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