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(const 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(const 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(const 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, const 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(const char *dir, const 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, const 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         const 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(const char *source_dir, struct btrfs_root *root,
1022                 int out_fd)
1023 {
1024         int ret;
1025         struct btrfs_trans_handle *trans;
1026         struct stat root_st;
1027         struct directory_name_entry dir_head;
1028         struct directory_name_entry *dir_entry = NULL;
1029
1030         ret = lstat(source_dir, &root_st);
1031         if (ret) {
1032                 error("unable to lstat %s: %s", source_dir, strerror(errno));
1033                 ret = -errno;
1034                 goto out;
1035         }
1036
1037         INIT_LIST_HEAD(&dir_head.list);
1038
1039         trans = btrfs_start_transaction(root, 1);
1040         ret = traverse_directory(trans, root, source_dir, &dir_head, out_fd);
1041         if (ret) {
1042                 error("unable to traverse directory %s: %d", source_dir, ret);
1043                 goto fail;
1044         }
1045         ret = btrfs_commit_transaction(trans, root);
1046         if (ret) {
1047                 error("transaction commit failed: %d", ret);
1048                 goto out;
1049         }
1050
1051         if (verbose)
1052                 printf("Making image is completed.\n");
1053         return 0;
1054 fail:
1055         while (!list_empty(&dir_head.list)) {
1056                 dir_entry = list_entry(dir_head.list.next,
1057                                        struct directory_name_entry, list);
1058                 list_del(&dir_entry->list);
1059                 free(dir_entry);
1060         }
1061 out:
1062         return ret;
1063 }
1064
1065 /*
1066  * This ignores symlinks with unreadable targets and subdirs that can't
1067  * be read.  It's a best-effort to give a rough estimate of the size of
1068  * a subdir.  It doesn't guarantee that prepopulating btrfs from this
1069  * tree won't still run out of space.
1070  */
1071 static u64 global_total_size;
1072 static u64 fs_block_size;
1073 static int ftw_add_entry_size(const char *fpath, const struct stat *st,
1074                               int type)
1075 {
1076         if (type == FTW_F || type == FTW_D)
1077                 global_total_size += round_up(st->st_size, fs_block_size);
1078
1079         return 0;
1080 }
1081
1082 static u64 size_sourcedir(const char *dir_name, u64 sectorsize,
1083                           u64 *num_of_meta_chunks_ret, u64 *size_of_data_ret)
1084 {
1085         u64 dir_size = 0;
1086         u64 total_size = 0;
1087         int ret;
1088         u64 default_chunk_size = 8 * 1024 * 1024;       /* 8MB */
1089         u64 allocated_meta_size = 8 * 1024 * 1024;      /* 8MB */
1090         u64 allocated_total_size = 20 * 1024 * 1024;    /* 20MB */
1091         u64 num_of_meta_chunks = 0;
1092         u64 num_of_data_chunks = 0;
1093         u64 num_of_allocated_meta_chunks =
1094                         allocated_meta_size / default_chunk_size;
1095
1096         global_total_size = 0;
1097         fs_block_size = sectorsize;
1098         ret = ftw(dir_name, ftw_add_entry_size, 10);
1099         dir_size = global_total_size;
1100         if (ret < 0) {
1101                 error("ftw subdir walk of %s failed: %s", dir_name,
1102                         strerror(errno));
1103                 exit(1);
1104         }
1105
1106         num_of_data_chunks = (dir_size + default_chunk_size - 1) /
1107                 default_chunk_size;
1108
1109         num_of_meta_chunks = (dir_size / 2) / default_chunk_size;
1110         if (((dir_size / 2) % default_chunk_size) != 0)
1111                 num_of_meta_chunks++;
1112         if (num_of_meta_chunks <= num_of_allocated_meta_chunks)
1113                 num_of_meta_chunks = 0;
1114         else
1115                 num_of_meta_chunks -= num_of_allocated_meta_chunks;
1116
1117         total_size = allocated_total_size +
1118                      (num_of_data_chunks * default_chunk_size) +
1119                      (num_of_meta_chunks * default_chunk_size);
1120
1121         *num_of_meta_chunks_ret = num_of_meta_chunks;
1122         *size_of_data_ret = num_of_data_chunks * default_chunk_size;
1123         return total_size;
1124 }
1125
1126 static int zero_output_file(int out_fd, u64 size)
1127 {
1128         int loop_num;
1129         u64 location = 0;
1130         char buf[4096];
1131         int ret = 0, i;
1132         ssize_t written;
1133
1134         memset(buf, 0, 4096);
1135         loop_num = size / 4096;
1136         for (i = 0; i < loop_num; i++) {
1137                 written = pwrite64(out_fd, buf, 4096, location);
1138                 if (written != 4096)
1139                         ret = -EIO;
1140                 location += 4096;
1141         }
1142         return ret;
1143 }
1144
1145 static int is_ssd(const char *file)
1146 {
1147         blkid_probe probe;
1148         char wholedisk[PATH_MAX];
1149         char sysfs_path[PATH_MAX];
1150         dev_t devno;
1151         int fd;
1152         char rotational;
1153         int ret;
1154
1155         probe = blkid_new_probe_from_filename(file);
1156         if (!probe)
1157                 return 0;
1158
1159         /* Device number of this disk (possibly a partition) */
1160         devno = blkid_probe_get_devno(probe);
1161         if (!devno) {
1162                 blkid_free_probe(probe);
1163                 return 0;
1164         }
1165
1166         /* Get whole disk name (not full path) for this devno */
1167         ret = blkid_devno_to_wholedisk(devno,
1168                         wholedisk, sizeof(wholedisk), NULL);
1169         if (ret) {
1170                 blkid_free_probe(probe);
1171                 return 0;
1172         }
1173
1174         snprintf(sysfs_path, PATH_MAX, "/sys/block/%s/queue/rotational",
1175                  wholedisk);
1176
1177         blkid_free_probe(probe);
1178
1179         fd = open(sysfs_path, O_RDONLY);
1180         if (fd < 0) {
1181                 return 0;
1182         }
1183
1184         if (read(fd, &rotational, 1) < 1) {
1185                 close(fd);
1186                 return 0;
1187         }
1188         close(fd);
1189
1190         return rotational == '0';
1191 }
1192
1193 static int _cmp_device_by_id(void *priv, struct list_head *a,
1194                              struct list_head *b)
1195 {
1196         return list_entry(a, struct btrfs_device, dev_list)->devid -
1197                list_entry(b, struct btrfs_device, dev_list)->devid;
1198 }
1199
1200 static void list_all_devices(struct btrfs_root *root)
1201 {
1202         struct btrfs_fs_devices *fs_devices;
1203         struct btrfs_device *device;
1204         int number_of_devices = 0;
1205         u64 total_block_count = 0;
1206
1207         fs_devices = root->fs_info->fs_devices;
1208
1209         list_for_each_entry(device, &fs_devices->devices, dev_list)
1210                 number_of_devices++;
1211
1212         list_sort(NULL, &fs_devices->devices, _cmp_device_by_id);
1213
1214         printf("Number of devices:  %d\n", number_of_devices);
1215         /* printf("Total devices size: %10s\n", */
1216                 /* pretty_size(total_block_count)); */
1217         printf("Devices:\n");
1218         printf("   ID        SIZE  PATH\n");
1219         list_for_each_entry(device, &fs_devices->devices, dev_list) {
1220                 printf("  %3llu  %10s  %s\n",
1221                         device->devid,
1222                         pretty_size(device->total_bytes),
1223                         device->name);
1224                 total_block_count += device->total_bytes;
1225         }
1226
1227         printf("\n");
1228 }
1229
1230 static int is_temp_block_group(struct extent_buffer *node,
1231                                struct btrfs_block_group_item *bgi,
1232                                u64 data_profile, u64 meta_profile,
1233                                u64 sys_profile)
1234 {
1235         u64 flag = btrfs_disk_block_group_flags(node, bgi);
1236         u64 flag_type = flag & BTRFS_BLOCK_GROUP_TYPE_MASK;
1237         u64 flag_profile = flag & BTRFS_BLOCK_GROUP_PROFILE_MASK;
1238         u64 used = btrfs_disk_block_group_used(node, bgi);
1239
1240         /*
1241          * Chunks meets all the following conditions is a temp chunk
1242          * 1) Empty chunk
1243          * Temp chunk is always empty.
1244          *
1245          * 2) profile mismatch with mkfs profile.
1246          * Temp chunk is always in SINGLE
1247          *
1248          * 3) Size differs with mkfs_alloc
1249          * Special case for SINGLE/SINGLE btrfs.
1250          * In that case, temp data chunk and real data chunk are always empty.
1251          * So we need to use mkfs_alloc to be sure which chunk is the newly
1252          * allocated.
1253          *
1254          * Normally, new chunk size is equal to mkfs one (One chunk)
1255          * If it has multiple chunks, we just refuse to delete any one.
1256          * As they are all single, so no real problem will happen.
1257          * So only use condition 1) and 2) to judge them.
1258          */
1259         if (used != 0)
1260                 return 0;
1261         switch (flag_type) {
1262         case BTRFS_BLOCK_GROUP_DATA:
1263         case BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA:
1264                 data_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1265                 if (flag_profile != data_profile)
1266                         return 1;
1267                 break;
1268         case BTRFS_BLOCK_GROUP_METADATA:
1269                 meta_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1270                 if (flag_profile != meta_profile)
1271                         return 1;
1272                 break;
1273         case BTRFS_BLOCK_GROUP_SYSTEM:
1274                 sys_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1275                 if (flag_profile != sys_profile)
1276                         return 1;
1277                 break;
1278         }
1279         return 0;
1280 }
1281
1282 /* Note: if current is a block group, it will skip it anyway */
1283 static int next_block_group(struct btrfs_root *root,
1284                             struct btrfs_path *path)
1285 {
1286         struct btrfs_key key;
1287         int ret = 0;
1288
1289         while (1) {
1290                 ret = btrfs_next_item(root, path);
1291                 if (ret)
1292                         goto out;
1293
1294                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1295                 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
1296                         goto out;
1297         }
1298 out:
1299         return ret;
1300 }
1301
1302 /* This function will cleanup  */
1303 static int cleanup_temp_chunks(struct btrfs_fs_info *fs_info,
1304                                struct mkfs_allocation *alloc,
1305                                u64 data_profile, u64 meta_profile,
1306                                u64 sys_profile)
1307 {
1308         struct btrfs_trans_handle *trans = NULL;
1309         struct btrfs_block_group_item *bgi;
1310         struct btrfs_root *root = fs_info->extent_root;
1311         struct btrfs_key key;
1312         struct btrfs_key found_key;
1313         struct btrfs_path path;
1314         int ret = 0;
1315
1316         btrfs_init_path(&path);
1317         trans = btrfs_start_transaction(root, 1);
1318
1319         key.objectid = 0;
1320         key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
1321         key.offset = 0;
1322
1323         while (1) {
1324                 /*
1325                  * as the rest of the loop may modify the tree, we need to
1326                  * start a new search each time.
1327                  */
1328                 ret = btrfs_search_slot(trans, root, &key, &path, 0, 0);
1329                 if (ret < 0)
1330                         goto out;
1331
1332                 btrfs_item_key_to_cpu(path.nodes[0], &found_key,
1333                                       path.slots[0]);
1334                 if (found_key.objectid < key.objectid)
1335                         goto out;
1336                 if (found_key.type != BTRFS_BLOCK_GROUP_ITEM_KEY) {
1337                         ret = next_block_group(root, &path);
1338                         if (ret < 0)
1339                                 goto out;
1340                         if (ret > 0) {
1341                                 ret = 0;
1342                                 goto out;
1343                         }
1344                         btrfs_item_key_to_cpu(path.nodes[0], &found_key,
1345                                               path.slots[0]);
1346                 }
1347
1348                 bgi = btrfs_item_ptr(path.nodes[0], path.slots[0],
1349                                      struct btrfs_block_group_item);
1350                 if (is_temp_block_group(path.nodes[0], bgi,
1351                                         data_profile, meta_profile,
1352                                         sys_profile)) {
1353                         u64 flags = btrfs_disk_block_group_flags(path.nodes[0],
1354                                                              bgi);
1355
1356                         ret = btrfs_free_block_group(trans, fs_info,
1357                                         found_key.objectid, found_key.offset);
1358                         if (ret < 0)
1359                                 goto out;
1360
1361                         if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
1362                             BTRFS_BLOCK_GROUP_DATA)
1363                                 alloc->data -= found_key.offset;
1364                         else if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
1365                                  BTRFS_BLOCK_GROUP_METADATA)
1366                                 alloc->metadata -= found_key.offset;
1367                         else if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
1368                                  BTRFS_BLOCK_GROUP_SYSTEM)
1369                                 alloc->system -= found_key.offset;
1370                         else if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
1371                                  (BTRFS_BLOCK_GROUP_METADATA |
1372                                   BTRFS_BLOCK_GROUP_DATA))
1373                                 alloc->mixed -= found_key.offset;
1374                 }
1375                 btrfs_release_path(&path);
1376                 key.objectid = found_key.objectid + found_key.offset;
1377         }
1378 out:
1379         if (trans)
1380                 btrfs_commit_transaction(trans, root);
1381         btrfs_release_path(&path);
1382         return ret;
1383 }
1384
1385 int main(int argc, char **argv)
1386 {
1387         char *file;
1388         struct btrfs_root *root;
1389         struct btrfs_fs_info *fs_info;
1390         struct btrfs_trans_handle *trans;
1391         char *label = NULL;
1392         u64 block_count = 0;
1393         u64 dev_block_count = 0;
1394         u64 blocks[7];
1395         u64 alloc_start = 0;
1396         u64 metadata_profile = 0;
1397         u64 data_profile = 0;
1398         u32 nodesize = max_t(u32, sysconf(_SC_PAGESIZE),
1399                         BTRFS_MKFS_DEFAULT_NODE_SIZE);
1400         u32 sectorsize = 4096;
1401         u32 stripesize = 4096;
1402         int zero_end = 1;
1403         int fd;
1404         int ret;
1405         int i;
1406         int mixed = 0;
1407         int nodesize_forced = 0;
1408         int data_profile_opt = 0;
1409         int metadata_profile_opt = 0;
1410         int discard = 1;
1411         int ssd = 0;
1412         int force_overwrite = 0;
1413         char *source_dir = NULL;
1414         int source_dir_set = 0;
1415         u64 num_of_meta_chunks = 0;
1416         u64 size_of_data = 0;
1417         u64 source_dir_size = 0;
1418         int dev_cnt = 0;
1419         int saved_optind;
1420         char fs_uuid[BTRFS_UUID_UNPARSED_SIZE] = { 0 };
1421         u64 features = BTRFS_MKFS_DEFAULT_FEATURES;
1422         struct mkfs_allocation allocation = { 0 };
1423         struct btrfs_mkfs_config mkfs_cfg;
1424
1425         while(1) {
1426                 int c;
1427                 static const struct option long_options[] = {
1428                         { "alloc-start", required_argument, NULL, 'A'},
1429                         { "byte-count", required_argument, NULL, 'b' },
1430                         { "force", no_argument, NULL, 'f' },
1431                         { "leafsize", required_argument, NULL, 'l' },
1432                         { "label", required_argument, NULL, 'L'},
1433                         { "metadata", required_argument, NULL, 'm' },
1434                         { "mixed", no_argument, NULL, 'M' },
1435                         { "nodesize", required_argument, NULL, 'n' },
1436                         { "sectorsize", required_argument, NULL, 's' },
1437                         { "data", required_argument, NULL, 'd' },
1438                         { "version", no_argument, NULL, 'V' },
1439                         { "rootdir", required_argument, NULL, 'r' },
1440                         { "nodiscard", no_argument, NULL, 'K' },
1441                         { "features", required_argument, NULL, 'O' },
1442                         { "uuid", required_argument, NULL, 'U' },
1443                         { "quiet", 0, NULL, 'q' },
1444                         { "help", no_argument, NULL, GETOPT_VAL_HELP },
1445                         { NULL, 0, NULL, 0}
1446                 };
1447
1448                 c = getopt_long(argc, argv, "A:b:fl:n:s:m:d:L:O:r:U:VMKq",
1449                                 long_options, NULL);
1450                 if (c < 0)
1451                         break;
1452                 switch(c) {
1453                         case 'A':
1454                                 alloc_start = parse_size(optarg);
1455                                 break;
1456                         case 'f':
1457                                 force_overwrite = 1;
1458                                 break;
1459                         case 'd':
1460                                 data_profile = parse_profile(optarg);
1461                                 data_profile_opt = 1;
1462                                 break;
1463                         case 'l':
1464                                 warning("--leafsize is deprecated, use --nodesize");
1465                         case 'n':
1466                                 nodesize = parse_size(optarg);
1467                                 nodesize_forced = 1;
1468                                 break;
1469                         case 'L':
1470                                 label = parse_label(optarg);
1471                                 break;
1472                         case 'm':
1473                                 metadata_profile = parse_profile(optarg);
1474                                 metadata_profile_opt = 1;
1475                                 break;
1476                         case 'M':
1477                                 mixed = 1;
1478                                 break;
1479                         case 'O': {
1480                                 char *orig = strdup(optarg);
1481                                 char *tmp = orig;
1482
1483                                 tmp = btrfs_parse_fs_features(tmp, &features);
1484                                 if (tmp) {
1485                                         error("unrecognized filesystem feature '%s'",
1486                                                         tmp);
1487                                         free(orig);
1488                                         exit(1);
1489                                 }
1490                                 free(orig);
1491                                 if (features & BTRFS_FEATURE_LIST_ALL) {
1492                                         btrfs_list_all_fs_features(0);
1493                                         exit(0);
1494                                 }
1495                                 break;
1496                                 }
1497                         case 's':
1498                                 sectorsize = parse_size(optarg);
1499                                 break;
1500                         case 'b':
1501                                 block_count = parse_size(optarg);
1502                                 zero_end = 0;
1503                                 break;
1504                         case 'V':
1505                                 printf("mkfs.btrfs, part of %s\n",
1506                                                 PACKAGE_STRING);
1507                                 exit(0);
1508                                 break;
1509                         case 'r':
1510                                 source_dir = optarg;
1511                                 source_dir_set = 1;
1512                                 break;
1513                         case 'U':
1514                                 strncpy(fs_uuid, optarg,
1515                                         BTRFS_UUID_UNPARSED_SIZE - 1);
1516                                 break;
1517                         case 'K':
1518                                 discard = 0;
1519                                 break;
1520                         case 'q':
1521                                 verbose = 0;
1522                                 break;
1523                         case GETOPT_VAL_HELP:
1524                         default:
1525                                 print_usage(c != GETOPT_VAL_HELP);
1526                 }
1527         }
1528
1529         if (verbose) {
1530                 printf("%s\n", PACKAGE_STRING);
1531                 printf("See %s for more information.\n\n", PACKAGE_URL);
1532         }
1533
1534         sectorsize = max(sectorsize, (u32)sysconf(_SC_PAGESIZE));
1535         stripesize = sectorsize;
1536         saved_optind = optind;
1537         dev_cnt = argc - optind;
1538         if (dev_cnt == 0)
1539                 print_usage(1);
1540
1541         if (source_dir_set && dev_cnt > 1) {
1542                 error("the option -r is limited to a single device");
1543                 exit(1);
1544         }
1545
1546         if (*fs_uuid) {
1547                 uuid_t dummy_uuid;
1548
1549                 if (uuid_parse(fs_uuid, dummy_uuid) != 0) {
1550                         error("could not parse UUID: %s", fs_uuid);
1551                         exit(1);
1552                 }
1553                 if (!test_uuid_unique(fs_uuid)) {
1554                         error("non-unique UUID: %s", fs_uuid);
1555                         exit(1);
1556                 }
1557         }
1558
1559         while (dev_cnt-- > 0) {
1560                 file = argv[optind++];
1561                 if (is_block_device(file) == 1)
1562                         if (test_dev_for_mkfs(file, force_overwrite))
1563                                 exit(1);
1564         }
1565
1566         optind = saved_optind;
1567         dev_cnt = argc - optind;
1568
1569         file = argv[optind++];
1570         ssd = is_ssd(file);
1571
1572         /*
1573         * Set default profiles according to number of added devices.
1574         * For mixed groups defaults are single/single.
1575         */
1576         if (!mixed) {
1577                 if (!metadata_profile_opt) {
1578                         if (dev_cnt == 1 && ssd && verbose)
1579                                 printf("Detected a SSD, turning off metadata "
1580                                 "duplication.  Mkfs with -m dup if you want to "
1581                                 "force metadata duplication.\n");
1582
1583                         metadata_profile = (dev_cnt > 1) ?
1584                                         BTRFS_BLOCK_GROUP_RAID1 : (ssd) ?
1585                                         0: BTRFS_BLOCK_GROUP_DUP;
1586                 }
1587                 if (!data_profile_opt) {
1588                         data_profile = (dev_cnt > 1) ?
1589                                 BTRFS_BLOCK_GROUP_RAID0 : 0; /* raid0 or single */
1590                 }
1591         } else {
1592                 u32 best_nodesize = max_t(u32, sysconf(_SC_PAGESIZE), sectorsize);
1593
1594                 if (metadata_profile_opt || data_profile_opt) {
1595                         if (metadata_profile != data_profile) {
1596                                 error(
1597         "with mixed block groups data and metadata profiles must be the same");
1598                                 exit(1);
1599                         }
1600                 }
1601
1602                 if (!nodesize_forced)
1603                         nodesize = best_nodesize;
1604         }
1605
1606         /*
1607          * FS features that can be set by other means than -O
1608          * just set the bit here
1609          */
1610         if (mixed)
1611                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS;
1612
1613         if ((data_profile | metadata_profile) &
1614             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
1615                 features |= BTRFS_FEATURE_INCOMPAT_RAID56;
1616         }
1617
1618         if (btrfs_check_nodesize(nodesize, sectorsize,
1619                                  features))
1620                 exit(1);
1621
1622         if (sectorsize < sizeof(struct btrfs_super_block)) {
1623                 error("sectorsize smaller than superblock: %u < %zu",
1624                                 sectorsize, sizeof(struct btrfs_super_block));
1625                 exit(1);
1626         }
1627
1628         /* Check device/block_count after the nodesize is determined */
1629         if (block_count && block_count < btrfs_min_dev_size(nodesize)) {
1630                 error("size %llu is too small to make a usable filesystem",
1631                         block_count);
1632                 error("minimum size for btrfs filesystem is %llu",
1633                         btrfs_min_dev_size(nodesize));
1634                 exit(1);
1635         }
1636         for (i = saved_optind; i < saved_optind + dev_cnt; i++) {
1637                 char *path;
1638
1639                 path = argv[i];
1640                 ret = test_minimum_size(path, nodesize);
1641                 if (ret < 0) {
1642                         error("failed to check size for %s: %s",
1643                                 path, strerror(-ret));
1644                         exit (1);
1645                 }
1646                 if (ret > 0) {
1647                         error("'%s' is too small to make a usable filesystem",
1648                                 path);
1649                         error("minimum size for each btrfs device is %llu",
1650                                 btrfs_min_dev_size(nodesize));
1651                         exit(1);
1652                 }
1653         }
1654         ret = test_num_disk_vs_raid(metadata_profile, data_profile,
1655                         dev_cnt, mixed, ssd);
1656         if (ret)
1657                 exit(1);
1658
1659         dev_cnt--;
1660
1661         if (!source_dir_set) {
1662                 /*
1663                  * open without O_EXCL so that the problem should not
1664                  * occur by the following processing.
1665                  * (btrfs_register_one_device() fails if O_EXCL is on)
1666                  */
1667                 fd = open(file, O_RDWR);
1668                 if (fd < 0) {
1669                         error("unable to open %s: %s", file, strerror(errno));
1670                         exit(1);
1671                 }
1672                 ret = btrfs_prepare_device(fd, file, &dev_block_count,
1673                                 block_count,
1674                                 (zero_end ? PREP_DEVICE_ZERO_END : 0) |
1675                                 (discard ? PREP_DEVICE_DISCARD : 0) |
1676                                 (verbose ? PREP_DEVICE_VERBOSE : 0));
1677                 if (ret) {
1678                         close(fd);
1679                         exit(1);
1680                 }
1681                 if (block_count && block_count > dev_block_count) {
1682                         error("%s is smaller than requested size, expected %llu, found %llu",
1683                                         file,
1684                                         (unsigned long long)block_count,
1685                                         (unsigned long long)dev_block_count);
1686                         exit(1);
1687                 }
1688         } else {
1689                 fd = open(file, O_CREAT | O_RDWR,
1690                                 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH);
1691                 if (fd < 0) {
1692                         error("unable to open %s: %s", file, strerror(errno));
1693                         exit(1);
1694                 }
1695
1696                 source_dir_size = size_sourcedir(source_dir, sectorsize,
1697                                              &num_of_meta_chunks, &size_of_data);
1698                 if(block_count < source_dir_size)
1699                         block_count = source_dir_size;
1700                 ret = zero_output_file(fd, block_count);
1701                 if (ret) {
1702                         error("unable to zero the output file");
1703                         exit(1);
1704                 }
1705                 /* our "device" is the new image file */
1706                 dev_block_count = block_count;
1707         }
1708
1709         /* To create the first block group and chunk 0 in make_btrfs */
1710         if (dev_block_count < BTRFS_MKFS_SYSTEM_GROUP_SIZE) {
1711                 error("device is too small to make filesystem, must be at least %llu",
1712                                 (unsigned long long)BTRFS_MKFS_SYSTEM_GROUP_SIZE);
1713                 exit(1);
1714         }
1715
1716         blocks[0] = BTRFS_SUPER_INFO_OFFSET;
1717         for (i = 1; i < 7; i++) {
1718                 blocks[i] = BTRFS_SUPER_INFO_OFFSET + 1024 * 1024 +
1719                         nodesize * i;
1720         }
1721
1722         if (group_profile_max_safe_loss(metadata_profile) <
1723                 group_profile_max_safe_loss(data_profile)){
1724                 warning("metadata has lower redundancy than data!\n");
1725         }
1726
1727         mkfs_cfg.label = label;
1728         memcpy(mkfs_cfg.fs_uuid, fs_uuid, sizeof(mkfs_cfg.fs_uuid));
1729         memcpy(mkfs_cfg.blocks, blocks, sizeof(blocks));
1730         mkfs_cfg.num_bytes = dev_block_count;
1731         mkfs_cfg.nodesize = nodesize;
1732         mkfs_cfg.sectorsize = sectorsize;
1733         mkfs_cfg.stripesize = stripesize;
1734         mkfs_cfg.features = features;
1735
1736         ret = make_btrfs(fd, &mkfs_cfg, NULL);
1737         if (ret) {
1738                 error("error during mkfs: %s", strerror(-ret));
1739                 exit(1);
1740         }
1741
1742         fs_info = open_ctree_fs_info(file, 0, 0, 0,
1743                         OPEN_CTREE_WRITES | OPEN_CTREE_FS_PARTIAL);
1744         if (!fs_info) {
1745                 error("open ctree failed");
1746                 close(fd);
1747                 exit(1);
1748         }
1749         root = fs_info->fs_root;
1750         fs_info->alloc_start = alloc_start;
1751
1752         ret = create_metadata_block_groups(root, mixed, &allocation);
1753         if (ret) {
1754                 error("failed to create default block groups: %d", ret);
1755                 exit(1);
1756         }
1757
1758         trans = btrfs_start_transaction(root, 1);
1759         if (!trans) {
1760                 error("failed to start transaction");
1761                 exit(1);
1762         }
1763
1764         ret = create_data_block_groups(trans, root, mixed, &allocation);
1765         if (ret) {
1766                 error("failed to create default data block groups: %d", ret);
1767                 exit(1);
1768         }
1769
1770         ret = make_root_dir(trans, root, &allocation);
1771         if (ret) {
1772                 error("failed to setup the root directory: %d", ret);
1773                 exit(1);
1774         }
1775
1776         ret = btrfs_commit_transaction(trans, root);
1777         if (ret) {
1778                 error("unable to commit transaction: %d", ret);
1779                 goto out;
1780         }
1781
1782         trans = btrfs_start_transaction(root, 1);
1783         if (!trans) {
1784                 error("failed to start transaction");
1785                 exit(1);
1786         }
1787
1788         if (dev_cnt == 0)
1789                 goto raid_groups;
1790
1791         while (dev_cnt-- > 0) {
1792                 file = argv[optind++];
1793
1794                 /*
1795                  * open without O_EXCL so that the problem should not
1796                  * occur by the following processing.
1797                  * (btrfs_register_one_device() fails if O_EXCL is on)
1798                  */
1799                 fd = open(file, O_RDWR);
1800                 if (fd < 0) {
1801                         error("unable to open %s: %s", file, strerror(errno));
1802                         exit(1);
1803                 }
1804                 ret = btrfs_device_already_in_root(root, fd,
1805                                                    BTRFS_SUPER_INFO_OFFSET);
1806                 if (ret) {
1807                         error("skipping duplicate device %s in the filesystem",
1808                                 file);
1809                         close(fd);
1810                         continue;
1811                 }
1812                 ret = btrfs_prepare_device(fd, file, &dev_block_count,
1813                                 block_count,
1814                                 (verbose ? PREP_DEVICE_VERBOSE : 0) |
1815                                 (zero_end ? PREP_DEVICE_ZERO_END : 0) |
1816                                 (discard ? PREP_DEVICE_DISCARD : 0));
1817                 if (ret) {
1818                         close(fd);
1819                         exit(1);
1820                 }
1821
1822                 ret = btrfs_add_to_fsid(trans, root, fd, file, dev_block_count,
1823                                         sectorsize, sectorsize, sectorsize);
1824                 if (ret) {
1825                         error("unable to add %s to filesystem: %d", file, ret);
1826                         goto out;
1827                 }
1828                 if (verbose >= 2) {
1829                         struct btrfs_device *device;
1830
1831                         device = container_of(fs_info->fs_devices->devices.next,
1832                                         struct btrfs_device, dev_list);
1833                         printf("adding device %s id %llu\n", file,
1834                                 (unsigned long long)device->devid);
1835                 }
1836         }
1837
1838 raid_groups:
1839         if (!source_dir_set) {
1840                 ret = create_raid_groups(trans, root, data_profile,
1841                                  metadata_profile, mixed, &allocation);
1842                 if (ret) {
1843                         error("unable to create raid groups: %d", ret);
1844                         goto out;
1845                 }
1846         }
1847
1848         ret = create_data_reloc_tree(trans, root);
1849         if (ret) {
1850                 error("unable to create data reloc tree: %d", ret);
1851                 goto out;
1852         }
1853
1854         ret = btrfs_commit_transaction(trans, root);
1855         if (ret) {
1856                 error("unable to commit transaction: %d", ret);
1857                 goto out;
1858         }
1859
1860         if (source_dir_set) {
1861                 trans = btrfs_start_transaction(root, 1);
1862                 ret = create_chunks(trans, root,
1863                                     num_of_meta_chunks, size_of_data,
1864                                     &allocation);
1865                 if (ret) {
1866                         error("unable to create chunks: %d", ret);
1867                         goto out;
1868                 }
1869                 ret = btrfs_commit_transaction(trans, root);
1870                 if (ret) {
1871                         error("transaction commit failed: %d", ret);
1872                         goto out;
1873                 }
1874
1875                 ret = make_image(source_dir, root, fd);
1876                 if (ret) {
1877                         error("error wihle filling filesystem: %d", ret);
1878                         goto out;
1879                 }
1880         }
1881         ret = cleanup_temp_chunks(fs_info, &allocation, data_profile,
1882                                   metadata_profile, metadata_profile);
1883         if (ret < 0) {
1884                 error("failed to cleanup temporary chunks: %d", ret);
1885                 goto out;
1886         }
1887
1888         if (verbose) {
1889                 char features_buf[64];
1890
1891                 printf("Label:              %s\n", label);
1892                 printf("UUID:               %s\n", fs_uuid);
1893                 printf("Node size:          %u\n", nodesize);
1894                 printf("Sector size:        %u\n", sectorsize);
1895                 printf("Filesystem size:    %s\n",
1896                         pretty_size(btrfs_super_total_bytes(fs_info->super_copy)));
1897                 printf("Block group profiles:\n");
1898                 if (allocation.data)
1899                         printf("  Data:             %-8s %16s\n",
1900                                 btrfs_group_profile_str(data_profile),
1901                                 pretty_size(allocation.data));
1902                 if (allocation.metadata)
1903                         printf("  Metadata:         %-8s %16s\n",
1904                                 btrfs_group_profile_str(metadata_profile),
1905                                 pretty_size(allocation.metadata));
1906                 if (allocation.mixed)
1907                         printf("  Data+Metadata:    %-8s %16s\n",
1908                                 btrfs_group_profile_str(data_profile),
1909                                 pretty_size(allocation.mixed));
1910                 printf("  System:           %-8s %16s\n",
1911                         btrfs_group_profile_str(metadata_profile),
1912                         pretty_size(allocation.system));
1913                 printf("SSD detected:       %s\n", ssd ? "yes" : "no");
1914                 btrfs_parse_features_to_string(features_buf, features);
1915                 printf("Incompat features:  %s", features_buf);
1916                 printf("\n");
1917
1918                 list_all_devices(root);
1919         }
1920
1921         /*
1922          * The filesystem is now fully set up, commit the remaining changes and
1923          * fix the signature as the last step before closing the devices.
1924          */
1925         fs_info->finalize_on_close = 1;
1926 out:
1927         ret = close_ctree(root);
1928
1929         if (!ret) {
1930                 optind = saved_optind;
1931                 dev_cnt = argc - optind;
1932                 while (dev_cnt-- > 0) {
1933                         file = argv[optind++];
1934                         if (is_block_device(file) == 1)
1935                                 btrfs_register_one_device(file);
1936                 }
1937         }
1938
1939         btrfs_close_all_devices();
1940         free(label);
1941
1942         return !!ret;
1943 }