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