir_entry->dir_name;
 894                 if (chdir(parent_dir_entry->path)) {
 895                         fprintf(stderr, "chdir error for %s\n",
 896                                 parent_dir_name);
 897                         goto fail_no_files;
 898                 }
 899
 900                 count = scandir(parent_dir_entry->path, &files,
 901                                 directory_select, NULL);
 902                 if (count == -1)
 903                 {
 904                         fprintf(stderr, "scandir for %s failed: %s\n",
 905                                 parent_dir_name, strerror (errno));
 906                         goto fail;
 907                 }
 908
 909                 for (i = 0; i < count; i++) {
 910                         cur_file = files[i];
 911
 912                         if (lstat(cur_file->d_name, &st) == -1) {
 913                                 fprintf(stderr, "lstat failed for file %s\n",
 914                                         cur_file->d_name);
 915                                 goto fail;
 916                         }
 917
 918                         cur_inum = ++highest_inum + BTRFS_FIRST_FREE_OBJECTID;
 919                         ret = add_directory_items(trans, root,
 920                                                   cur_inum, parent_inum,
 921                                                   cur_file->d_name,
 922                                                   &st, &dir_index_cnt);
 923                         if (ret) {
 924                                 fprintf(stderr, "add_directory_items failed\n");
 925                                 goto fail;
 926                         }
 927
 928                         ret = add_inode_items(trans, root, &st,
 929                                               cur_file->d_name, cur_inum,
 930                                               parent_inum, dir_index_cnt,
 931                                               &cur_inode);
 932                         if (ret) {
 933                                 fprintf(stderr, "add_inode_items failed\n");
 934                                 goto fail;
 935                         }
 936
 937                         ret = add_xattr_item(trans, root,
 938                                              cur_inum, cur_file->d_name);
 939                         if (ret) {
 940                                 fprintf(stderr, "add_xattr_item failed\n");
 941                                 if(ret != -ENOTSUP)
 942                                         goto fail;
 943                         }
 944
 945                         if (S_ISDIR(st.st_mode)) {
 946                                 dir_entry = malloc(sizeof(struct directory_name_entry));
 947                                 dir_entry->dir_name = cur_file->d_name;
 948                                 dir_entry->path = make_path(parent_dir_entry->path,
 949                                                             cur_file->d_name);
 950                                 dir_entry->inum = cur_inum;
 951                                 list_add_tail(&dir_entry->list, &dir_head->list);
 952                         } else if (S_ISREG(st.st_mode)) {
 953                                 ret = add_file_items(trans, root, &cur_inode,
 954                                                      cur_inum, parent_inum, &st,
 955                                                      cur_file->d_name, out_fd);
 956                                 if (ret) {
 957                                         fprintf(stderr, "add_file_items failed\n");
 958                                         goto fail;
 959                                 }
 960                         } else if (S_ISLNK(st.st_mode)) {
 961                                 ret = add_symbolic_link(trans, root,
 962                                                         cur_inum, cur_file->d_name);
 963                                 if (ret) {
 964                                         fprintf(stderr, "add_symbolic_link failed\n");
 965                                         goto fail;
 966                                 }
 967                         }
 968                 }
 969
 970                 free_namelist(files, count);
 971                 free(parent_dir_entry->path);
 972                 free(parent_dir_entry);
 973
 974                 index_cnt = 2;
 975
 976         } while (!list_empty(&dir_head->list));
 977
 978         return 0;
 979 fail:
 980         free_namelist(files, count);
 981 fail_no_files:
 982         free(parent_dir_entry->path);
 983         free(parent_dir_entry);
 984         return -1;
 985 }
 986
 987 static int open_target(char *output_name)
 988 {
 989         int output_fd;
 990         output_fd = open(output_name, O_CREAT | O_RDWR | O_TRUNC,
 991                          S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH);
 992
 993         return output_fd;
 994 }
 995
 996 static int create_chunks(struct btrfs_trans_handle *trans,
 997                          struct btrfs_root *root, u64 num_of_meta_chunks,
 998                          u64 size_of_data)
 999 {
1000         u64 chunk_start;
1001         u64 chunk_size;
1002         u64 meta_type = BTRFS_BLOCK_GROUP_METADATA;
1003         u64 data_type = BTRFS_BLOCK_GROUP_DATA;
1004         u64 minimum_data_chunk_size = 8 * 1024 * 1024;
1005         u64 i;
1006         int ret;
1007
1008         for (i = 0; i < num_of_meta_chunks; i++) {
1009                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
1010                                         &chunk_start, &chunk_size, meta_type);
1011                 BUG_ON(ret);
1012                 ret = btrfs_make_block_group(trans, root->fs_info->extent_root, 0,
1013                                              meta_type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1014                                              chunk_start, chunk_size);
1015                 BUG_ON(ret);
1016                 set_extent_dirty(&root->fs_info->free_space_cache,
1017                                  chunk_start, chunk_start + chunk_size - 1, 0);
1018         }
1019
1020         if (size_of_data < minimum_data_chunk_size)
1021                 size_of_data = minimum_data_chunk_size;
1022         ret = btrfs_alloc_data_chunk(trans, root->fs_info->extent_root,
1023                                      &chunk_start, size_of_data, data_type);
1024         BUG_ON(ret);
1025         ret = btrfs_make_block_group(trans, root->fs_info->extent_root, 0,
1026                                      data_type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1027                                      chunk_start, size_of_data);
1028         BUG_ON(ret);
1029         set_extent_dirty(&root->fs_info->free_space_cache,
1030                          chunk_start, chunk_start + size_of_data - 1, 0);
1031         return ret;
1032 }
1033
1034 static int make_image(char *source_dir, struct btrfs_root *root, int out_fd)
1035 {
1036         int ret;
1037         struct btrfs_trans_handle *trans;
1038
1039         struct stat root_st;
1040
1041         struct directory_name_entry dir_head;
1042
1043         struct directory_name_entry *dir_entry = NULL;
1044
1045         ret = lstat(source_dir, &root_st);
1046         if (ret) {
1047                 fprintf(stderr, "unable to lstat the %s\n", source_dir);
1048                 goto fail;
1049         }
1050
1051         INIT_LIST_HEAD(&dir_head.list);
1052
1053         trans = btrfs_start_transaction(root, 1);
1054         ret = traverse_directory(trans, root, source_dir, &dir_head, out_fd);
1055         if (ret) {
1056                 fprintf(stderr, "unable to traverse_directory\n");
1057                 goto fail;
1058         }
1059         btrfs_commit_transaction(trans, root);
1060
1061         printf("Making image is completed.\n");
1062         return 0;
1063 fail:
1064         while (!list_empty(&dir_head.list)) {
1065                 dir_entry = list_entry(dir_head.list.next,
1066                                        struct directory_name_entry, list);
1067                 list_del(&dir_entry->list);
1068                 free(dir_entry);
1069         }
1070         fprintf(stderr, "Making image is aborted.\n");
1071         return -1;
1072 }
1073
1074 /*
1075  * This ignores symlinks with unreadable targets and subdirs that can't
1076  * be read.  It's a best-effort to give a rough estimate of the size of
1077  * a subdir.  It doesn't guarantee that prepopulating btrfs from this
1078  * tree won't still run out of space.
1079  *
1080  * The rounding up to 4096 is questionable.  Previous code used du -B 4096.
1081  */
1082 static u64 global_total_size;
1083 static int ftw_add_entry_size(const char *fpath, const struct stat *st,
1084                               int type)
1085 {
1086         if (type == FTW_F || type == FTW_D)
1087                 global_total_size += round_up(st->st_size, 4096);
1088
1089         return 0;
1090 }
1091
1092 static u64 size_sourcedir(char *dir_name, u64 sectorsize,
1093                           u64 *num_of_meta_chunks_ret, u64 *size_of_data_ret)
1094 {
1095         u64 dir_size = 0;
1096         u64 total_size = 0;
1097         int ret;
1098         u64 default_chunk_size = 8 * 1024 * 1024;       /* 8MB */
1099         u64 allocated_meta_size = 8 * 1024 * 1024;      /* 8MB */
1100         u64 allocated_total_size = 20 * 1024 * 1024;    /* 20MB */
1101         u64 num_of_meta_chunks = 0;
1102         u64 num_of_allocated_meta_chunks =
1103                         allocated_meta_size / default_chunk_size;
1104
1105         global_total_size = 0;
1106         ret = ftw(dir_name, ftw_add_entry_size, 10);
1107         dir_size = global_total_size;
1108         if (ret < 0) {
1109                 fprintf(stderr, "ftw subdir walk of '%s' failed: %s\n",
1110                         dir_name, strerror(errno));
1111                 exit(1);
1112         }
1113
1114         num_of_meta_chunks = (dir_size / 2) / default_chunk_size;
1115         if (((dir_size / 2) % default_chunk_size) != 0)
1116                 num_of_meta_chunks++;
1117         if (num_of_meta_chunks <= num_of_allocated_meta_chunks)
1118                 num_of_meta_chunks = 0;
1119         else
1120                 num_of_meta_chunks -= num_of_allocated_meta_chunks;
1121
1122         total_size = allocated_total_size + dir_size +
1123                      (num_of_meta_chunks * default_chunk_size);
1124
1125         *num_of_meta_chunks_ret = num_of_meta_chunks;
1126
1127         return total_size;
1128 }
1129
1130 static int zero_output_file(int out_fd, u64 size, u32 sectorsize)
1131 {
1132         int len = sectorsize;
1133         int loop_num = size / sectorsize;
1134         u64 location = 0;
1135         char *buf = malloc(len);
1136         int ret = 0, i;
1137         ssize_t written;
1138
1139         if (!buf)
1140                 return -ENOMEM;
1141         memset(buf, 0, len);
1142         for (i = 0; i < loop_num; i++) {
1143                 written = pwrite64(out_fd, buf, len, location);
1144                 if (written != len)
1145                         ret = -EIO;
1146                 location += sectorsize;
1147         }
1148         free(buf);
1149         return ret;
1150 }
1151
1152 static int check_leaf_or_node_size(u32 size, u32 sectorsize)
1153 {
1154         if (size < sectorsize) {
1155                 fprintf(stderr,
1156                         "Illegal leafsize (or nodesize) %u (smaller than %u)\n",
1157                         size, sectorsize);
1158                 return -1;
1159         } else if (size > BTRFS_MAX_METADATA_BLOCKSIZE) {
1160                 fprintf(stderr,
1161                         "Illegal leafsize (or nodesize) %u (larger than %u)\n",
1162                         size, BTRFS_MAX_METADATA_BLOCKSIZE);
1163                 return -1;
1164         } else if (size & (sectorsize - 1)) {
1165                 fprintf(stderr,
1166                         "Illegal leafsize (or nodesize) %u (not align to %u)\n",
1167                         size, sectorsize);
1168                 return -1;
1169         }
1170         return 0;
1171 }
1172
1173 static int is_ssd(const char *file)
1174 {
1175         blkid_probe probe;
1176         char wholedisk[32];
1177         char sysfs_path[PATH_MAX];
1178         dev_t devno;
1179         int fd;
1180         char rotational;
1181         int ret;
1182
1183         probe = blkid_new_probe_from_filename(file);
1184         if (!probe)
1185                 return 0;
1186
1187         /* Device number of this disk (possibly a partition) */
1188         devno = blkid_probe_get_devno(probe);
1189         if (!devno) {
1190                 blkid_free_probe(probe);
1191                 return 0;
1192         }
1193
1194         /* Get whole disk name (not full path) for this devno */
1195         ret = blkid_devno_to_wholedisk(devno,
1196                         wholedisk, sizeof(wholedisk), NULL);
1197         if (ret) {
1198                 blkid_free_probe(probe);
1199                 return 0;
1200         }
1201
1202         snprintf(sysfs_path, PATH_MAX, "/sys/block/%s/queue/rotational",
1203                  wholedisk);
1204
1205         blkid_free_probe(probe);
1206
1207         fd = open(sysfs_path, O_RDONLY);
1208         if (fd < 0) {
1209                 return 0;
1210         }
1211
1212         if (read(fd, &rotational, sizeof(char)) < sizeof(char)) {
1213                 close(fd);
1214                 return 0;
1215         }
1216         close(fd);
1217
1218         return !atoi((const char *)&rotational);
1219 }
1220
1221 #define BTRFS_FEATURE_LIST_ALL          (1ULL << 63)
1222
1223 static const struct btrfs_fs_feature {
1224         const char *name;
1225         u64 flag;
1226         const char *desc;
1227 } mkfs_features[] = {
1228         { "mixed-bg", BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS,
1229                 "mixed data and metadata block groups" },
1230         { "extref", BTRFS_FEATURE_INCOMPAT_EXTENDED_IREF,
1231                 "increased hardlink limit per file to 65536" },
1232         { "raid56", BTRFS_FEATURE_INCOMPAT_RAID56,
1233                 "raid56 extended format" },
1234         { "skinny-metadata", BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA,
1235                 "reduced-size metadata extent refs" },
1236         /* Keep this one last */
1237         { "list-all", BTRFS_FEATURE_LIST_ALL, NULL }
1238 };
1239
1240 static void list_all_fs_features(void)
1241 {
1242         int i;
1243
1244         fprintf(stderr, "Filesystem features available at mkfs time:\n");
1245         for (i = 0; i < ARRAY_SIZE(mkfs_features) - 1; i++) {
1246                 fprintf(stderr, "%-20s- %s (0x%llx)\n",
1247                                 mkfs_features[i].name,
1248                                 mkfs_features[i].desc,
1249                                 mkfs_features[i].flag);
1250         }
1251 }
1252
1253 static int parse_one_fs_feature(const char *name, u64 *flags)
1254 {
1255         int i;
1256         int found = 0;
1257
1258         for (i = 0; i < ARRAY_SIZE(mkfs_features); i++) {
1259                 if (!strcmp(mkfs_features[i].name, name)) {
1260                         *flags |= mkfs_features[i].flag;
1261                         found = 1;
1262                 }
1263         }
1264
1265         return !found;
1266 }
1267
1268 static void process_fs_features(u64 flags)
1269 {
1270         int i;
1271
1272         for (i = 0; i < ARRAY_SIZE(mkfs_features); i++) {
1273                 if (flags & mkfs_features[i].flag) {
1274                         fprintf(stderr,
1275                                 "Turning ON incompat feature '%s': %s\n",
1276                                 mkfs_features[i].name,
1277                                 mkfs_features[i].desc);
1278                 }
1279         }
1280 }
1281
1282
1283 /*
1284  * Return NULL if all features were parsed fine, otherwise return the name of
1285  * the first unparsed.
1286  */
1287 static char* parse_fs_features(char *namelist, u64 *flags)
1288 {
1289         char *this_char;
1290         char *save_ptr = NULL; /* Satisfy static checkers */
1291
1292         for (this_char = strtok_r(namelist, ",", &save_ptr);
1293              this_char != NULL;
1294              this_char = strtok_r(NULL, ",", &save_ptr)) {
1295                 if (parse_one_fs_feature(this_char, flags))
1296                         return this_char;
1297         }
1298
1299         return NULL;
1300 }
1301
1302 int main(int ac, char **av)
1303 {
1304         char *file;
1305         struct btrfs_root *root;
1306         struct btrfs_trans_handle *trans;
1307         char *label = NULL;
1308         char *first_file;
1309         u64 block_count = 0;
1310         u64 dev_block_count = 0;
1311         u64 blocks[7];
1312         u64 alloc_start = 0;
1313         u64 metadata_profile = 0;
1314         u64 data_profile = 0;
1315         u32 leafsize = sysconf(_SC_PAGESIZE);
1316         u32 sectorsize = 4096;
1317         u32 nodesize = leafsize;
1318         u32 stripesize = 4096;
1319         int zero_end = 1;
1320         int option_index = 0;
1321         int fd;
1322         int ret;
1323         int i;
1324         int mixed = 0;
1325         int data_profile_opt = 0;
1326         int metadata_profile_opt = 0;
1327         int discard = 1;
1328         int ssd = 0;
1329         int force_overwrite = 0;
1330
1331         char *source_dir = NULL;
1332         int source_dir_set = 0;
1333         u64 num_of_meta_chunks = 0;
1334         u64 size_of_data = 0;
1335         u64 source_dir_size = 0;
1336         int dev_cnt = 0;
1337         int saved_optind;
1338         char estr[100];
1339         u64 features = 0;
1340
1341         while(1) {
1342                 int c;
1343                 c = getopt_long(ac, av, "A:b:fl:n:s:m:d:L:O:r:VMK",
1344                                 long_options, &option_index);
1345                 if (c < 0)
1346                         break;
1347                 switch(c) {
1348                         case 'A':
1349                                 alloc_start = parse_size(optarg);
1350                                 break;
1351                         case 'f':
1352                                 force_overwrite = 1;
1353                                 break;
1354                         case 'd':
1355                                 data_profile = parse_profile(optarg);
1356                                 data_profile_opt = 1;
1357                                 break;
1358                         case 'l':
1359                         case 'n':
1360                                 nodesize = parse_size(optarg);
1361                                 leafsize = parse_size(optarg);
1362                                 break;
1363                         case 'L':
1364                                 label = parse_label(optarg);
1365                                 break;
1366                         case 'm':
1367                                 metadata_profile = parse_profile(optarg);
1368                                 metadata_profile_opt = 1;
1369                                 break;
1370                         case 'M':
1371                                 mixed = 1;
1372                                 break;
1373                         case 'O': {
1374                                 char *orig = strdup(optarg);
1375                                 char *tmp = orig;
1376
1377                                 tmp = parse_fs_features(tmp, &features);
1378                                 if (tmp) {
1379                                         fprintf(stderr,
1380                                                 "Unrecognized filesystem feature '%s'\n",
1381                                                         tmp);
1382                                         free(orig);
1383                                         exit(1);
1384                                 }
1385                                 free(orig);
1386                                 if (features & BTRFS_FEATURE_LIST_ALL) {
1387                                         list_all_fs_features();
1388                                         exit(0);
1389                                 }
1390                                 break;
1391                                 }
1392                         case 's':
1393                                 sectorsize = parse_size(optarg);
1394                                 break;
1395                         case 'b':
1396                                 block_count = parse_size(optarg);
1397                                 if (block_count <= 1024*1024*1024) {
1398                                         printf("SMALL VOLUME: forcing mixed "
1399                                                "metadata/data groups\n");
1400                                         mixed = 1;
1401                                 }
1402                                 zero_end = 0;
1403                                 break;
1404                         case 'V':
1405                                 print_version();
1406                                 break;
1407                         case 'r':
1408                                 source_dir = optarg;
1409                                 source_dir_set = 1;
1410                                 break;
1411                         case 'K':
1412                                 discard = 0;
1413                                 break;
1414                         default:
1415                                 print_usage();
1416                 }
1417         }
1418         sectorsize = max(sectorsize, (u32)sysconf(_SC_PAGESIZE));
1419         if (check_leaf_or_node_size(leafsize, sectorsize))
1420                 exit(1);
1421         if (check_leaf_or_node_size(nodesize, sectorsize))
1422                 exit(1);
1423         saved_optind = optind;
1424         dev_cnt = ac - optind;
1425         if (dev_cnt == 0)
1426                 print_usage();
1427
1428         if (source_dir_set && dev_cnt > 1) {
1429                 fprintf(stderr,
1430                         "The -r option is limited to a single device\n");
1431                 exit(1);
1432         }
1433         while (dev_cnt-- > 0) {
1434                 file = av[optind++];
1435                 if (is_block_device(file))
1436                         if (test_dev_for_mkfs(file, force_overwrite, estr)) {
1437                                 fprintf(stderr, "Error: %s", estr);
1438                                 exit(1);
1439                         }
1440         }
1441
1442         optind = saved_optind;
1443         dev_cnt = ac - optind;
1444
1445         file = av[optind++];
1446         ssd = is_ssd(file);
1447
1448         if (is_vol_small(file)) {
1449                 printf("SMALL VOLUME: forcing mixed metadata/data groups\n");
1450                 mixed = 1;
1451                 if (metadata_profile != data_profile) {
1452                         if (metadata_profile_opt || data_profile_opt) {
1453                                 fprintf(stderr,
1454         "With mixed block groups data and metadata profiles must be the same\n");
1455                                 exit(1);
1456                         }
1457                 }
1458         }
1459         /*
1460         * Set default profiles according to number of added devices.
1461         * For mixed groups defaults are single/single.
1462         */
1463         if (!mixed) {
1464                 if (!metadata_profile_opt) {
1465                         if (dev_cnt == 1 && ssd)
1466                                 printf("Detected a SSD, turning off metadata "
1467                                 "duplication.  Mkfs with -m dup if you want to "
1468                                 "force metadata duplication.\n");
1469
1470                         metadata_profile = (dev_cnt > 1) ?
1471                                         BTRFS_BLOCK_GROUP_RAID1 : (ssd) ?
1472                                         0: BTRFS_BLOCK_GROUP_DUP;
1473                 }
1474                 if (!data_profile_opt) {
1475                         data_profile = (dev_cnt > 1) ?
1476                                 BTRFS_BLOCK_GROUP_RAID0 : 0; /* raid0 or single */
1477                 }
1478         } else {
1479                 metadata_profile = 0;
1480                 data_profile = 0;
1481         }
1482
1483         ret = test_num_disk_vs_raid(metadata_profile, data_profile,
1484                         dev_cnt, mixed, estr);
1485         if (ret) {
1486                 fprintf(stderr, "Error: %s\n", estr);
1487                 exit(1);
1488         }
1489
1490         /* if we are here that means all devs are good to btrfsify */
1491         printf("\nWARNING! - %s IS EXPERIMENTAL\n", BTRFS_BUILD_VERSION);
1492         printf("WARNING! - see http://btrfs.wiki.kernel.org before using\n\n");
1493
1494         dev_cnt--;
1495
1496         if (!source_dir_set) {
1497                 /*
1498                  * open without O_EXCL so that the problem should not
1499                  * occur by the following processing.
1500                  * (btrfs_register_one_device() fails if O_EXCL is on)
1501                  */
1502                 fd = open(file, O_RDWR);
1503                 if (fd < 0) {
1504                         fprintf(stderr, "unable to open %s: %s\n", file,
1505                                 strerror(errno));
1506                         exit(1);
1507                 }
1508                 first_file = file;
1509                 ret = btrfs_prepare_device(fd, file, zero_end, &dev_block_count,
1510                                            block_count, &mixed, discard);
1511                 if (block_count && block_count > dev_block_count) {
1512                         fprintf(stderr, "%s is smaller than requested size\n", file);
1513                         exit(1);
1514                 }
1515         } else {
1516                 fd = open_target(file);
1517                 if (fd < 0) {
1518                         fprintf(stderr, "unable to open the %s\n", file);
1519                         exit(1);
1520                 }
1521
1522                 first_file = file;
1523                 source_dir_size = size_sourcedir(source_dir, sectorsize,
1524                                              &num_of_meta_chunks, &size_of_data);
1525                 if(block_count < source_dir_size)
1526                         block_count = source_dir_size;
1527                 ret = zero_output_file(fd, block_count, sectorsize);
1528                 if (ret) {
1529                         fprintf(stderr, "unable to zero the output file\n");
1530                         exit(1);
1531                 }
1532                 /* our "device" is the new image file */
1533                 dev_block_count = block_count;
1534         }
1535
1536         /* To create the first block group and chunk 0 in make_btrfs */
1537         if (dev_block_count < BTRFS_MKFS_SYSTEM_GROUP_SIZE) {
1538                 fprintf(stderr, "device is too small to make filesystem\n");
1539                 exit(1);
1540         }
1541
1542         blocks[0] = BTRFS_SUPER_INFO_OFFSET;
1543         for (i = 1; i < 7; i++) {
1544                 blocks[i] = BTRFS_SUPER_INFO_OFFSET + 1024 * 1024 +
1545                         leafsize * i;
1546         }
1547
1548         /*
1549          * FS features that can be set by other means than -O
1550          * just set the bit here
1551          */
1552         if (mixed)
1553                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS;
1554
1555         if ((data_profile | metadata_profile) &
1556             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
1557                 features |= BTRFS_FEATURE_INCOMPAT_RAID56;
1558         }
1559
1560         process_fs_features(features);
1561
1562         ret = make_btrfs(fd, file, label, blocks, dev_block_count,
1563                          nodesize, leafsize,
1564                          sectorsize, stripesize, features);
1565         if (ret) {
1566                 fprintf(stderr, "error during mkfs: %s\n", strerror(-ret));
1567                 exit(1);
1568         }
1569
1570         root = open_ctree(file, 0, O_RDWR);
1571         if (!root) {
1572                 fprintf(stderr, "Open ctree failed\n");
1573                 close(fd);
1574                 exit(1);
1575         }
1576         root->fs_info->alloc_start = alloc_start;
1577
1578         ret = make_root_dir(root, mixed);
1579         if (ret) {
1580                 fprintf(stderr, "failed to setup the root directory\n");
1581                 exit(1);
1582         }
1583
1584         trans = btrfs_start_transaction(root, 1);
1585
1586         if (dev_cnt == 0)
1587                 goto raid_groups;
1588
1589         btrfs_register_one_device(file);
1590
1591         zero_end = 1;
1592         while (dev_cnt-- > 0) {
1593                 int old_mixed = mixed;
1594
1595                 file = av[optind++];
1596
1597                 /*
1598                  * open without O_EXCL so that the problem should not
1599                  * occur by the following processing.
1600                  * (btrfs_register_one_device() fails if O_EXCL is on)
1601                  */
1602                 fd = open(file, O_RDWR);
1603                 if (fd < 0) {
1604                         fprintf(stderr, "unable to open %s: %s\n", file,
1605                                 strerror(errno));
1606                         exit(1);
1607                 }
1608                 ret = btrfs_device_already_in_root(root, fd,
1609                                                    BTRFS_SUPER_INFO_OFFSET);
1610                 if (ret) {
1611                         fprintf(stderr, "skipping duplicate device %s in FS\n",
1612                                 file);
1613                         close(fd);
1614                         continue;
1615                 }
1616                 ret = btrfs_prepare_device(fd, file, zero_end, &dev_block_count,
1617                                            block_count, &mixed, discard);
1618                 mixed = old_mixed;
1619                 BUG_ON(ret);
1620
1621                 ret = btrfs_add_to_fsid(trans, root, fd, file, dev_block_count,
1622                                         sectorsize, sectorsize, sectorsize);
1623                 BUG_ON(ret);
1624                 btrfs_register_one_device(file);
1625         }
1626
1627 raid_groups:
1628         if (!source_dir_set) {
1629                 ret = create_raid_groups(trans, root, data_profile,
1630                                  data_profile_opt, metadata_profile,
1631                                  metadata_profile_opt, mixed, ssd);
1632                 BUG_ON(ret);
1633         }
1634
1635         ret = create_data_reloc_tree(trans, root);
1636         BUG_ON(ret);
1637
1638         printf("fs created label %s on %s\n\tnodesize %u leafsize %u "
1639             "sectorsize %u size %s\n",
1640             label, first_file, nodesize, leafsize, sectorsize,
1641             pretty_size(btrfs_super_total_bytes(root->fs_info->super_copy)));
1642
1643         printf("%s\n", BTRFS_BUILD_VERSION);
1644         btrfs_commit_transaction(trans, root);
1645
1646         if (source_dir_set) {
1647                 trans = btrfs_start_transaction(root, 1);
1648                 ret = create_chunks(trans, root,
1649                                     num_of_meta_chunks, size_of_data);
1650                 BUG_ON(ret);
1651                 btrfs_commit_transaction(trans, root);
1652
1653                 ret = make_image(source_dir, root, fd);
1654                 BUG_ON(ret);
1655         }
1656
1657         ret = close_ctree(root);
1658         BUG_ON(ret);
1659         free(label);
1660         return 0;
1661 }