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         char *dir_name;
  52         char *path;
  53         ino_t inum;
  54         struct list_head list;
  55 };
  56
  57 struct mkfs_allocation {
  58         u64 data;
  59         u64 metadata;
  60         u64 mixed;
  61         u64 system;
  62 };
  63
  64 static int create_metadata_block_groups(struct btrfs_root *root, int mixed,
  65                                 struct mkfs_allocation *allocation)
  66 {
  67         struct btrfs_trans_handle *trans;
  68         u64 bytes_used;
  69         u64 chunk_start = 0;
  70         u64 chunk_size = 0;
  71         int ret;
  72
  73         trans = btrfs_start_transaction(root, 1);
  74         bytes_used = btrfs_super_bytes_used(root->fs_info->super_copy);
  75
  76         root->fs_info->system_allocs = 1;
  77         ret = btrfs_make_block_group(trans, root, bytes_used,
  78                                      BTRFS_BLOCK_GROUP_SYSTEM,
  79                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
  80                                      0, BTRFS_MKFS_SYSTEM_GROUP_SIZE);
  81         allocation->system += BTRFS_MKFS_SYSTEM_GROUP_SIZE;
  82         if (ret)
  83                 return ret;
  84
  85         if (mixed) {
  86                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
  87                                         &chunk_start, &chunk_size,
  88                                         BTRFS_BLOCK_GROUP_METADATA |
  89                                         BTRFS_BLOCK_GROUP_DATA);
  90                 if (ret == -ENOSPC) {
  91                         error("no space to allocate data/metadata chunk");
  92                         goto err;
  93                 }
  94                 if (ret)
  95                         return ret;
  96                 ret = btrfs_make_block_group(trans, root, 0,
  97                                              BTRFS_BLOCK_GROUP_METADATA |
  98                                              BTRFS_BLOCK_GROUP_DATA,
  99                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 100                                              chunk_start, chunk_size);
 101                 if (ret)
 102                         return ret;
 103                 allocation->mixed += chunk_size;
 104         } else {
 105                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 106                                         &chunk_start, &chunk_size,
 107                                         BTRFS_BLOCK_GROUP_METADATA);
 108                 if (ret == -ENOSPC) {
 109                         error("no space to allocate metadata chunk");
 110                         goto err;
 111                 }
 112                 if (ret)
 113                         return ret;
 114                 ret = btrfs_make_block_group(trans, root, 0,
 115                                              BTRFS_BLOCK_GROUP_METADATA,
 116                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 117                                              chunk_start, chunk_size);
 118                 allocation->metadata += chunk_size;
 119                 if (ret)
 120                         return ret;
 121         }
 122
 123         root->fs_info->system_allocs = 0;
 124         ret = btrfs_commit_transaction(trans, root);
 125
 126 err:
 127         return ret;
 128 }
 129
 130 static int create_data_block_groups(struct btrfs_trans_handle *trans,
 131                 struct btrfs_root *root, int mixed,
 132                 struct mkfs_allocation *allocation)
 133 {
 134         u64 chunk_start = 0;
 135         u64 chunk_size = 0;
 136         int ret = 0;
 137
 138         if (!mixed) {
 139                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 140                                         &chunk_start, &chunk_size,
 141                                         BTRFS_BLOCK_GROUP_DATA);
 142                 if (ret == -ENOSPC) {
 143                         error("no space to allocate data chunk");
 144                         goto err;
 145                 }
 146                 if (ret)
 147                         return ret;
 148                 ret = btrfs_make_block_group(trans, root, 0,
 149                                              BTRFS_BLOCK_GROUP_DATA,
 150                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 151                                              chunk_start, chunk_size);
 152                 allocation->data += chunk_size;
 153                 if (ret)
 154                         return ret;
 155         }
 156
 157 err:
 158         return ret;
 159 }
 160
 161 static int make_root_dir(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 162                 struct mkfs_allocation *allocation)
 163 {
 164         struct btrfs_key location;
 165         int ret;
 166
 167         ret = btrfs_make_root_dir(trans, root->fs_info->tree_root,
 168                               BTRFS_ROOT_TREE_DIR_OBJECTID);
 169         if (ret)
 170                 goto err;
 171         ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
 172         if (ret)
 173                 goto err;
 174         memcpy(&location, &root->fs_info->fs_root->root_key, sizeof(location));
 175         location.offset = (u64)-1;
 176         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
 177                         "default", 7,
 178                         btrfs_super_root_dir(root->fs_info->super_copy),
 179                         &location, BTRFS_FT_DIR, 0);
 180         if (ret)
 181                 goto err;
 182
 183         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 184                              "default", 7, location.objectid,
 185                              BTRFS_ROOT_TREE_DIR_OBJECTID, 0);
 186         if (ret)
 187                 goto err;
 188
 189 err:
 190         return ret;
 191 }
 192
 193 static int __recow_root(struct btrfs_trans_handle *trans,
 194                          struct btrfs_root *root)
 195 {
 196         struct extent_buffer *tmp;
 197         int ret;
 198
 199         if (trans->transid != btrfs_root_generation(&root->root_item)) {
 200                 extent_buffer_get(root->node);
 201                 ret = __btrfs_cow_block(trans, root, root->node,
 202                                         NULL, 0, &tmp, 0, 0);
 203                 if (ret)
 204                         return ret;
 205                 free_extent_buffer(tmp);
 206         }
 207
 208         return 0;
 209 }
 210
 211 static int recow_roots(struct btrfs_trans_handle *trans,
 212                        struct btrfs_root *root)
 213 {
 214         struct btrfs_fs_info *info = root->fs_info;
 215         int ret;
 216
 217         ret = __recow_root(trans, info->fs_root);
 218         if (ret)
 219                 return ret;
 220         ret = __recow_root(trans, info->tree_root);
 221         if (ret)
 222                 return ret;
 223         ret = __recow_root(trans, info->extent_root);
 224         if (ret)
 225                 return ret;
 226         ret = __recow_root(trans, info->chunk_root);
 227         if (ret)
 228                 return ret;
 229         ret = __recow_root(trans, info->dev_root);
 230         if (ret)
 231                 return ret;
 232         ret = __recow_root(trans, info->csum_root);
 233         if (ret)
 234                 return ret;
 235
 236         return 0;
 237 }
 238
 239 static int create_one_raid_group(struct btrfs_trans_handle *trans,
 240                               struct btrfs_root *root, u64 type,
 241                               struct mkfs_allocation *allocation)
 242
 243 {
 244         u64 chunk_start;
 245         u64 chunk_size;
 246         int ret;
 247
 248         ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 249                                 &chunk_start, &chunk_size, type);
 250         if (ret == -ENOSPC) {
 251                 error("not enough free space to allocate chunk");
 252                 exit(1);
 253         }
 254         if (ret)
 255                 return ret;
 256
 257         ret = btrfs_make_block_group(trans, root->fs_info->extent_root, 0,
 258                                      type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 259                                      chunk_start, chunk_size);
 260
 261         type &= BTRFS_BLOCK_GROUP_TYPE_MASK;
 262         if (type == BTRFS_BLOCK_GROUP_DATA) {
 263                 allocation->data += chunk_size;
 264         } else if (type == BTRFS_BLOCK_GROUP_METADATA) {
 265                 allocation->metadata += chunk_size;
 266         } else if (type == BTRFS_BLOCK_GROUP_SYSTEM) {
 267                 allocation->system += chunk_size;
 268         } else if (type ==
 269                         (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA)) {
 270                 allocation->mixed += chunk_size;
 271         } else {
 272                 error("unrecognized profile type: 0x%llx",
 273                                 (unsigned long long)type);
 274                 ret = -EINVAL;
 275         }
 276
 277         return ret;
 278 }
 279
 280 static int create_raid_groups(struct btrfs_trans_handle *trans,
 281                               struct btrfs_root *root, u64 data_profile,
 282                               u64 metadata_profile, int mixed,
 283                               struct mkfs_allocation *allocation)
 284 {
 285         int ret;
 286
 287         if (metadata_profile) {
 288                 u64 meta_flags = BTRFS_BLOCK_GROUP_METADATA;
 289
 290                 ret = create_one_raid_group(trans, root,
 291                                             BTRFS_BLOCK_GROUP_SYSTEM |
 292                                             metadata_profile, allocation);
 293                 if (ret)
 294                         return ret;
 295
 296                 if (mixed)
 297                         meta_flags |= BTRFS_BLOCK_GROUP_DATA;
 298
 299                 ret = create_one_raid_group(trans, root, meta_flags |
 300                                             metadata_profile, allocation);
 301                 if (ret)
 302                         return ret;
 303
 304         }
 305         if (!mixed && data_profile) {
 306                 ret = create_one_raid_group(trans, root,
 307                                             BTRFS_BLOCK_GROUP_DATA |
 308                                             data_profile, allocation);
 309                 if (ret)
 310                         return ret;
 311         }
 312         ret = recow_roots(trans, root);
 313
 314         return ret;
 315 }
 316
 317 static int create_data_reloc_tree(struct btrfs_trans_handle *trans,
 318                                   struct btrfs_root *root)
 319 {
 320         struct btrfs_key location;
 321         struct btrfs_root_item root_item;
 322         struct extent_buffer *tmp;
 323         u64 objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
 324         int ret;
 325
 326         ret = btrfs_copy_root(trans, root, root->node, &tmp, objectid);
 327         if (ret)
 328                 return ret;
 329
 330         memcpy(&root_item, &root->root_item, sizeof(root_item));
 331         btrfs_set_root_bytenr(&root_item, tmp->start);
 332         btrfs_set_root_level(&root_item, btrfs_header_level(tmp));
 333         btrfs_set_root_generation(&root_item, trans->transid);
 334         free_extent_buffer(tmp);
 335
 336         location.objectid = objectid;
 337         location.type = BTRFS_ROOT_ITEM_KEY;
 338         location.offset = 0;
 339         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
 340                                 &location, &root_item);
 341
 342         return ret;
 343 }
 344
 345 static void print_usage(int ret)
 346 {
 347         printf("usage: mkfs.btrfs [options] dev [ dev ... ]\n");
 348         printf("options:\n");
 349         printf("\t-A|--alloc-start START  the offset to start the FS\n");
 350         printf("\t-b|--byte-count SIZE    total number of bytes in the FS\n");
 351         printf("\t-d|--data PROFILE       data profile, raid0, raid1, raid5, raid6, raid10, dup or single\n");
 352         printf("\t-f|--force              force overwrite of existing filesystem\n");
 353         printf("\t-l|--leafsize SIZE      deprecated, alias for nodesize\n");
 354         printf("\t-L|--label LABEL        set a label\n");
 355         printf("\t-m|--metadata PROFILE   metadata profile, values like data profile\n");
 356         printf("\t-M|--mixed              mix metadata and data together\n");
 357         printf("\t-n|--nodesize SIZE      size of btree nodes\n");
 358         printf("\t-s|--sectorsize SIZE    min block allocation (may not mountable by current kernel)\n");
 359         printf("\t-r|--rootdir DIR        the source directory\n");
 360         printf("\t-K|--nodiscard          do not perform whole device TRIM\n");
 361         printf("\t-O|--features LIST      comma separated list of filesystem features, use '-O list-all' to list features\n");
 362         printf("\t-U|--uuid UUID          specify the filesystem UUID\n");
 363         printf("\t-q|--quiet              no messages except errors\n");
 364         printf("\t-V|--version            print the mkfs.btrfs version and exit\n");
 365         exit(ret);
 366 }
 367
 368 static u64 parse_profile(char *s)
 369 {
 370         if (strcasecmp(s, "raid0") == 0) {
 371                 return BTRFS_BLOCK_GROUP_RAID0;
 372         } else if (strcasecmp(s, "raid1") == 0) {
 373                 return BTRFS_BLOCK_GROUP_RAID1;
 374         } else if (strcasecmp(s, "raid5") == 0) {
 375                 return BTRFS_BLOCK_GROUP_RAID5;
 376         } else if (strcasecmp(s, "raid6") == 0) {
 377                 return BTRFS_BLOCK_GROUP_RAID6;
 378         } else if (strcasecmp(s, "raid10") == 0) {
 379                 return BTRFS_BLOCK_GROUP_RAID10;
 380         } else if (strcasecmp(s, "dup") == 0) {
 381                 return BTRFS_BLOCK_GROUP_DUP;
 382         } else if (strcasecmp(s, "single") == 0) {
 383                 return 0;
 384         } else {
 385                 error("unknown profile %s", s);
 386                 exit(1);
 387         }
 388         /* not reached */
 389         return 0;
 390 }
 391
 392 static char *parse_label(char *input)
 393 {
 394         int len = strlen(input);
 395
 396         if (len >= BTRFS_LABEL_SIZE) {
 397                 error("label %s is too long (max %d)", input,
 398                         BTRFS_LABEL_SIZE - 1);
 399                 exit(1);
 400         }
 401         return strdup(input);
 402 }
 403
 404 static int add_directory_items(struct btrfs_trans_handle *trans,
 405                                struct btrfs_root *root, u64 objectid,
 406                                ino_t parent_inum, const char *name,
 407                                struct stat *st, int *dir_index_cnt)
 408 {
 409         int ret;
 410         int name_len;
 411         struct btrfs_key location;
 412         u8 filetype = 0;
 413
 414         name_len = strlen(name);
 415
 416         location.objectid = objectid;
 417         location.offset = 0;
 418         btrfs_set_key_type(&location, BTRFS_INODE_ITEM_KEY);
 419
 420         if (S_ISDIR(st->st_mode))
 421                 filetype = BTRFS_FT_DIR;
 422         if (S_ISREG(st->st_mode))
 423                 filetype = BTRFS_FT_REG_FILE;
 424         if (S_ISLNK(st->st_mode))
 425                 filetype = BTRFS_FT_SYMLINK;
 426
 427         ret = btrfs_insert_dir_item(trans, root, name, name_len,
 428                                     parent_inum, &location,
 429                                     filetype, index_cnt);
 430         if (ret)
 431                 return ret;
 432         ret = btrfs_insert_inode_ref(trans, root, name, name_len,
 433                                      objectid, parent_inum, index_cnt);
 434         *dir_index_cnt = index_cnt;
 435         index_cnt++;
 436
 437         return ret;
 438 }
 439
 440 static int fill_inode_item(struct btrfs_trans_handle *trans,
 441                            struct btrfs_root *root,
 442                            struct btrfs_inode_item *dst, struct stat *src)
 443 {
 444         u64 blocks = 0;
 445         u64 sectorsize = root->sectorsize;
 446
 447         /*
 448          * btrfs_inode_item has some reserved fields
 449          * and represents on-disk inode entry, so
 450          * zero everything to prevent information leak
 451          */
 452         memset(dst, 0, sizeof (*dst));
 453
 454         btrfs_set_stack_inode_generation(dst, trans->transid);
 455         btrfs_set_stack_inode_size(dst, src->st_size);
 456         btrfs_set_stack_inode_nbytes(dst, 0);
 457         btrfs_set_stack_inode_block_group(dst, 0);
 458         btrfs_set_stack_inode_nlink(dst, src->st_nlink);
 459         btrfs_set_stack_inode_uid(dst, src->st_uid);
 460         btrfs_set_stack_inode_gid(dst, src->st_gid);
 461         btrfs_set_stack_inode_mode(dst, src->st_mode);
 462         btrfs_set_stack_inode_rdev(dst, 0);
 463         btrfs_set_stack_inode_flags(dst, 0);
 464         btrfs_set_stack_timespec_sec(&dst->atime, src->st_atime);
 465         btrfs_set_stack_timespec_nsec(&dst->atime, 0);
 466         btrfs_set_stack_timespec_sec(&dst->ctime, src->st_ctime);
 467         btrfs_set_stack_timespec_nsec(&dst->ctime, 0);
 468         btrfs_set_stack_timespec_sec(&dst->mtime, src->st_mtime);
 469         btrfs_set_stack_timespec_nsec(&dst->mtime, 0);
 470         btrfs_set_stack_timespec_sec(&dst->otime, 0);
 471         btrfs_set_stack_timespec_nsec(&dst->otime, 0);
 472
 473         if (S_ISDIR(src->st_mode)) {
 474                 btrfs_set_stack_inode_size(dst, 0);
 475                 btrfs_set_stack_inode_nlink(dst, 1);
 476         }
 477         if (S_ISREG(src->st_mode)) {
 478                 btrfs_set_stack_inode_size(dst, (u64)src->st_size);
 479                 if (src->st_size <= BTRFS_MAX_INLINE_DATA_SIZE(root))
 480                         btrfs_set_stack_inode_nbytes(dst, src->st_size);
 481                 else {
 482                         blocks = src->st_size / sectorsize;
 483                         if (src->st_size % sectorsize)
 484                                 blocks += 1;
 485                         blocks *= sectorsize;
 486                         btrfs_set_stack_inode_nbytes(dst, blocks);
 487                 }
 488         }
 489         if (S_ISLNK(src->st_mode))
 490                 btrfs_set_stack_inode_nbytes(dst, src->st_size + 1);
 491
 492         return 0;
 493 }
 494
 495 static int directory_select(const struct direct *entry)
 496 {
 497         if ((strncmp(entry->d_name, ".", entry->d_reclen) == 0) ||
 498                 (strncmp(entry->d_name, "..", entry->d_reclen) == 0))
 499                 return 0;
 500         else
 501                 return 1;
 502 }
 503
 504 static void free_namelist(struct direct **files, int count)
 505 {
 506         int i;
 507
 508         if (count < 0)
 509                 return;
 510
 511         for (i = 0; i < count; ++i)
 512                 free(files[i]);
 513         free(files);
 514 }
 515
 516 static u64 calculate_dir_inode_size(char *dirname)
 517 {
 518         int count, i;
 519         struct direct **files, *cur_file;
 520         u64 dir_inode_size = 0;
 521
 522         count = scandir(dirname, &files, directory_select, NULL);
 523
 524         for (i = 0; i < count; i++) {
 525                 cur_file = files[i];
 526                 dir_inode_size += strlen(cur_file->d_name);
 527         }
 528
 529         free_namelist(files, count);
 530
 531         dir_inode_size *= 2;
 532         return dir_inode_size;
 533 }
 534
 535 static int add_inode_items(struct btrfs_trans_handle *trans,
 536                            struct btrfs_root *root,
 537                            struct stat *st, char *name,
 538                            u64 self_objectid, ino_t parent_inum,
 539                            int dir_index_cnt, struct btrfs_inode_item *inode_ret)
 540 {
 541         int ret;
 542         struct btrfs_key inode_key;
 543         struct btrfs_inode_item btrfs_inode;
 544         u64 objectid;
 545         u64 inode_size = 0;
 546
 547         fill_inode_item(trans, root, &btrfs_inode, st);
 548         objectid = self_objectid;
 549
 550         if (S_ISDIR(st->st_mode)) {
 551                 inode_size = calculate_dir_inode_size(name);
 552                 btrfs_set_stack_inode_size(&btrfs_inode, inode_size);
 553         }
 554
 555         inode_key.objectid = objectid;
 556         inode_key.offset = 0;
 557         btrfs_set_key_type(&inode_key, BTRFS_INODE_ITEM_KEY);
 558
 559         ret = btrfs_insert_inode(trans, root, objectid, &btrfs_inode);
 560
 561         *inode_ret = btrfs_inode;
 562         return ret;
 563 }
 564
 565 static int add_xattr_item(struct btrfs_trans_handle *trans,
 566                           struct btrfs_root *root, u64 objectid,
 567                           const char *file_name)
 568 {
 569         int ret;
 570         int cur_name_len;
 571         char xattr_list[XATTR_LIST_MAX];
 572         char *cur_name;
 573         char cur_value[XATTR_SIZE_MAX];
 574         char delimiter = '\0';
 575         char *next_location = xattr_list;
 576
 577         ret = llistxattr(file_name, xattr_list, XATTR_LIST_MAX);
 578         if (ret < 0) {
 579                 if(errno == ENOTSUP)
 580                         return 0;
 581                 error("getting a list of xattr failed for %s: %s", file_name,
 582                                 strerror(errno));
 583                 return ret;
 584         }
 585         if (ret == 0)
 586                 return ret;
 587
 588         cur_name = strtok(xattr_list, &delimiter);
 589         while (cur_name != NULL) {
 590                 cur_name_len = strlen(cur_name);
 591                 next_location += cur_name_len + 1;
 592
 593                 ret = getxattr(file_name, cur_name, cur_value, XATTR_SIZE_MAX);
 594                 if (ret < 0) {
 595                         if(errno == ENOTSUP)
 596                                 return 0;
 597                         error("gettig a xattr value failed for %s attr %s: %s",
 598                                 file_name, cur_name, strerror(errno));
 599                         return ret;
 600                 }
 601
 602                 ret = btrfs_insert_xattr_item(trans, root, cur_name,
 603                                               cur_name_len, cur_value,
 604                                               ret, objectid);
 605                 if (ret) {
 606                         error("inserting a xattr item failed for %s: %s",
 607                                         file_name, strerror(-ret));
 608                 }
 609
 610                 cur_name = strtok(next_location, &delimiter);
 611         }
 612
 613         return ret;
 614 }
 615
 616 static int add_symbolic_link(struct btrfs_trans_handle *trans,
 617                              struct btrfs_root *root,
 618                              u64 objectid, const char *path_name)
 619 {
 620         int ret;
 621         char buf[PATH_MAX];
 622
 623         ret = readlink(path_name, buf, sizeof(buf));
 624         if (ret <= 0) {
 625                 error("readlink failed for %s: %s", path_name, strerror(errno));
 626                 goto fail;
 627         }
 628         if (ret >= sizeof(buf)) {
 629                 error("symlink too long for %s", path_name);
 630                 ret = -1;
 631                 goto fail;
 632         }
 633
 634         buf[ret] = '\0'; /* readlink does not do it for us */
 635         ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
 636                                          buf, ret + 1);
 637 fail:
 638         return ret;
 639 }
 640
 641 static int add_file_items(struct btrfs_trans_handle *trans,
 642                           struct btrfs_root *root,
 643                           struct btrfs_inode_item *btrfs_inode, u64 objectid,
 644                           ino_t parent_inum, struct stat *st,
 645                           const char *path_name, int out_fd)
 646 {
 647         int ret = -1;
 648         ssize_t ret_read;
 649         u64 bytes_read = 0;
 650         struct btrfs_key key;
 651         int blocks;
 652         u32 sectorsize = root->sectorsize;
 653         u64 first_block = 0;
 654         u64 file_pos = 0;
 655         u64 cur_bytes;
 656         u64 total_bytes;
 657         struct extent_buffer *eb = NULL;
 658         int fd;
 659
 660         if (st->st_size == 0)
 661                 return 0;
 662
 663         fd = open(path_name, O_RDONLY);
 664         if (fd == -1) {
 665                 error("cannot open %s: %s", path_name, strerror(errno));
 666                 return ret;
 667         }
 668
 669         blocks = st->st_size / sectorsize;
 670         if (st->st_size % sectorsize)
 671                 blocks += 1;
 672
 673         if (st->st_size <= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 674                 char *buffer = malloc(st->st_size);
 675
 676                 if (!buffer) {
 677                         ret = -ENOMEM;
 678                         goto end;
 679                 }
 680
 681                 ret_read = pread64(fd, buffer, st->st_size, bytes_read);
 682                 if (ret_read == -1) {
 683                         error("cannot read %s at offset %llu length %llu: %s",
 684                                 path_name, (unsigned long long)bytes_read,
 685                                 (unsigned long long)st->st_size,
 686                                 strerror(errno));
 687                         free(buffer);
 688                         goto end;
 689                 }
 690
 691                 ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
 692                                                  buffer, st->st_size);
 693                 free(buffer);
 694                 goto end;
 695         }
 696
 697         /* round up our st_size to the FS blocksize */
 698         total_bytes = (u64)blocks * sectorsize;
 699
 700         /*
 701          * do our IO in extent buffers so it can work
 702          * against any raid type
 703          */
 704         eb = calloc(1, sizeof(*eb) + sectorsize);
 705         if (!eb) {
 706                 ret = -ENOMEM;
 707                 goto end;
 708         }
 709
 710 again:
 711
 712         /*
 713          * keep our extent size at 1MB max, this makes it easier to work inside
 714          * the tiny block groups created during mkfs
 715          */
 716         cur_bytes = min(total_bytes, 1024ULL * 1024);
 717         ret = btrfs_reserve_extent(trans, root, cur_bytes, 0, 0, (u64)-1,
 718                                    &key, 1);
 719         if (ret)
 720                 goto end;
 721
 722         first_block = key.objectid;
 723         bytes_read = 0;
 724
 725         while (bytes_read < cur_bytes) {
 726
 727                 memset(eb->data, 0, sectorsize);
 728
 729                 ret_read = pread64(fd, eb->data, sectorsize, file_pos + bytes_read);
 730                 if (ret_read == -1) {
 731                         error("cannot read %s at offset %llu length %llu: %s",
 732                                 path_name,
 733                                 (unsigned long long)file_pos + bytes_read,
 734                                 (unsigned long long)sectorsize,
 735                                 strerror(errno));
 736                         goto end;
 737                 }
 738
 739                 eb->start = first_block + bytes_read;
 740                 eb->len = sectorsize;
 741
 742                 /*
 743                  * we're doing the csum before we record the extent, but
 744                  * that's ok
 745                  */
 746                 ret = btrfs_csum_file_block(trans, root->fs_info->csum_root,
 747                                             first_block + bytes_read + sectorsize,
 748                                             first_block + bytes_read,
 749                                             eb->data, sectorsize);
 750                 if (ret)
 751                         goto end;
 752
 753                 ret = write_and_map_eb(trans, root, eb);
 754                 if (ret) {
 755                         error("failed to write %s", path_name);
 756                         goto end;
 757                 }
 758
 759                 bytes_read += sectorsize;
 760         }
 761
 762         if (bytes_read) {
 763                 ret = btrfs_record_file_extent(trans, root, objectid, btrfs_inode,
 764                                                file_pos, first_block, cur_bytes);
 765                 if (ret)
 766                         goto end;
 767
 768         }
 769
 770         file_pos += cur_bytes;
 771         total_bytes -= cur_bytes;
 772
 773         if (total_bytes)
 774                 goto again;
 775
 776 end:
 777         free(eb);
 778         close(fd);
 779         return ret;
 780 }
 781
 782 static char *make_path(char *dir, char *name)
 783 {
 784         char *path;
 785
 786         path = malloc(strlen(dir) + strlen(name) + 2);
 787         if (!path)
 788                 return NULL;
 789         strcpy(path, dir);
 790         if (dir[strlen(dir) - 1] != '/')
 791                 strcat(path, "/");
 792         strcat(path, name);
 793         return path;
 794 }
 795
 796 static int traverse_directory(struct btrfs_trans_handle *trans,
 797                               struct btrfs_root *root, char *dir_name,
 798                               struct directory_name_entry *dir_head, int out_fd)
 799 {
 800         int ret = 0;
 801
 802         struct btrfs_inode_item cur_inode;
 803         struct btrfs_inode_item *inode_item;
 804         int count, i, dir_index_cnt;
 805         struct direct **files;
 806         struct stat st;
 807         struct directory_name_entry *dir_entry, *parent_dir_entry;
 808         struct direct *cur_file;
 809         ino_t parent_inum, cur_inum;
 810         ino_t highest_inum = 0;
 811         char *parent_dir_name;
 812         char real_path[PATH_MAX];
 813         struct btrfs_path path;
 814         struct extent_buffer *leaf;
 815         struct btrfs_key root_dir_key;
 816         u64 root_dir_inode_size = 0;
 817
 818         /* Add list for source directory */
 819         dir_entry = malloc(sizeof(struct directory_name_entry));
 820         if (!dir_entry)
 821                 return -ENOMEM;
 822         dir_entry->dir_name = dir_name;
 823         dir_entry->path = realpath(dir_name, real_path);
 824         if (!dir_entry->path) {
 825                 error("realpath  failed for %s: %s", dir_name, strerror(errno));
 826                 ret = -1;
 827                 goto fail_no_dir;
 828         }
 829
 830         parent_inum = highest_inum + BTRFS_FIRST_FREE_OBJECTID;
 831         dir_entry->inum = parent_inum;
 832         list_add_tail(&dir_entry->list, &dir_head->list);
 833
 834         btrfs_init_path(&path);
 835
 836         root_dir_key.objectid = btrfs_root_dirid(&root->root_item);
 837         root_dir_key.offset = 0;
 838         btrfs_set_key_type(&root_dir_key, BTRFS_INODE_ITEM_KEY);
 839         ret = btrfs_lookup_inode(trans, root, &path, &root_dir_key, 1);
 840         if (ret) {
 841                 error("failed to lookup root dir: %d", ret);
 842                 goto fail_no_dir;
 843         }
 844
 845         leaf = path.nodes[0];
 846         inode_item = btrfs_item_ptr(leaf, path.slots[0],
 847                                     struct btrfs_inode_item);
 848
 849         root_dir_inode_size = calculate_dir_inode_size(dir_name);
 850         btrfs_set_inode_size(leaf, inode_item, root_dir_inode_size);
 851         btrfs_mark_buffer_dirty(leaf);
 852
 853         btrfs_release_path(&path);
 854
 855         do {
 856                 parent_dir_entry = list_entry(dir_head->list.next,
 857                                               struct directory_name_entry,
 858                                               list);
 859                 list_del(&parent_dir_entry->list);
 860
 861                 parent_inum = parent_dir_entry->inum;
 862                 parent_dir_name = parent_dir_entry->dir_name;
 863                 if (chdir(parent_dir_entry->path)) {
 864                         error("chdir failed for %s: %s",
 865                                 parent_dir_name, strerror(errno));
 866                         ret = -1;
 867                         goto fail_no_files;
 868                 }
 869
 870                 count = scandir(parent_dir_entry->path, &files,
 871                                 directory_select, NULL);
 872                 if (count == -1)
 873                 {
 874                         error("scandir failed for %s: %s",
 875                                 parent_dir_name, strerror (errno));
 876                         ret = -1;
 877                         goto fail;
 878                 }
 879
 880                 for (i = 0; i < count; i++) {
 881                         cur_file = files[i];
 882
 883                         if (lstat(cur_file->d_name, &st) == -1) {
 884                                 error("lstat failed for %s: %s",
 885                                         cur_file->d_name, strerror(errno));
 886                                 ret = -1;
 887                                 goto fail;
 888                         }
 889
 890                         cur_inum = st.st_ino;
 891                         ret = add_directory_items(trans, root,
 892                                                   cur_inum, parent_inum,
 893                                                   cur_file->d_name,
 894                                                   &st, &dir_index_cnt);
 895                         if (ret) {
 896                                 error("unable to add directory items for %s: %d",
 897                                         cur_file->d_name, ret);
 898                                 goto fail;
 899                         }
 900
 901                         ret = add_inode_items(trans, root, &st,
 902                                               cur_file->d_name, cur_inum,
 903                                               parent_inum, dir_index_cnt,
 904                                               &cur_inode);
 905                         if (ret == -EEXIST) {
 906                                 if (st.st_nlink <= 1) {
 907                                         error(
 908                         "item %s already exists but has wrong st_nlink %ld <= 1",
 909                                                 cur_file->d_name, st.st_nlink);
 910                                         goto fail;
 911                                 }
 912                                 continue;
 913                         }
 914                         if (ret) {
 915                                 error("unable to add inode items for %s: %d",
 916                                         cur_file->d_name, ret);
 917                                 goto fail;
 918                         }
 919
 920                         ret = add_xattr_item(trans, root,
 921                                              cur_inum, cur_file->d_name);
 922                         if (ret) {
 923                                 error("unable to add xattr items for %s: %d",
 924                                         cur_file->d_name, ret);
 925                                 if(ret != -ENOTSUP)
 926                                         goto fail;
 927                         }
 928
 929                         if (S_ISDIR(st.st_mode)) {
 930                                 dir_entry = malloc(sizeof(struct directory_name_entry));
 931                                 if (!dir_entry) {
 932                                         ret = -ENOMEM;
 933                                         goto fail;
 934                                 }
 935                                 dir_entry->dir_name = cur_file->d_name;
 936                                 dir_entry->path = make_path(parent_dir_entry->path,
 937                                                             cur_file->d_name);
 938                                 dir_entry->inum = cur_inum;
 939                                 list_add_tail(&dir_entry->list, &dir_head->list);
 940                         } else if (S_ISREG(st.st_mode)) {
 941                                 ret = add_file_items(trans, root, &cur_inode,
 942                                                      cur_inum, parent_inum, &st,
 943                                                      cur_file->d_name, out_fd);
 944                                 if (ret) {
 945                                         error("unable to add file items for %s: %d",
 946                                                 cur_file->d_name, ret);
 947                                         goto fail;
 948                                 }
 949                         } else if (S_ISLNK(st.st_mode)) {
 950                                 ret = add_symbolic_link(trans, root,
 951                                                         cur_inum, cur_file->d_name);
 952                                 if (ret) {
 953                                         error("unable to add symlink for %s: %d",
 954                                                 cur_file->d_name, ret);
 955                                         goto fail;
 956                                 }
 957                         }
 958                 }
 959
 960                 free_namelist(files, count);
 961                 free(parent_dir_entry);
 962
 963                 index_cnt = 2;
 964
 965         } while (!list_empty(&dir_head->list));
 966
 967 out:
 968         return !!ret;
 969 fail:
 970         free_namelist(files, count);
 971 fail_no_files:
 972         free(parent_dir_entry);
 973         goto out;
 974 fail_no_dir:
 975         free(dir_entry);
 976         goto out;
 977 }
 978
 979 static int open_target(char *output_name)
 980 {
 981         int output_fd;
 982         output_fd = open(output_name, O_CREAT | O_RDWR,
 983                          S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH);
 984
 985         return output_fd;
 986 }
 987
 988 static int create_chunks(struct btrfs_trans_handle *trans,
 989                          struct btrfs_root *root, u64 num_of_meta_chunks,
 990                          u64 size_of_data,
 991                          struct mkfs_allocation *allocation)
 992 {
 993         u64 chunk_start;
 994         u64 chunk_size;
 995         u64 meta_type = BTRFS_BLOCK_GROUP_METADATA;
 996         u64 data_type = BTRFS_BLOCK_GROUP_DATA;
 997         u64 minimum_data_chunk_size = 8 * 1024 * 1024;
 998         u64 i;
 999         int ret;
1000
1001         for (i = 0; i < num_of_meta_chunks; i++) {
1002                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
1003                                         &chunk_start, &chunk_size, meta_type);
1004                 if (ret)
1005                         return ret;
1006                 ret = btrfs_make_block_group(trans, root->fs_info->extent_root, 0,
1007                                              meta_type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1008                                              chunk_start, chunk_size);
1009                 allocation->metadata += chunk_size;
1010                 if (ret)
1011                         return ret;
1012                 set_extent_dirty(&root->fs_info->free_space_cache,
1013                                  chunk_start, chunk_start + chunk_size - 1, 0);
1014         }
1015
1016         if (size_of_data < minimum_data_chunk_size)
1017                 size_of_data = minimum_data_chunk_size;
1018
1019         ret = btrfs_alloc_data_chunk(trans, root->fs_info->extent_root,
1020                                      &chunk_start, size_of_data, data_type, 0);
1021         if (ret)
1022                 return ret;
1023         ret = btrfs_make_block_group(trans, root->fs_info->extent_root, 0,
1024                                      data_type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1025                                      chunk_start, size_of_data);
1026         allocation->data += size_of_data;
1027         if (ret)
1028                 return ret;
1029         set_extent_dirty(&root->fs_info->free_space_cache,
1030                          chunk_start, chunk_start + size_of_data - 1, 0);
1031         return ret;
1032 }
1033
1034 static int make_image(char *source_dir, struct btrfs_root *root, int out_fd)
1035 {
1036         int ret;
1037         struct btrfs_trans_handle *trans;
1038         struct stat root_st;
1039         struct directory_name_entry dir_head;
1040         struct directory_name_entry *dir_entry = NULL;
1041
1042         ret = lstat(source_dir, &root_st);
1043         if (ret) {
1044                 error("unable to lstat %s: %s", source_dir, strerror(errno));
1045                 ret = -errno;
1046                 goto out;
1047         }
1048
1049         INIT_LIST_HEAD(&dir_head.list);
1050
1051         trans = btrfs_start_transaction(root, 1);
1052         ret = traverse_directory(trans, root, source_dir, &dir_head, out_fd);
1053         if (ret) {
1054                 error("unable to traverse directory %s: %d", source_dir, ret);
1055                 goto fail;
1056         }
1057         ret = btrfs_commit_transaction(trans, root);
1058         if (ret) {
1059                 error("transaction commit failed: %d", ret);
1060                 goto out;
1061         }
1062
1063         if (verbose)
1064                 printf("Making image is completed.\n");
1065         return 0;
1066 fail:
1067         while (!list_empty(&dir_head.list)) {
1068                 dir_entry = list_entry(dir_head.list.next,
1069                                        struct directory_name_entry, list);
1070                 list_del(&dir_entry->list);
1071                 free(dir_entry);
1072         }
1073 out:
1074         return ret;
1075 }
1076
1077 /*
1078  * This ignores symlinks with unreadable targets and subdirs that can't
1079  * be read.  It's a best-effort to give a rough estimate of the size of
1080  * a subdir.  It doesn't guarantee that prepopulating btrfs from this
1081  * tree won't still run out of space.
1082  */
1083 static u64 global_total_size;
1084 static u64 fs_block_size;
1085 static int ftw_add_entry_size(const char *fpath, const struct stat *st,
1086                               int type)
1087 {
1088         if (type == FTW_F || type == FTW_D)
1089                 global_total_size += round_up(st->st_size, fs_block_size);
1090
1091         return 0;
1092 }
1093
1094 static u64 size_sourcedir(char *dir_name, u64 sectorsize,
1095                           u64 *num_of_meta_chunks_ret, u64 *size_of_data_ret)
1096 {
1097         u64 dir_size = 0;
1098         u64 total_size = 0;
1099         int ret;
1100         u64 default_chunk_size = 8 * 1024 * 1024;       /* 8MB */
1101         u64 allocated_meta_size = 8 * 1024 * 1024;      /* 8MB */
1102         u64 allocated_total_size = 20 * 1024 * 1024;    /* 20MB */
1103         u64 num_of_meta_chunks = 0;
1104         u64 num_of_data_chunks = 0;
1105         u64 num_of_allocated_meta_chunks =
1106                         allocated_meta_size / default_chunk_size;
1107
1108         global_total_size = 0;
1109         fs_block_size = sectorsize;
1110         ret = ftw(dir_name, ftw_add_entry_size, 10);
1111         dir_size = global_total_size;
1112         if (ret < 0) {
1113                 error("ftw subdir walk of %s failed: %s", dir_name,
1114                         strerror(errno));
1115                 exit(1);
1116         }
1117
1118         num_of_data_chunks = (dir_size + default_chunk_size - 1) /
1119                 default_chunk_size;
1120
1121         num_of_meta_chunks = (dir_size / 2) / default_chunk_size;
1122         if (((dir_size / 2) % default_chunk_size) != 0)
1123                 num_of_meta_chunks++;
1124         if (num_of_meta_chunks <= num_of_allocated_meta_chunks)
1125                 num_of_meta_chunks = 0;
1126         else
1127                 num_of_meta_chunks -= num_of_allocated_meta_chunks;
1128
1129         total_size = allocated_total_size +
1130                      (num_of_data_chunks * default_chunk_size) +
1131                      (num_of_meta_chunks * default_chunk_size);
1132
1133         *num_of_meta_chunks_ret = num_of_meta_chunks;
1134         *size_of_data_ret = num_of_data_chunks * default_chunk_size;
1135         return total_size;
1136 }
1137
1138 static int zero_output_file(int out_fd, u64 size)
1139 {
1140         int loop_num;
1141         u64 location = 0;
1142         char buf[4096];
1143         int ret = 0, i;
1144         ssize_t written;
1145
1146         memset(buf, 0, 4096);
1147         loop_num = size / 4096;
1148         for (i = 0; i < loop_num; i++) {
1149                 written = pwrite64(out_fd, buf, 4096, location);
1150                 if (written != 4096)
1151                         ret = -EIO;
1152                 location += 4096;
1153         }
1154         return ret;
1155 }
1156
1157 static int is_ssd(const char *file)
1158 {
1159         blkid_probe probe;
1160         char wholedisk[PATH_MAX];
1161         char sysfs_path[PATH_MAX];
1162         dev_t devno;
1163         int fd;
1164         char rotational;
1165         int ret;
1166
1167         probe = blkid_new_probe_from_filename(file);
1168         if (!probe)
1169                 return 0;
1170
1171         /* Device number of this disk (possibly a partition) */
1172         devno = blkid_probe_get_devno(probe);
1173         if (!devno) {
1174                 blkid_free_probe(probe);
1175                 return 0;
1176         }
1177
1178         /* Get whole disk name (not full path) for this devno */
1179         ret = blkid_devno_to_wholedisk(devno,
1180                         wholedisk, sizeof(wholedisk), NULL);
1181         if (ret) {
1182                 blkid_free_probe(probe);
1183                 return 0;
1184         }
1185
1186         snprintf(sysfs_path, PATH_MAX, "/sys/block/%s/queue/rotational",
1187                  wholedisk);
1188
1189         blkid_free_probe(probe);
1190
1191         fd = open(sysfs_path, O_RDONLY);
1192         if (fd < 0) {
1193                 return 0;
1194         }
1195
1196         if (read(fd, &rotational, 1) < 1) {
1197                 close(fd);
1198                 return 0;
1199         }
1200         close(fd);
1201
1202         return rotational == '0';
1203 }
1204
1205 static int _cmp_device_by_id(void *priv, struct list_head *a,
1206                              struct list_head *b)
1207 {
1208         return list_entry(a, struct btrfs_device, dev_list)->devid -
1209                list_entry(b, struct btrfs_device, dev_list)->devid;
1210 }
1211
1212 static void list_all_devices(struct btrfs_root *root)
1213 {
1214         struct btrfs_fs_devices *fs_devices;
1215         struct btrfs_device *device;
1216         int number_of_devices = 0;
1217         u64 total_block_count = 0;
1218
1219         fs_devices = root->fs_info->fs_devices;
1220
1221         list_for_each_entry(device, &fs_devices->devices, dev_list)
1222                 number_of_devices++;
1223
1224         list_sort(NULL, &fs_devices->devices, _cmp_device_by_id);
1225
1226         printf("Number of devices:  %d\n", number_of_devices);
1227         /* printf("Total devices size: %10s\n", */
1228                 /* pretty_size(total_block_count)); */
1229         printf("Devices:\n");
1230         printf("   ID        SIZE  PATH\n");
1231         list_for_each_entry(device, &fs_devices->devices, dev_list) {
1232                 printf("  %3llu  %10s  %s\n",
1233                         device->devid,
1234                         pretty_size(device->total_bytes),
1235                         device->name);
1236                 total_block_count += device->total_bytes;
1237         }
1238
1239         printf("\n");
1240 }
1241
1242 static int is_temp_block_group(struct extent_buffer *node,
1243                                struct btrfs_block_group_item *bgi,
1244                                u64 data_profile, u64 meta_profile,
1245                                u64 sys_profile)
1246 {
1247         u64 flag = btrfs_disk_block_group_flags(node, bgi);
1248         u64 flag_type = flag & BTRFS_BLOCK_GROUP_TYPE_MASK;
1249         u64 flag_profile = flag & BTRFS_BLOCK_GROUP_PROFILE_MASK;
1250         u64 used = btrfs_disk_block_group_used(node, bgi);
1251
1252         /*
1253          * Chunks meets all the following conditions is a temp chunk
1254          * 1) Empty chunk
1255          * Temp chunk is always empty.
1256          *
1257          * 2) profile mismatch with mkfs profile.
1258          * Temp chunk is always in SINGLE
1259          *
1260          * 3) Size differs with mkfs_alloc
1261          * Special case for SINGLE/SINGLE btrfs.
1262          * In that case, temp data chunk and real data chunk are always empty.
1263          * So we need to use mkfs_alloc to be sure which chunk is the newly
1264          * allocated.
1265          *
1266          * Normally, new chunk size is equal to mkfs one (One chunk)
1267          * If it has multiple chunks, we just refuse to delete any one.
1268          * As they are all single, so no real problem will happen.
1269          * So only use condition 1) and 2) to judge them.
1270          */
1271         if (used != 0)
1272                 return 0;
1273         switch (flag_type) {
1274         case BTRFS_BLOCK_GROUP_DATA:
1275         case BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA:
1276                 data_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1277                 if (flag_profile != data_profile)
1278                         return 1;
1279                 break;
1280         case BTRFS_BLOCK_GROUP_METADATA:
1281                 meta_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1282                 if (flag_profile != meta_profile)
1283                         return 1;
1284                 break;
1285         case BTRFS_BLOCK_GROUP_SYSTEM:
1286                 sys_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1287                 if (flag_profile != sys_profile)
1288                         return 1;
1289                 break;
1290         }
1291         return 0;
1292 }
1293
1294 /* Note: if current is a block group, it will skip it anyway */
1295 static int next_block_group(struct btrfs_root *root,
1296                             struct btrfs_path *path)
1297 {
1298         struct btrfs_key key;
1299         int ret = 0;
1300
1301         while (1) {
1302                 ret = btrfs_next_item(root, path);
1303                 if (ret)
1304                         goto out;
1305
1306                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1307                 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
1308                         goto out;
1309         }
1310 out:
1311         return ret;
1312 }
1313
1314 /* This function will cleanup  */
1315 static int cleanup_temp_chunks(struct btrfs_fs_info *fs_info,
1316                                struct mkfs_allocation *alloc,
1317                                u64 data_profile, u64 meta_profile,
1318                                u64 sys_profile)
1319 {
1320         struct btrfs_trans_handle *trans = NULL;
1321         struct btrfs_block_group_item *bgi;
1322         struct btrfs_root *root = fs_info->extent_root;
1323         struct btrfs_key key;
1324         struct btrfs_key found_key;
1325         struct btrfs_path *path;
1326         int ret = 0;
1327
1328         path = btrfs_alloc_path();
1329         if (!path) {
1330                 ret = -ENOMEM;
1331                 goto out;
1332         }
1333
1334         trans = btrfs_start_transaction(root, 1);
1335
1336         key.objectid = 0;
1337         key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
1338         key.offset = 0;
1339
1340         while (1) {
1341                 /*
1342                  * as the rest of the loop may modify the tree, we need to
1343                  * start a new search each time.
1344                  */
1345                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
1346                 if (ret < 0)
1347                         goto out;
1348
1349                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1350                                       path->slots[0]);
1351                 if (found_key.objectid < key.objectid)
1352                         goto out;
1353                 if (found_key.type != BTRFS_BLOCK_GROUP_ITEM_KEY) {
1354                         ret = next_block_group(root, path);
1355                         if (ret < 0)
1356                                 goto out;
1357                         if (ret > 0) {
1358                                 ret = 0;
1359                                 goto out;
1360                         }
1361                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1362                                               path->slots[0]);
1363                 }
1364
1365                 bgi = btrfs_item_ptr(path->nodes[0], path->slots[0],
1366                                      struct btrfs_block_group_item);
1367                 if (is_temp_block_group(path->nodes[0], bgi,
1368                                         data_profile, meta_profile,
1369                                         sys_profile)) {
1370                         u64 flags = btrfs_disk_block_group_flags(path->nodes[0],
1371                                                              bgi);
1372
1373                         ret = btrfs_free_block_group(trans, fs_info,
1374                                         found_key.objectid, found_key.offset);
1375                         if (ret < 0)
1376                                 goto out;
1377
1378                         if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
1379                             BTRFS_BLOCK_GROUP_DATA)
1380                                 alloc->data -= found_key.offset;
1381                         else if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
1382                                  BTRFS_BLOCK_GROUP_METADATA)
1383                                 alloc->metadata -= found_key.offset;
1384                         else if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
1385                                  BTRFS_BLOCK_GROUP_SYSTEM)
1386                                 alloc->system -= found_key.offset;
1387                         else if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
1388                                  (BTRFS_BLOCK_GROUP_METADATA |
1389                                   BTRFS_BLOCK_GROUP_DATA))
1390                                 alloc->mixed -= found_key.offset;
1391                 }
1392                 btrfs_release_path(path);
1393                 key.objectid = found_key.objectid + found_key.offset;
1394         }
1395 out:
1396         if (trans)
1397                 btrfs_commit_transaction(trans, root);
1398         btrfs_free_path(path);
1399         return ret;
1400 }
1401
1402 int main(int argc, char **argv)
1403 {
1404         char *file;
1405         struct btrfs_root *root;
1406         struct btrfs_fs_info *fs_info;
1407         struct btrfs_trans_handle *trans;
1408         char *label = NULL;
1409         u64 block_count = 0;
1410         u64 dev_block_count = 0;
1411         u64 blocks[7];
1412         u64 alloc_start = 0;
1413         u64 metadata_profile = 0;
1414         u64 data_profile = 0;
1415         u32 nodesize = max_t(u32, sysconf(_SC_PAGESIZE),
1416                         BTRFS_MKFS_DEFAULT_NODE_SIZE);
1417         u32 sectorsize = 4096;
1418         u32 stripesize = 4096;
1419         int zero_end = 1;
1420         int fd;
1421         int ret;
1422         int i;
1423         int mixed = 0;
1424         int nodesize_forced = 0;
1425         int data_profile_opt = 0;
1426         int metadata_profile_opt = 0;
1427         int discard = 1;
1428         int ssd = 0;
1429         int force_overwrite = 0;
1430         char *source_dir = NULL;
1431         int source_dir_set = 0;
1432         u64 num_of_meta_chunks = 0;
1433         u64 size_of_data = 0;
1434         u64 source_dir_size = 0;
1435         int dev_cnt = 0;
1436         int saved_optind;
1437         char fs_uuid[BTRFS_UUID_UNPARSED_SIZE] = { 0 };
1438         u64 features = BTRFS_MKFS_DEFAULT_FEATURES;
1439         struct mkfs_allocation allocation = { 0 };
1440         struct btrfs_mkfs_config mkfs_cfg;
1441
1442         while(1) {
1443                 int c;
1444                 static const struct option long_options[] = {
1445                         { "alloc-start", required_argument, NULL, 'A'},
1446                         { "byte-count", required_argument, NULL, 'b' },
1447                         { "force", no_argument, NULL, 'f' },
1448                         { "leafsize", required_argument, NULL, 'l' },
1449                         { "label", required_argument, NULL, 'L'},
1450                         { "metadata", required_argument, NULL, 'm' },
1451                         { "mixed", no_argument, NULL, 'M' },
1452                         { "nodesize", required_argument, NULL, 'n' },
1453                         { "sectorsize", required_argument, NULL, 's' },
1454                         { "data", required_argument, NULL, 'd' },
1455                         { "version", no_argument, NULL, 'V' },
1456                         { "rootdir", required_argument, NULL, 'r' },
1457                         { "nodiscard", no_argument, NULL, 'K' },
1458                         { "features", required_argument, NULL, 'O' },
1459                         { "uuid", required_argument, NULL, 'U' },
1460                         { "quiet", 0, NULL, 'q' },
1461                         { "help", no_argument, NULL, GETOPT_VAL_HELP },
1462                         { NULL, 0, NULL, 0}
1463                 };
1464
1465                 c = getopt_long(argc, argv, "A:b:fl:n:s:m:d:L:O:r:U:VMKq",
1466                                 long_options, NULL);
1467                 if (c < 0)
1468                         break;
1469                 switch(c) {
1470                         case 'A':
1471                                 alloc_start = parse_size(optarg);
1472                                 break;
1473                         case 'f':
1474                                 force_overwrite = 1;
1475                                 break;
1476                         case 'd':
1477                                 data_profile = parse_profile(optarg);
1478                                 data_profile_opt = 1;
1479                                 break;
1480                         case 'l':
1481                                 warning("--leafsize is deprecated, use --nodesize");
1482                         case 'n':
1483                                 nodesize = parse_size(optarg);
1484                                 nodesize_forced = 1;
1485                                 break;
1486                         case 'L':
1487                                 label = parse_label(optarg);
1488                                 break;
1489                         case 'm':
1490                                 metadata_profile = parse_profile(optarg);
1491                                 metadata_profile_opt = 1;
1492                                 break;
1493                         case 'M':
1494                                 mixed = 1;
1495                                 break;
1496                         case 'O': {
1497                                 char *orig = strdup(optarg);
1498                                 char *tmp = orig;
1499
1500                                 tmp = btrfs_parse_fs_features(tmp, &features);
1501                                 if (tmp) {
1502                                         error("unrecognized filesystem feature '%s'",
1503                                                         tmp);
1504                                         free(orig);
1505                                         exit(1);
1506                                 }
1507                                 free(orig);
1508                                 if (features & BTRFS_FEATURE_LIST_ALL) {
1509                                         btrfs_list_all_fs_features(0);
1510                                         exit(0);
1511                                 }
1512                                 break;
1513                                 }
1514                         case 's':
1515                                 sectorsize = parse_size(optarg);
1516                                 break;
1517                         case 'b':
1518                                 block_count = parse_size(optarg);
1519                                 zero_end = 0;
1520                                 break;
1521                         case 'V':
1522                                 printf("mkfs.btrfs, part of %s\n",
1523                                                 PACKAGE_STRING);
1524                                 exit(0);
1525                                 break;
1526                         case 'r':
1527                                 source_dir = optarg;
1528                                 source_dir_set = 1;
1529                                 break;
1530                         case 'U':
1531                                 strncpy(fs_uuid, optarg,
1532                                         BTRFS_UUID_UNPARSED_SIZE - 1);
1533                                 break;
1534                         case 'K':
1535                                 discard = 0;
1536                                 break;
1537                         case 'q':
1538                                 verbose = 0;
1539                                 break;
1540                         case GETOPT_VAL_HELP:
1541                         default:
1542                                 print_usage(c != GETOPT_VAL_HELP);
1543                 }
1544         }
1545
1546         if (verbose) {
1547                 printf("%s\n", PACKAGE_STRING);
1548                 printf("See %s for more information.\n\n", PACKAGE_URL);
1549         }
1550
1551         sectorsize = max(sectorsize, (u32)sysconf(_SC_PAGESIZE));
1552         stripesize = sectorsize;
1553         saved_optind = optind;
1554         dev_cnt = argc - optind;
1555         if (dev_cnt == 0)
1556                 print_usage(1);
1557
1558         if (source_dir_set && dev_cnt > 1) {
1559                 error("the option -r is limited to a single device");
1560                 exit(1);
1561         }
1562
1563         if (*fs_uuid) {
1564                 uuid_t dummy_uuid;
1565
1566                 if (uuid_parse(fs_uuid, dummy_uuid) != 0) {
1567                         error("could not parse UUID: %s", fs_uuid);
1568                         exit(1);
1569                 }
1570                 if (!test_uuid_unique(fs_uuid)) {
1571                         error("non-unique UUID: %s", fs_uuid);
1572                         exit(1);
1573                 }
1574         }
1575
1576         while (dev_cnt-- > 0) {
1577                 file = argv[optind++];
1578                 if (is_block_device(file) == 1)
1579                         if (test_dev_for_mkfs(file, force_overwrite))
1580                                 exit(1);
1581         }
1582
1583         optind = saved_optind;
1584         dev_cnt = argc - optind;
1585
1586         file = argv[optind++];
1587         ssd = is_ssd(file);
1588
1589         /*
1590         * Set default profiles according to number of added devices.
1591         * For mixed groups defaults are single/single.
1592         */
1593         if (!mixed) {
1594                 if (!metadata_profile_opt) {
1595                         if (dev_cnt == 1 && ssd && verbose)
1596                                 printf("Detected a SSD, turning off metadata "
1597                                 "duplication.  Mkfs with -m dup if you want to "
1598                                 "force metadata duplication.\n");
1599
1600                         metadata_profile = (dev_cnt > 1) ?
1601                                         BTRFS_BLOCK_GROUP_RAID1 : (ssd) ?
1602                                         0: BTRFS_BLOCK_GROUP_DUP;
1603                 }
1604                 if (!data_profile_opt) {
1605                         data_profile = (dev_cnt > 1) ?
1606                                 BTRFS_BLOCK_GROUP_RAID0 : 0; /* raid0 or single */
1607                 }
1608         } else {
1609                 u32 best_nodesize = max_t(u32, sysconf(_SC_PAGESIZE), sectorsize);
1610
1611                 if (metadata_profile_opt || data_profile_opt) {
1612                         if (metadata_profile != data_profile) {
1613                                 error(
1614         "with mixed block groups data and metadata profiles must be the same");
1615                                 exit(1);
1616                         }
1617                 }
1618
1619                 if (!nodesize_forced)
1620                         nodesize = best_nodesize;
1621         }
1622
1623         /*
1624          * FS features that can be set by other means than -O
1625          * just set the bit here
1626          */
1627         if (mixed)
1628                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS;
1629
1630         if ((data_profile | metadata_profile) &
1631             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
1632                 features |= BTRFS_FEATURE_INCOMPAT_RAID56;
1633         }
1634
1635         if (btrfs_check_nodesize(nodesize, sectorsize,
1636                                  features))
1637                 exit(1);
1638
1639         if (sectorsize < sizeof(struct btrfs_super_block)) {
1640                 error("sectorsize smaller than superblock: %u < %zu",
1641                                 sectorsize, sizeof(struct btrfs_super_block));
1642                 exit(1);
1643         }
1644
1645         /* Check device/block_count after the nodesize is determined */
1646         if (block_count && block_count < btrfs_min_dev_size(nodesize)) {
1647                 error("size %llu is too small to make a usable filesystem",
1648                         block_count);
1649                 error("minimum size for btrfs filesystem is %llu",
1650                         btrfs_min_dev_size(nodesize));
1651                 exit(1);
1652         }
1653         for (i = saved_optind; i < saved_optind + dev_cnt; i++) {
1654                 char *path;
1655
1656                 path = argv[i];
1657                 ret = test_minimum_size(path, nodesize);
1658                 if (ret < 0) {
1659                         error("failed to check size for %s: %s",
1660                                 path, strerror(-ret));
1661                         exit (1);
1662                 }
1663                 if (ret > 0) {
1664                         error("'%s' is too small to make a usable filesystem",
1665                                 path);
1666                         error("minimum size for each btrfs device is %llu",
1667                                 btrfs_min_dev_size(nodesize));
1668                         exit(1);
1669                 }
1670         }
1671         ret = test_num_disk_vs_raid(metadata_profile, data_profile,
1672                         dev_cnt, mixed, ssd);
1673         if (ret)
1674                 exit(1);
1675
1676         dev_cnt--;
1677
1678         if (!source_dir_set) {
1679                 /*
1680                  * open without O_EXCL so that the problem should not
1681                  * occur by the following processing.
1682                  * (btrfs_register_one_device() fails if O_EXCL is on)
1683                  */
1684                 fd = open(file, O_RDWR);
1685                 if (fd < 0) {
1686                         error("unable to open %s: %s", file, strerror(errno));
1687                         exit(1);
1688                 }
1689                 ret = btrfs_prepare_device(fd, file, &dev_block_count,
1690                                 block_count,
1691                                 (zero_end ? PREP_DEVICE_ZERO_END : 0) |
1692                                 (discard ? PREP_DEVICE_DISCARD : 0) |
1693                                 (verbose ? PREP_DEVICE_VERBOSE : 0));
1694                 if (ret) {
1695                         close(fd);
1696                         exit(1);
1697                 }
1698                 if (block_count && block_count > dev_block_count) {
1699                         error("%s is smaller than requested size, expected %llu, found %llu",
1700                                         file,
1701                                         (unsigned long long)block_count,
1702                                         (unsigned long long)dev_block_count);
1703                         exit(1);
1704                 }
1705         } else {
1706                 fd = open_target(file);
1707                 if (fd < 0) {
1708                         error("unable to open %s: %s", file, strerror(errno));
1709                         exit(1);
1710                 }
1711
1712                 source_dir_size = size_sourcedir(source_dir, sectorsize,
1713                                              &num_of_meta_chunks, &size_of_data);
1714                 if(block_count < source_dir_size)
1715                         block_count = source_dir_size;
1716                 ret = zero_output_file(fd, block_count);
1717                 if (ret) {
1718                         error("unable to zero the output file");
1719                         exit(1);
1720                 }
1721                 /* our "device" is the new image file */
1722                 dev_block_count = block_count;
1723         }
1724
1725         /* To create the first block group and chunk 0 in make_btrfs */
1726         if (dev_block_count < BTRFS_MKFS_SYSTEM_GROUP_SIZE) {
1727                 error("device is too small to make filesystem, must be at least %llu",
1728                                 (unsigned long long)BTRFS_MKFS_SYSTEM_GROUP_SIZE);
1729                 exit(1);
1730         }
1731
1732         blocks[0] = BTRFS_SUPER_INFO_OFFSET;
1733         for (i = 1; i < 7; i++) {
1734                 blocks[i] = BTRFS_SUPER_INFO_OFFSET + 1024 * 1024 +
1735                         nodesize * i;
1736         }
1737
1738         if (group_profile_max_safe_loss(metadata_profile) <
1739                 group_profile_max_safe_loss(data_profile)){
1740                 warning("metadata has lower redundancy than data!\n");
1741         }
1742
1743         mkfs_cfg.label = label;
1744         memcpy(mkfs_cfg.fs_uuid, fs_uuid, sizeof(mkfs_cfg.fs_uuid));
1745         memcpy(mkfs_cfg.blocks, blocks, sizeof(blocks));
1746         mkfs_cfg.num_bytes = dev_block_count;
1747         mkfs_cfg.nodesize = nodesize;
1748         mkfs_cfg.sectorsize = sectorsize;
1749         mkfs_cfg.stripesize = stripesize;
1750         mkfs_cfg.features = features;
1751
1752         ret = make_btrfs(fd, &mkfs_cfg, NULL);
1753         if (ret) {
1754                 error("error during mkfs: %s", strerror(-ret));
1755                 exit(1);
1756         }
1757
1758         fs_info = open_ctree_fs_info(file, 0, 0, 0,
1759                         OPEN_CTREE_WRITES | OPEN_CTREE_FS_PARTIAL);
1760         if (!fs_info) {
1761                 error("open ctree failed");
1762                 close(fd);
1763                 exit(1);
1764         }
1765         root = fs_info->fs_root;
1766         fs_info->alloc_start = alloc_start;
1767
1768         ret = create_metadata_block_groups(root, mixed, &allocation);
1769         if (ret) {
1770                 error("failed to create default block groups: %d", ret);
1771                 exit(1);
1772         }
1773
1774         trans = btrfs_start_transaction(root, 1);
1775         if (!trans) {
1776                 error("failed to start transaction");
1777                 exit(1);
1778         }
1779
1780         ret = create_data_block_groups(trans, root, mixed, &allocation);
1781         if (ret) {
1782                 error("failed to create default data block groups: %d", ret);
1783                 exit(1);
1784         }
1785
1786         ret = make_root_dir(trans, root, &allocation);
1787         if (ret) {
1788                 error("failed to setup the root directory: %d", ret);
1789                 exit(1);
1790         }
1791
1792         ret = btrfs_commit_transaction(trans, root);
1793         if (ret) {
1794                 error("unable to commit transaction: %d", ret);
1795                 goto out;
1796         }
1797
1798         trans = btrfs_start_transaction(root, 1);
1799         if (!trans) {
1800                 error("failed to start transaction");
1801                 exit(1);
1802         }
1803
1804         if (dev_cnt == 0)
1805                 goto raid_groups;
1806
1807         while (dev_cnt-- > 0) {
1808                 file = argv[optind++];
1809
1810                 /*
1811                  * open without O_EXCL so that the problem should not
1812                  * occur by the following processing.
1813                  * (btrfs_register_one_device() fails if O_EXCL is on)
1814                  */
1815                 fd = open(file, O_RDWR);
1816                 if (fd < 0) {
1817                         error("unable to open %s: %s", file, strerror(errno));
1818                         exit(1);
1819                 }
1820                 ret = btrfs_device_already_in_root(root, fd,
1821                                                    BTRFS_SUPER_INFO_OFFSET);
1822                 if (ret) {
1823                         error("skipping duplicate device %s in the filesystem",
1824                                 file);
1825                         close(fd);
1826                         continue;
1827                 }
1828                 ret = btrfs_prepare_device(fd, file, &dev_block_count,
1829                                 block_count,
1830                                 (verbose ? PREP_DEVICE_VERBOSE : 0) |
1831                                 (zero_end ? PREP_DEVICE_ZERO_END : 0) |
1832                                 (discard ? PREP_DEVICE_DISCARD : 0));
1833                 if (ret) {
1834                         close(fd);
1835                         exit(1);
1836                 }
1837
1838                 ret = btrfs_add_to_fsid(trans, root, fd, file, dev_block_count,
1839                                         sectorsize, sectorsize, sectorsize);
1840                 if (ret) {
1841                         error("unable to add %s to filesystem: %d", file, ret);
1842                         goto out;
1843                 }
1844                 if (verbose >= 2) {
1845                         struct btrfs_device *device;
1846
1847                         device = container_of(fs_info->fs_devices->devices.next,
1848                                         struct btrfs_device, dev_list);
1849                         printf("adding device %s id %llu\n", file,
1850                                 (unsigned long long)device->devid);
1851                 }
1852         }
1853
1854 raid_groups:
1855         if (!source_dir_set) {
1856                 ret = create_raid_groups(trans, root, data_profile,
1857                                  metadata_profile, mixed, &allocation);
1858                 if (ret) {
1859                         error("unable to create raid groups: %d", ret);
1860                         goto out;
1861                 }
1862         }
1863
1864         ret = create_data_reloc_tree(trans, root);
1865         if (ret) {
1866                 error("unable to create data reloc tree: %d", ret);
1867                 goto out;
1868         }
1869
1870         ret = btrfs_commit_transaction(trans, root);
1871         if (ret) {
1872                 error("unable to commit transaction: %d", ret);
1873                 goto out;
1874         }
1875
1876         if (source_dir_set) {
1877                 trans = btrfs_start_transaction(root, 1);
1878                 ret = create_chunks(trans, root,
1879                                     num_of_meta_chunks, size_of_data,
1880                                     &allocation);
1881                 if (ret) {
1882                         error("unable to create chunks: %d", ret);
1883                         goto out;
1884                 }
1885                 ret = btrfs_commit_transaction(trans, root);
1886                 if (ret) {
1887                         error("transaction commit failed: %d", ret);
1888                         goto out;
1889                 }
1890
1891                 ret = make_image(source_dir, root, fd);
1892                 if (ret) {
1893                         error("error wihle filling filesystem: %d", ret);
1894                         goto out;
1895                 }
1896         }
1897         ret = cleanup_temp_chunks(fs_info, &allocation, data_profile,
1898                                   metadata_profile, metadata_profile);
1899         if (ret < 0) {
1900                 error("failed to cleanup temporary chunks: %d", ret);
1901                 goto out;
1902         }
1903
1904         if (verbose) {
1905                 char features_buf[64];
1906
1907                 printf("Label:              %s\n", label);
1908                 printf("UUID:               %s\n", fs_uuid);
1909                 printf("Node size:          %u\n", nodesize);
1910                 printf("Sector size:        %u\n", sectorsize);
1911                 printf("Filesystem size:    %s\n",
1912                         pretty_size(btrfs_super_total_bytes(fs_info->super_copy)));
1913                 printf("Block group profiles:\n");
1914                 if (allocation.data)
1915                         printf("  Data:             %-8s %16s\n",
1916                                 btrfs_group_profile_str(data_profile),
1917                                 pretty_size(allocation.data));
1918                 if (allocation.metadata)
1919                         printf("  Metadata:         %-8s %16s\n",
1920                                 btrfs_group_profile_str(metadata_profile),
1921                                 pretty_size(allocation.metadata));
1922                 if (allocation.mixed)
1923                         printf("  Data+Metadata:    %-8s %16s\n",
1924                                 btrfs_group_profile_str(data_profile),
1925                                 pretty_size(allocation.mixed));
1926                 printf("  System:           %-8s %16s\n",
1927                         btrfs_group_profile_str(metadata_profile),
1928                         pretty_size(allocation.system));
1929                 printf("SSD detected:       %s\n", ssd ? "yes" : "no");
1930                 btrfs_parse_features_to_string(features_buf, features);
1931                 printf("Incompat features:  %s", features_buf);
1932                 printf("\n");
1933
1934                 list_all_devices(root);
1935         }
1936
1937         /*
1938          * The filesystem is now fully set up, commit the remaining changes and
1939          * fix the signature as the last step before closing the devices.
1940          */
1941         fs_info->finalize_on_close = 1;
1942 out:
1943         ret = close_ctree(root);
1944
1945         if (!ret) {
1946                 optind = saved_optind;
1947                 dev_cnt = argc - optind;
1948                 while (dev_cnt-- > 0) {
1949                         file = argv[optind++];
1950                         if (is_block_device(file) == 1)
1951                                 btrfs_register_one_device(file);
1952                 }
1953         }
1954
1955         btrfs_close_all_devices();
1956         free(label);
1957
1958         return !!ret;
1959 }