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