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