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
  46 static u64 index_cnt = 2;
  47 static int verbose = 1;
  48
  49 struct directory_name_entry {
  50         char *dir_name;
  51         char *path;
  52         ino_t inum;
  53         struct list_head list;
  54 };
  55
  56 struct mkfs_allocation {
  57         u64 data;
  58         u64 metadata;
  59         u64 mixed;
  60         u64 system;
  61 };
  62
  63 static int create_metadata_block_groups(struct btrfs_root *root, int mixed,
  64                                 struct mkfs_allocation *allocation)
  65 {
  66         struct btrfs_trans_handle *trans;
  67         u64 bytes_used;
  68         u64 chunk_start = 0;
  69         u64 chunk_size = 0;
  70         int ret;
  71
  72         trans = btrfs_start_transaction(root, 1);
  73         bytes_used = btrfs_super_bytes_used(root->fs_info->super_copy);
  74
  75         root->fs_info->system_allocs = 1;
  76         ret = btrfs_make_block_group(trans, root, bytes_used,
  77                                      BTRFS_BLOCK_GROUP_SYSTEM,
  78                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
  79                                      0, BTRFS_MKFS_SYSTEM_GROUP_SIZE);
  80         allocation->system += BTRFS_MKFS_SYSTEM_GROUP_SIZE;
  81         BUG_ON(ret);
  82
  83         if (mixed) {
  84                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
  85                                         &chunk_start, &chunk_size,
  86                                         BTRFS_BLOCK_GROUP_METADATA |
  87                                         BTRFS_BLOCK_GROUP_DATA);
  88                 if (ret == -ENOSPC) {
  89                         fprintf(stderr,
  90                                 "no space to alloc data/metadata chunk\n");
  91                         goto err;
  92                 }
  93                 BUG_ON(ret);
  94                 ret = btrfs_make_block_group(trans, root, 0,
  95                                              BTRFS_BLOCK_GROUP_METADATA |
  96                                              BTRFS_BLOCK_GROUP_DATA,
  97                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
  98                                              chunk_start, chunk_size);
  99                 BUG_ON(ret);
 100                 allocation->mixed += chunk_size;
 101         } else {
 102                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 103                                         &chunk_start, &chunk_size,
 104                                         BTRFS_BLOCK_GROUP_METADATA);
 105                 if (ret == -ENOSPC) {
 106                         fprintf(stderr, "no space to alloc metadata chunk\n");
 107                         goto err;
 108                 }
 109                 BUG_ON(ret);
 110                 ret = btrfs_make_block_group(trans, root, 0,
 111                                              BTRFS_BLOCK_GROUP_METADATA,
 112                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 113                                              chunk_start, chunk_size);
 114                 allocation->metadata += chunk_size;
 115                 BUG_ON(ret);
 116         }
 117
 118         root->fs_info->system_allocs = 0;
 119         btrfs_commit_transaction(trans, root);
 120
 121 err:
 122         return ret;
 123 }
 124
 125 static int create_data_block_groups(struct btrfs_trans_handle *trans,
 126                 struct btrfs_root *root, int mixed,
 127                 struct mkfs_allocation *allocation)
 128 {
 129         u64 chunk_start = 0;
 130         u64 chunk_size = 0;
 131         int ret = 0;
 132
 133         if (!mixed) {
 134                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 135                                         &chunk_start, &chunk_size,
 136                                         BTRFS_BLOCK_GROUP_DATA);
 137                 if (ret == -ENOSPC) {
 138                         fprintf(stderr, "no space to alloc data chunk\n");
 139                         goto err;
 140                 }
 141                 BUG_ON(ret);
 142                 ret = btrfs_make_block_group(trans, root, 0,
 143                                              BTRFS_BLOCK_GROUP_DATA,
 144                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 145                                              chunk_start, chunk_size);
 146                 allocation->data += chunk_size;
 147                 BUG_ON(ret);
 148         }
 149
 150 err:
 151         return ret;
 152 }
 153
 154 static int make_root_dir(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 155                 int mixed, struct mkfs_allocation *allocation)
 156 {
 157         struct btrfs_key location;
 158         int ret;
 159
 160         ret = btrfs_make_root_dir(trans, root->fs_info->tree_root,
 161                               BTRFS_ROOT_TREE_DIR_OBJECTID);
 162         if (ret)
 163                 goto err;
 164         ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
 165         if (ret)
 166                 goto err;
 167         memcpy(&location, &root->fs_info->fs_root->root_key, sizeof(location));
 168         location.offset = (u64)-1;
 169         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
 170                         "default", 7,
 171                         btrfs_super_root_dir(root->fs_info->super_copy),
 172                         &location, BTRFS_FT_DIR, 0);
 173         if (ret)
 174                 goto err;
 175
 176         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 177                              "default", 7, location.objectid,
 178                              BTRFS_ROOT_TREE_DIR_OBJECTID, 0);
 179         if (ret)
 180                 goto err;
 181
 182 err:
 183         return ret;
 184 }
 185
 186 static void __recow_root(struct btrfs_trans_handle *trans,
 187                          struct btrfs_root *root)
 188 {
 189         int ret;
 190         struct extent_buffer *tmp;
 191
 192         if (trans->transid != btrfs_root_generation(&root->root_item)) {
 193                 extent_buffer_get(root->node);
 194                 ret = __btrfs_cow_block(trans, root, root->node,
 195                                         NULL, 0, &tmp, 0, 0);
 196                 BUG_ON(ret);
 197                 free_extent_buffer(tmp);
 198         }
 199 }
 200
 201 static void recow_roots(struct btrfs_trans_handle *trans,
 202                        struct btrfs_root *root)
 203 {
 204         struct btrfs_fs_info *info = root->fs_info;
 205
 206         __recow_root(trans, info->fs_root);
 207         __recow_root(trans, info->tree_root);
 208         __recow_root(trans, info->extent_root);
 209         __recow_root(trans, info->chunk_root);
 210         __recow_root(trans, info->dev_root);
 211         __recow_root(trans, info->csum_root);
 212 }
 213
 214 static int create_one_raid_group(struct btrfs_trans_handle *trans,
 215                               struct btrfs_root *root, u64 type,
 216                               struct mkfs_allocation *allocation)
 217
 218 {
 219         u64 chunk_start;
 220         u64 chunk_size;
 221         int ret;
 222
 223         ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 224                                 &chunk_start, &chunk_size, type);
 225         if (ret == -ENOSPC) {
 226                 fprintf(stderr, "not enough free space\n");
 227                 exit(1);
 228         }
 229         BUG_ON(ret);
 230         ret = btrfs_make_block_group(trans, root->fs_info->extent_root, 0,
 231                                      type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 232                                      chunk_start, chunk_size);
 233         if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == BTRFS_BLOCK_GROUP_DATA)
 234                 allocation->data += chunk_size;
 235         else if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == BTRFS_BLOCK_GROUP_METADATA)
 236                 allocation->metadata += chunk_size;
 237         else if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == BTRFS_BLOCK_GROUP_SYSTEM)
 238                 allocation->system += chunk_size;
 239         else if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
 240                         (BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA))
 241                 allocation->mixed += chunk_size;
 242         else
 243                 BUG_ON(1);
 244
 245         BUG_ON(ret);
 246         return ret;
 247 }
 248
 249 static int create_raid_groups(struct btrfs_trans_handle *trans,
 250                               struct btrfs_root *root, u64 data_profile,
 251                               u64 metadata_profile, int mixed,
 252                               struct mkfs_allocation *allocation)
 253 {
 254         u64 num_devices = btrfs_super_num_devices(root->fs_info->super_copy);
 255         int ret;
 256
 257         if (metadata_profile) {
 258                 u64 meta_flags = BTRFS_BLOCK_GROUP_METADATA;
 259
 260                 ret = create_one_raid_group(trans, root,
 261                                             BTRFS_BLOCK_GROUP_SYSTEM |
 262                                             metadata_profile, allocation);
 263                 BUG_ON(ret);
 264
 265                 if (mixed)
 266                         meta_flags |= BTRFS_BLOCK_GROUP_DATA;
 267
 268                 ret = create_one_raid_group(trans, root, meta_flags |
 269                                             metadata_profile, allocation);
 270                 BUG_ON(ret);
 271
 272         }
 273         if (!mixed && num_devices > 1 && data_profile) {
 274                 ret = create_one_raid_group(trans, root,
 275                                             BTRFS_BLOCK_GROUP_DATA |
 276                                             data_profile, allocation);
 277                 BUG_ON(ret);
 278         }
 279         recow_roots(trans, root);
 280
 281         return 0;
 282 }
 283
 284 static int create_data_reloc_tree(struct btrfs_trans_handle *trans,
 285                                   struct btrfs_root *root)
 286 {
 287         struct btrfs_key location;
 288         struct btrfs_root_item root_item;
 289         struct extent_buffer *tmp;
 290         u64 objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
 291         int ret;
 292
 293         ret = btrfs_copy_root(trans, root, root->node, &tmp, objectid);
 294         BUG_ON(ret);
 295
 296         memcpy(&root_item, &root->root_item, sizeof(root_item));
 297         btrfs_set_root_bytenr(&root_item, tmp->start);
 298         btrfs_set_root_level(&root_item, btrfs_header_level(tmp));
 299         btrfs_set_root_generation(&root_item, trans->transid);
 300         free_extent_buffer(tmp);
 301
 302         location.objectid = objectid;
 303         location.type = BTRFS_ROOT_ITEM_KEY;
 304         location.offset = 0;
 305         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
 306                                 &location, &root_item);
 307         BUG_ON(ret);
 308         return 0;
 309 }
 310
 311 static void print_usage(int ret)
 312 {
 313         fprintf(stderr, "usage: mkfs.btrfs [options] dev [ dev ... ]\n");
 314         fprintf(stderr, "options:\n");
 315         fprintf(stderr, "\t-A|--alloc-start START  the offset to start the FS\n");
 316         fprintf(stderr, "\t-b|--byte-count SIZE    total number of bytes in the FS\n");
 317         fprintf(stderr, "\t-d|--data PROFILE       data profile, raid0, raid1, raid5, raid6, raid10, dup or single\n");
 318         fprintf(stderr, "\t-f|--force              force overwrite of existing filesystem\n");
 319         fprintf(stderr, "\t-l|--leafsize SIZE      deprecated, alias for nodesize\n");
 320         fprintf(stderr, "\t-L|--label LABEL        set a label\n");
 321         fprintf(stderr, "\t-m|--metadata PROFILE   metadata profile, values like data profile\n");
 322         fprintf(stderr, "\t-M|--mixed              mix metadata and data together\n");
 323         fprintf(stderr, "\t-n|--nodesize SIZE      size of btree nodes\n");
 324         fprintf(stderr, "\t-s|--sectorsize SIZE    min block allocation (may not mountable by current kernel)\n");
 325         fprintf(stderr, "\t-r|--rootdir DIR        the source directory\n");
 326         fprintf(stderr, "\t-K|--nodiscard          do not perform whole device TRIM\n");
 327         fprintf(stderr, "\t-O|--features LIST      comma separated list of filesystem features, use '-O list-all' to list features\n");
 328         fprintf(stderr, "\t-U|--uuid UUID          specify the filesystem UUID\n");
 329         fprintf(stderr, "\t-q|--quiet              no messages except errors\n");
 330         fprintf(stderr, "\t-V|--version            print the mkfs.btrfs version and exit\n");
 331         fprintf(stderr, "%s\n", PACKAGE_STRING);
 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 void list_all_devices(struct btrfs_root *root)
1151 {
1152         struct btrfs_fs_devices *fs_devices;
1153         struct btrfs_device *device;
1154         int number_of_devices = 0;
1155         u64 total_block_count = 0;
1156
1157         fs_devices = root->fs_info->fs_devices;
1158
1159         list_for_each_entry(device, &fs_devices->devices, dev_list)
1160                 number_of_devices++;
1161
1162         printf("Number of devices:  %d\n", number_of_devices);
1163         /* printf("Total devices size: %10s\n", */
1164                 /* pretty_size(total_block_count)); */
1165         printf("Devices:\n");
1166         printf("   ID        SIZE  PATH\n");
1167         list_for_each_entry_reverse(device, &fs_devices->devices, dev_list) {
1168                 char dev_uuid[BTRFS_UUID_UNPARSED_SIZE];
1169
1170                 uuid_unparse(device->uuid, dev_uuid);
1171                 printf("  %3llu  %10s  %s\n",
1172                         device->devid,
1173                         pretty_size(device->total_bytes),
1174                         device->name);
1175                 total_block_count += device->total_bytes;
1176         }
1177
1178         printf("\n");
1179 }
1180
1181 static int is_temp_block_group(struct extent_buffer *node,
1182                                struct btrfs_block_group_item *bgi,
1183                                u64 data_profile, u64 meta_profile,
1184                                u64 sys_profile)
1185 {
1186         u64 flag = btrfs_disk_block_group_flags(node, bgi);
1187         u64 flag_type = flag & BTRFS_BLOCK_GROUP_TYPE_MASK;
1188         u64 flag_profile = flag & BTRFS_BLOCK_GROUP_PROFILE_MASK;
1189         u64 used = btrfs_disk_block_group_used(node, bgi);
1190
1191         /*
1192          * Chunks meets all the following conditions is a temp chunk
1193          * 1) Empty chunk
1194          * Temp chunk is always empty.
1195          *
1196          * 2) profile dismatch with mkfs profile.
1197          * Temp chunk is always in SINGLE
1198          *
1199          * 3) Size differs with mkfs_alloc
1200          * Special case for SINGLE/SINGLE btrfs.
1201          * In that case, temp data chunk and real data chunk are always empty.
1202          * So we need to use mkfs_alloc to be sure which chunk is the newly
1203          * allocated.
1204          *
1205          * Normally, new chunk size is equal to mkfs one (One chunk)
1206          * If it has multiple chunks, we just refuse to delete any one.
1207          * As they are all single, so no real problem will happen.
1208          * So only use condition 1) and 2) to judge them.
1209          */
1210         if (used != 0)
1211                 return 0;
1212         switch (flag_type) {
1213         case BTRFS_BLOCK_GROUP_DATA:
1214         case BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA:
1215                 data_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1216                 if (flag_profile != data_profile)
1217                         return 1;
1218                 break;
1219         case BTRFS_BLOCK_GROUP_METADATA:
1220                 meta_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1221                 if (flag_profile != meta_profile)
1222                         return 1;
1223                 break;
1224         case BTRFS_BLOCK_GROUP_SYSTEM:
1225                 sys_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1226                 if (flag_profile != sys_profile)
1227                         return 1;
1228                 break;
1229         }
1230         return 0;
1231 }
1232
1233 /* Note: if current is a block group, it will skip it anyway */
1234 static int next_block_group(struct btrfs_root *root,
1235                             struct btrfs_path *path)
1236 {
1237         struct btrfs_key key;
1238         int ret = 0;
1239
1240         while (1) {
1241                 ret = btrfs_next_item(root, path);
1242                 if (ret)
1243                         goto out;
1244
1245                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1246                 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
1247                         goto out;
1248         }
1249 out:
1250         return ret;
1251 }
1252
1253 /* This function will cleanup  */
1254 static int cleanup_temp_chunks(struct btrfs_fs_info *fs_info,
1255                                struct mkfs_allocation *alloc,
1256                                u64 data_profile, u64 meta_profile,
1257                                u64 sys_profile)
1258 {
1259         struct btrfs_trans_handle *trans = NULL;
1260         struct btrfs_block_group_item *bgi;
1261         struct btrfs_root *root = fs_info->extent_root;
1262         struct btrfs_key key;
1263         struct btrfs_key found_key;
1264         struct btrfs_path *path;
1265         int ret = 0;
1266
1267         path = btrfs_alloc_path();
1268         if (!path) {
1269                 ret = -ENOMEM;
1270                 goto out;
1271         }
1272
1273         trans = btrfs_start_transaction(root, 1);
1274
1275         key.objectid = 0;
1276         key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
1277         key.offset = 0;
1278
1279         while (1) {
1280                 /*
1281                  * as the rest of the loop may modify the tree, we need to
1282                  * start a new search each time.
1283                  */
1284                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
1285                 if (ret < 0)
1286                         goto out;
1287
1288                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1289                                       path->slots[0]);
1290                 if (found_key.objectid < key.objectid)
1291                         goto out;
1292                 if (found_key.type != BTRFS_BLOCK_GROUP_ITEM_KEY) {
1293                         ret = next_block_group(root, path);
1294                         if (ret < 0)
1295                                 goto out;
1296                         if (ret > 0) {
1297                                 ret = 0;
1298                                 goto out;
1299                         }
1300                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1301                                               path->slots[0]);
1302                 }
1303
1304                 bgi = btrfs_item_ptr(path->nodes[0], path->slots[0],
1305                                      struct btrfs_block_group_item);
1306                 if (is_temp_block_group(path->nodes[0], bgi,
1307                                         data_profile, meta_profile,
1308                                         sys_profile)) {
1309                         ret = btrfs_free_block_group(trans, fs_info,
1310                                         found_key.objectid, found_key.offset);
1311                         if (ret < 0)
1312                                 goto out;
1313                 }
1314                 btrfs_release_path(path);
1315                 key.objectid = found_key.objectid + found_key.offset;
1316         }
1317 out:
1318         if (trans)
1319                 btrfs_commit_transaction(trans, root);
1320         btrfs_free_path(path);
1321         return ret;
1322 }
1323
1324 int main(int ac, char **av)
1325 {
1326         char *file;
1327         struct btrfs_root *root;
1328         struct btrfs_trans_handle *trans;
1329         char *label = NULL;
1330         u64 block_count = 0;
1331         u64 dev_block_count = 0;
1332         u64 blocks[7];
1333         u64 alloc_start = 0;
1334         u64 metadata_profile = 0;
1335         u64 data_profile = 0;
1336         u32 nodesize = max_t(u32, sysconf(_SC_PAGESIZE),
1337                         BTRFS_MKFS_DEFAULT_NODE_SIZE);
1338         u32 sectorsize = 4096;
1339         u32 stripesize = 4096;
1340         int zero_end = 1;
1341         int fd;
1342         int ret;
1343         int i;
1344         int mixed = 0;
1345         int nodesize_forced = 0;
1346         int data_profile_opt = 0;
1347         int metadata_profile_opt = 0;
1348         int discard = 1;
1349         int ssd = 0;
1350         int force_overwrite = 0;
1351         char *source_dir = NULL;
1352         int source_dir_set = 0;
1353         u64 num_of_meta_chunks = 0;
1354         u64 size_of_data = 0;
1355         u64 source_dir_size = 0;
1356         int dev_cnt = 0;
1357         int saved_optind;
1358         char fs_uuid[BTRFS_UUID_UNPARSED_SIZE] = { 0 };
1359         u64 features = BTRFS_MKFS_DEFAULT_FEATURES;
1360         struct mkfs_allocation allocation = { 0 };
1361         struct btrfs_mkfs_config mkfs_cfg;
1362
1363         while(1) {
1364                 int c;
1365                 static const struct option long_options[] = {
1366                         { "alloc-start", required_argument, NULL, 'A'},
1367                         { "byte-count", required_argument, NULL, 'b' },
1368                         { "force", no_argument, NULL, 'f' },
1369                         { "leafsize", required_argument, NULL, 'l' },
1370                         { "label", required_argument, NULL, 'L'},
1371                         { "metadata", required_argument, NULL, 'm' },
1372                         { "mixed", no_argument, NULL, 'M' },
1373                         { "nodesize", required_argument, NULL, 'n' },
1374                         { "sectorsize", required_argument, NULL, 's' },
1375                         { "data", required_argument, NULL, 'd' },
1376                         { "version", no_argument, NULL, 'V' },
1377                         { "rootdir", required_argument, NULL, 'r' },
1378                         { "nodiscard", no_argument, NULL, 'K' },
1379                         { "features", required_argument, NULL, 'O' },
1380                         { "uuid", required_argument, NULL, 'U' },
1381                         { "quiet", 0, NULL, 'q' },
1382                         { "help", no_argument, NULL, GETOPT_VAL_HELP },
1383                         { NULL, 0, NULL, 0}
1384                 };
1385
1386                 c = getopt_long(ac, av, "A:b:fl:n:s:m:d:L:O:r:U:VMKq",
1387                                 long_options, NULL);
1388                 if (c < 0)
1389                         break;
1390                 switch(c) {
1391                         case 'A':
1392                                 alloc_start = parse_size(optarg);
1393                                 break;
1394                         case 'f':
1395                                 force_overwrite = 1;
1396                                 break;
1397                         case 'd':
1398                                 data_profile = parse_profile(optarg);
1399                                 data_profile_opt = 1;
1400                                 break;
1401                         case 'l':
1402                                 fprintf(stderr,
1403                         "WARNING: --leafsize is deprecated, use --nodesize\n");
1404                         case 'n':
1405                                 nodesize = parse_size(optarg);
1406                                 nodesize_forced = 1;
1407                                 break;
1408                         case 'L':
1409                                 label = parse_label(optarg);
1410                                 break;
1411                         case 'm':
1412                                 metadata_profile = parse_profile(optarg);
1413                                 metadata_profile_opt = 1;
1414                                 break;
1415                         case 'M':
1416                                 mixed = 1;
1417                                 break;
1418                         case 'O': {
1419                                 char *orig = strdup(optarg);
1420                                 char *tmp = orig;
1421
1422                                 tmp = btrfs_parse_fs_features(tmp, &features);
1423                                 if (tmp) {
1424                                         fprintf(stderr,
1425                                                 "Unrecognized filesystem feature '%s'\n",
1426                                                         tmp);
1427                                         free(orig);
1428                                         exit(1);
1429                                 }
1430                                 free(orig);
1431                                 if (features & BTRFS_FEATURE_LIST_ALL) {
1432                                         btrfs_list_all_fs_features(0);
1433                                         exit(0);
1434                                 }
1435                                 break;
1436                                 }
1437                         case 's':
1438                                 sectorsize = parse_size(optarg);
1439                                 break;
1440                         case 'b':
1441                                 block_count = parse_size(optarg);
1442                                 if (block_count <= BTRFS_MKFS_SMALL_VOLUME_SIZE)
1443                                         mixed = 1;
1444                                 zero_end = 0;
1445                                 break;
1446                         case 'V':
1447                                 print_version();
1448                                 break;
1449                         case 'r':
1450                                 source_dir = optarg;
1451                                 source_dir_set = 1;
1452                                 break;
1453                         case 'U':
1454                                 strncpy(fs_uuid, optarg,
1455                                         BTRFS_UUID_UNPARSED_SIZE - 1);
1456                                 break;
1457                         case 'K':
1458                                 discard = 0;
1459                                 break;
1460                         case 'q':
1461                                 verbose = 0;
1462                                 break;
1463                         case GETOPT_VAL_HELP:
1464                         default:
1465                                 print_usage(c != GETOPT_VAL_HELP);
1466                 }
1467         }
1468
1469         sectorsize = max(sectorsize, (u32)sysconf(_SC_PAGESIZE));
1470         saved_optind = optind;
1471         dev_cnt = ac - optind;
1472         if (dev_cnt == 0)
1473                 print_usage(1);
1474
1475         if (source_dir_set && dev_cnt > 1) {
1476                 fprintf(stderr,
1477                         "The -r option is limited to a single device\n");
1478                 exit(1);
1479         }
1480
1481         if (*fs_uuid) {
1482                 uuid_t dummy_uuid;
1483
1484                 if (uuid_parse(fs_uuid, dummy_uuid) != 0) {
1485                         fprintf(stderr, "could not parse UUID: %s\n", fs_uuid);
1486                         exit(1);
1487                 }
1488                 if (!test_uuid_unique(fs_uuid)) {
1489                         fprintf(stderr, "non-unique UUID: %s\n", fs_uuid);
1490                         exit(1);
1491                 }
1492         }
1493
1494         while (dev_cnt-- > 0) {
1495                 file = av[optind++];
1496                 if (is_block_device(file) == 1)
1497                         if (test_dev_for_mkfs(file, force_overwrite))
1498                                 exit(1);
1499         }
1500
1501         optind = saved_optind;
1502         dev_cnt = ac - optind;
1503
1504         file = av[optind++];
1505         ssd = is_ssd(file);
1506
1507         if (is_vol_small(file) || mixed) {
1508                 if (verbose)
1509                         printf("SMALL VOLUME: forcing mixed metadata/data groups\n");
1510                 mixed = 1;
1511         }
1512
1513         /*
1514         * Set default profiles according to number of added devices.
1515         * For mixed groups defaults are single/single.
1516         */
1517         if (!mixed) {
1518                 if (!metadata_profile_opt) {
1519                         if (dev_cnt == 1 && ssd && verbose)
1520                                 printf("Detected a SSD, turning off metadata "
1521                                 "duplication.  Mkfs with -m dup if you want to "
1522                                 "force metadata duplication.\n");
1523
1524                         metadata_profile = (dev_cnt > 1) ?
1525                                         BTRFS_BLOCK_GROUP_RAID1 : (ssd) ?
1526                                         0: BTRFS_BLOCK_GROUP_DUP;
1527                 }
1528                 if (!data_profile_opt) {
1529                         data_profile = (dev_cnt > 1) ?
1530                                 BTRFS_BLOCK_GROUP_RAID0 : 0; /* raid0 or single */
1531                 }
1532         } else {
1533                 u32 best_nodesize = max_t(u32, sysconf(_SC_PAGESIZE), sectorsize);
1534
1535                 if (metadata_profile_opt || data_profile_opt) {
1536                         if (metadata_profile != data_profile) {
1537                                 fprintf(stderr,
1538         "ERROR: With mixed block groups data and metadata profiles must be the same\n");
1539                                 exit(1);
1540                         }
1541                 }
1542
1543                 if (!nodesize_forced)
1544                         nodesize = best_nodesize;
1545         }
1546         if (btrfs_check_nodesize(nodesize, sectorsize,
1547                                  features))
1548                 exit(1);
1549
1550         /* Check device/block_count after the nodesize is determined */
1551         if (block_count && block_count < btrfs_min_dev_size(nodesize)) {
1552                 fprintf(stderr,
1553                         "Size '%llu' is too small to make a usable filesystem\n",
1554                         block_count);
1555                 fprintf(stderr,
1556                         "Minimum size for btrfs filesystem is %llu\n",
1557                         btrfs_min_dev_size(nodesize));
1558                 exit(1);
1559         }
1560         for (i = saved_optind; i < saved_optind + dev_cnt; i++) {
1561                 char *path;
1562
1563                 path = av[i];
1564                 ret = test_minimum_size(path, nodesize);
1565                 if (ret < 0) {
1566                         fprintf(stderr, "Failed to check size for '%s': %s\n",
1567                                 path, strerror(-ret));
1568                         exit (1);
1569                 }
1570                 if (ret > 0) {
1571                         fprintf(stderr,
1572                                 "'%s' is too small to make a usable filesystem\n",
1573                                 path);
1574                         fprintf(stderr,
1575                                 "Minimum size for each btrfs device is %llu.\n",
1576                                 btrfs_min_dev_size(nodesize));
1577                         exit(1);
1578                 }
1579         }
1580         ret = test_num_disk_vs_raid(metadata_profile, data_profile,
1581                         dev_cnt, mixed);
1582         if (ret)
1583                 exit(1);
1584
1585         /* if we are here that means all devs are good to btrfsify */
1586         if (verbose) {
1587                 printf("%s\n", PACKAGE_STRING);
1588                 printf("See %s for more information.\n\n", PACKAGE_URL);
1589         }
1590
1591         dev_cnt--;
1592
1593         if (!source_dir_set) {
1594                 /*
1595                  * open without O_EXCL so that the problem should not
1596                  * occur by the following processing.
1597                  * (btrfs_register_one_device() fails if O_EXCL is on)
1598                  */
1599                 fd = open(file, O_RDWR);
1600                 if (fd < 0) {
1601                         fprintf(stderr, "unable to open %s: %s\n", file,
1602                                 strerror(errno));
1603                         exit(1);
1604                 }
1605                 ret = btrfs_prepare_device(fd, file, zero_end, &dev_block_count,
1606                                            block_count, &mixed, discard);
1607                 if (ret) {
1608                         close(fd);
1609                         exit(1);
1610                 }
1611                 if (block_count && block_count > dev_block_count) {
1612                         fprintf(stderr, "%s is smaller than requested size\n", file);
1613                         exit(1);
1614                 }
1615         } else {
1616                 fd = open_target(file);
1617                 if (fd < 0) {
1618                         fprintf(stderr, "unable to open the %s\n", file);
1619                         exit(1);
1620                 }
1621
1622                 source_dir_size = size_sourcedir(source_dir, sectorsize,
1623                                              &num_of_meta_chunks, &size_of_data);
1624                 if(block_count < source_dir_size)
1625                         block_count = source_dir_size;
1626                 ret = zero_output_file(fd, block_count, sectorsize);
1627                 if (ret) {
1628                         fprintf(stderr, "unable to zero the output file\n");
1629                         exit(1);
1630                 }
1631                 /* our "device" is the new image file */
1632                 dev_block_count = block_count;
1633         }
1634
1635         /* To create the first block group and chunk 0 in make_btrfs */
1636         if (dev_block_count < BTRFS_MKFS_SYSTEM_GROUP_SIZE) {
1637                 fprintf(stderr, "device is too small to make filesystem\n");
1638                 exit(1);
1639         }
1640
1641         blocks[0] = BTRFS_SUPER_INFO_OFFSET;
1642         for (i = 1; i < 7; i++) {
1643                 blocks[i] = BTRFS_SUPER_INFO_OFFSET + 1024 * 1024 +
1644                         nodesize * i;
1645         }
1646
1647         if (group_profile_max_safe_loss(metadata_profile) <
1648                 group_profile_max_safe_loss(data_profile)){
1649                 fprintf(stderr,
1650                         "WARNING: metatdata has lower redundancy than data!\n\n");
1651         }
1652
1653         /*
1654          * FS features that can be set by other means than -O
1655          * just set the bit here
1656          */
1657         if (mixed)
1658                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS;
1659
1660         if ((data_profile | metadata_profile) &
1661             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
1662                 features |= BTRFS_FEATURE_INCOMPAT_RAID56;
1663         }
1664
1665         mkfs_cfg.label = label;
1666         mkfs_cfg.fs_uuid = fs_uuid;
1667         memcpy(mkfs_cfg.blocks, blocks, sizeof(blocks));
1668         mkfs_cfg.num_bytes = dev_block_count;
1669         mkfs_cfg.nodesize = nodesize;
1670         mkfs_cfg.sectorsize = sectorsize;
1671         mkfs_cfg.stripesize = stripesize;
1672         mkfs_cfg.features = features;
1673
1674         ret = make_btrfs(fd, &mkfs_cfg);
1675         if (ret) {
1676                 fprintf(stderr, "error during mkfs: %s\n", strerror(-ret));
1677                 exit(1);
1678         }
1679
1680         root = open_ctree(file, 0, OPEN_CTREE_WRITES);
1681         if (!root) {
1682                 fprintf(stderr, "Open ctree failed\n");
1683                 close(fd);
1684                 exit(1);
1685         }
1686         root->fs_info->alloc_start = alloc_start;
1687
1688         ret = create_metadata_block_groups(root, mixed, &allocation);
1689         if (ret) {
1690                 fprintf(stderr, "failed to create default block groups\n");
1691                 exit(1);
1692         }
1693
1694         trans = btrfs_start_transaction(root, 1);
1695         if (!trans) {
1696                 fprintf(stderr, "failed to start transaction\n");
1697                 exit(1);
1698         }
1699
1700         ret = create_data_block_groups(trans, root, mixed, &allocation);
1701         if (ret) {
1702                 fprintf(stderr, "failed to create default data block groups\n");
1703                 exit(1);
1704         }
1705
1706         ret = make_root_dir(trans, root, mixed, &allocation);
1707         if (ret) {
1708                 fprintf(stderr, "failed to setup the root directory\n");
1709                 exit(1);
1710         }
1711
1712         btrfs_commit_transaction(trans, root);
1713
1714         trans = btrfs_start_transaction(root, 1);
1715         if (!trans) {
1716                 fprintf(stderr, "failed to start transaction\n");
1717                 exit(1);
1718         }
1719
1720         if (is_block_device(file) == 1)
1721                 btrfs_register_one_device(file);
1722
1723         if (dev_cnt == 0)
1724                 goto raid_groups;
1725
1726         while (dev_cnt-- > 0) {
1727                 int old_mixed = mixed;
1728
1729                 file = av[optind++];
1730
1731                 /*
1732                  * open without O_EXCL so that the problem should not
1733                  * occur by the following processing.
1734                  * (btrfs_register_one_device() fails if O_EXCL is on)
1735                  */
1736                 fd = open(file, O_RDWR);
1737                 if (fd < 0) {
1738                         fprintf(stderr, "unable to open %s: %s\n", file,
1739                                 strerror(errno));
1740                         exit(1);
1741                 }
1742                 ret = btrfs_device_already_in_root(root, fd,
1743                                                    BTRFS_SUPER_INFO_OFFSET);
1744                 if (ret) {
1745                         fprintf(stderr, "skipping duplicate device %s in FS\n",
1746                                 file);
1747                         close(fd);
1748                         continue;
1749                 }
1750                 ret = btrfs_prepare_device(fd, file, zero_end, &dev_block_count,
1751                                            block_count, &mixed, discard);
1752                 if (ret) {
1753                         close(fd);
1754                         exit(1);
1755                 }
1756                 mixed = old_mixed;
1757
1758                 ret = btrfs_add_to_fsid(trans, root, fd, file, dev_block_count,
1759                                         sectorsize, sectorsize, sectorsize);
1760                 BUG_ON(ret);
1761                 if (verbose >= 2) {
1762                         struct btrfs_device *device;
1763
1764                         device = container_of(root->fs_info->fs_devices->devices.next,
1765                                         struct btrfs_device, dev_list);
1766                         printf("adding device %s id %llu\n", file,
1767                                 (unsigned long long)device->devid);
1768                 }
1769
1770                 if (is_block_device(file) == 1)
1771                         btrfs_register_one_device(file);
1772         }
1773
1774 raid_groups:
1775         if (!source_dir_set) {
1776                 ret = create_raid_groups(trans, root, data_profile,
1777                                  metadata_profile, mixed, &allocation);
1778                 BUG_ON(ret);
1779         }
1780
1781         ret = create_data_reloc_tree(trans, root);
1782         BUG_ON(ret);
1783
1784         btrfs_commit_transaction(trans, root);
1785
1786         if (source_dir_set) {
1787                 trans = btrfs_start_transaction(root, 1);
1788                 ret = create_chunks(trans, root,
1789                                     num_of_meta_chunks, size_of_data,
1790                                     &allocation);
1791                 BUG_ON(ret);
1792                 btrfs_commit_transaction(trans, root);
1793
1794                 ret = make_image(source_dir, root, fd);
1795                 BUG_ON(ret);
1796         }
1797         ret = cleanup_temp_chunks(root->fs_info, &allocation, data_profile,
1798                                   metadata_profile, metadata_profile);
1799         if (ret < 0) {
1800                 fprintf(stderr, "Failed to cleanup temporary chunks\n");
1801                 goto out;
1802         }
1803
1804         if (verbose) {
1805                 char features_buf[64];
1806
1807                 printf("Label:              %s\n", label);
1808                 printf("UUID:               %s\n", fs_uuid);
1809                 printf("Node size:          %u\n", nodesize);
1810                 printf("Sector size:        %u\n", sectorsize);
1811                 printf("Filesystem size:    %s\n",
1812                         pretty_size(btrfs_super_total_bytes(root->fs_info->super_copy)));
1813                 printf("Block group profiles:\n");
1814                 if (allocation.data)
1815                         printf("  Data:             %-8s %16s\n",
1816                                 btrfs_group_profile_str(data_profile),
1817                                 pretty_size(allocation.data));
1818                 if (allocation.metadata)
1819                         printf("  Metadata:         %-8s %16s\n",
1820                                 btrfs_group_profile_str(metadata_profile),
1821                                 pretty_size(allocation.metadata));
1822                 if (allocation.mixed)
1823                         printf("  Data+Metadata:    %-8s %16s\n",
1824                                 btrfs_group_profile_str(data_profile),
1825                                 pretty_size(allocation.mixed));
1826                 printf("  System:           %-8s %16s\n",
1827                         btrfs_group_profile_str(metadata_profile),
1828                         pretty_size(allocation.system));
1829                 printf("SSD detected:       %s\n", ssd ? "yes" : "no");
1830                 btrfs_parse_features_to_string(features_buf, features);
1831                 printf("Incompat features:  %s", features_buf);
1832                 printf("\n");
1833
1834                 list_all_devices(root);
1835         }
1836
1837 out:
1838         ret = close_ctree(root);
1839         BUG_ON(ret);
1840         btrfs_close_all_devices();
1841         free(label);
1842         return 0;
1843 }