mkfs.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include "kerncompat.h"
  20 #include "androidcompat.h"
  21
  22 #include <sys/ioctl.h>
  23 #include <sys/mount.h>
  24 #include "ioctl.h"
  25 #include <stdio.h>
  26 #include <stdlib.h>
  27 #include <sys/types.h>
  28 #include <sys/stat.h>
  29 /* #include <sys/dir.h> included via androidcompat.h */
  30 #include <fcntl.h>
  31 #include <unistd.h>
  32 #include <getopt.h>
  33 #include <uuid/uuid.h>
  34 #include <ctype.h>
  35 #include <sys/xattr.h>
  36 #include <limits.h>
  37 #include <linux/limits.h>
  38 #include <blkid/blkid.h>
  39 #include <ftw.h>
  40 #include "ctree.h"
  41 #include "disk-io.h"
  42 #include "volumes.h"
  43 #include "transaction.h"
  44 #include "utils.h"
  45 #include "list_sort.h"
  46
  47 static u64 index_cnt = 2;
  48 static int verbose = 1;
  49
  50 struct directory_name_entry {
  51         char *dir_name;
  52         char *path;
  53         ino_t inum;
  54         struct list_head list;
  55 };
  56
  57 struct mkfs_allocation {
  58         u64 data;
  59         u64 metadata;
  60         u64 mixed;
  61         u64 system;
  62 };
  63
  64 static int create_metadata_block_groups(struct btrfs_root *root, int mixed,
  65                                 struct mkfs_allocation *allocation)
  66 {
  67         struct btrfs_trans_handle *trans;
  68         u64 bytes_used;
  69         u64 chunk_start = 0;
  70         u64 chunk_size = 0;
  71         int ret;
  72
  73         trans = btrfs_start_transaction(root, 1);
  74         bytes_used = btrfs_super_bytes_used(root->fs_info->super_copy);
  75
  76         root->fs_info->system_allocs = 1;
  77         ret = btrfs_make_block_group(trans, root, bytes_used,
  78                                      BTRFS_BLOCK_GROUP_SYSTEM,
  79                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
  80                                      0, BTRFS_MKFS_SYSTEM_GROUP_SIZE);
  81         allocation->system += BTRFS_MKFS_SYSTEM_GROUP_SIZE;
  82         BUG_ON(ret);
  83
  84         if (mixed) {
  85                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
  86                                         &chunk_start, &chunk_size,
  87                                         BTRFS_BLOCK_GROUP_METADATA |
  88                                         BTRFS_BLOCK_GROUP_DATA);
  89                 if (ret == -ENOSPC) {
  90                         fprintf(stderr,
  91                                 "no space to alloc data/metadata chunk\n");
  92                         goto err;
  93                 }
  94                 BUG_ON(ret);
  95                 ret = btrfs_make_block_group(trans, root, 0,
  96                                              BTRFS_BLOCK_GROUP_METADATA |
  97                                              BTRFS_BLOCK_GROUP_DATA,
  98                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
  99                                              chunk_start, chunk_size);
 100                 BUG_ON(ret);
 101                 allocation->mixed += chunk_size;
 102         } else {
 103                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 104                                         &chunk_start, &chunk_size,
 105                                         BTRFS_BLOCK_GROUP_METADATA);
 106                 if (ret == -ENOSPC) {
 107                         fprintf(stderr, "no space to alloc metadata chunk\n");
 108                         goto err;
 109                 }
 110                 BUG_ON(ret);
 111                 ret = btrfs_make_block_group(trans, root, 0,
 112                                              BTRFS_BLOCK_GROUP_METADATA,
 113                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 114                                              chunk_start, chunk_size);
 115                 allocation->metadata += chunk_size;
 116                 BUG_ON(ret);
 117         }
 118
 119         root->fs_info->system_allocs = 0;
 120         btrfs_commit_transaction(trans, root);
 121
 122 err:
 123         return ret;
 124 }
 125
 126 static int create_data_block_groups(struct btrfs_trans_handle *trans,
 127                 struct btrfs_root *root, int mixed,
 128                 struct mkfs_allocation *allocation)
 129 {
 130         u64 chunk_start = 0;
 131         u64 chunk_size = 0;
 132         int ret = 0;
 133
 134         if (!mixed) {
 135                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 136                                         &chunk_start, &chunk_size,
 137                                         BTRFS_BLOCK_GROUP_DATA);
 138                 if (ret == -ENOSPC) {
 139                         fprintf(stderr, "no space to alloc data chunk\n");
 140                         goto err;
 141                 }
 142                 BUG_ON(ret);
 143                 ret = btrfs_make_block_group(trans, root, 0,
 144                                              BTRFS_BLOCK_GROUP_DATA,
 145                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 146                                              chunk_start, chunk_size);
 147                 allocation->data += chunk_size;
 148                 BUG_ON(ret);
 149         }
 150
 151 err:
 152         return ret;
 153 }
 154
 155 static int make_root_dir(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 156                 struct mkfs_allocation *allocation)
 157 {
 158         struct btrfs_key location;
 159         int ret;
 160
 161         ret = btrfs_make_root_dir(trans, root->fs_info->tree_root,
 162                               BTRFS_ROOT_TREE_DIR_OBJECTID);
 163         if (ret)
 164                 goto err;
 165         ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
 166         if (ret)
 167                 goto err;
 168         memcpy(&location, &root->fs_info->fs_root->root_key, sizeof(location));
 169         location.offset = (u64)-1;
 170         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
 171                         "default", 7,
 172                         btrfs_super_root_dir(root->fs_info->super_copy),
 173                         &location, BTRFS_FT_DIR, 0);
 174         if (ret)
 175                 goto err;
 176
 177         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 178                              "default", 7, location.objectid,
 179                              BTRFS_ROOT_TREE_DIR_OBJECTID, 0);
 180         if (ret)
 181                 goto err;
 182
 183 err:
 184         return ret;
 185 }
 186
 187 static void __recow_root(struct btrfs_trans_handle *trans,
 188                          struct btrfs_root *root)
 189 {
 190         int ret;
 191         struct extent_buffer *tmp;
 192
 193         if (trans->transid != btrfs_root_generation(&root->root_item)) {
 194                 extent_buffer_get(root->node);
 195                 ret = __btrfs_cow_block(trans, root, root->node,
 196                                         NULL, 0, &tmp, 0, 0);
 197                 BUG_ON(ret);
 198                 free_extent_buffer(tmp);
 199         }
 200 }
 201
 202 static void recow_roots(struct btrfs_trans_handle *trans,
 203                        struct btrfs_root *root)
 204 {
 205         struct btrfs_fs_info *info = root->fs_info;
 206
 207         __recow_root(trans, info->fs_root);
 208         __recow_root(trans, info->tree_root);
 209         __recow_root(trans, info->extent_root);
 210         __recow_root(trans, info->chunk_root);
 211         __recow_root(trans, info->dev_root);
 212         __recow_root(trans, info->csum_root);
 213 }
 214
 215 static int create_one_raid_group(struct btrfs_trans_handle *trans,
 216                               struct btrfs_root *root, u64 type,
 217                               struct mkfs_allocation *allocation)
 218
 219 {
 220         u64 chunk_start;
 221         u64 chunk_size;
 222         int ret;
 223
 224         ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 225                                 &chunk_start, &chunk_size, type);
 226         if (ret == -ENOSPC) {
 227                 fprintf(stderr, "not enough free space\n");
 228                 exit(1);
 229         }
 230         BUG_ON(ret);
 231         ret = btrfs_make_block_group(trans, root->fs_info->extent_root, 0,
 232                                      type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 233                                      chunk_start, chunk_size);
 234         if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == BTRFS_BLOCK_GROUP_DATA)
 235                 allocation->data += chunk_size;
 236         else if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == BTRFS_BLOCK_GROUP_METADATA)
 237                 allocation->metadata += chunk_size;
 238         else if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == BTRFS_BLOCK_GROUP_SYSTEM)
 239                 allocation->system += chunk_size;
 240         else if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
 241                         (BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA))
 242                 allocation->mixed += chunk_size;
 243         else
 244                 BUG_ON(1);
 245
 246         BUG_ON(ret);
 247         return ret;
 248 }
 249
 250 static int create_raid_groups(struct btrfs_trans_handle *trans,
 251                               struct btrfs_root *root, u64 data_profile,
 252                               u64 metadata_profile, int mixed,
 253                               struct mkfs_allocation *allocation)
 254 {
 255         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 && 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         char buf[PATH_MAX];
 595
 596         ret = readlink(path_name, buf, sizeof(buf));
 597         if (ret <= 0) {
 598                 fprintf(stderr, "readlink failed for %s\n", path_name);
 599                 goto fail;
 600         }
 601         if (ret >= sizeof(buf)) {
 602                 fprintf(stderr, "symlink too long for %s\n", path_name);
 603                 ret = -1;
 604                 goto fail;
 605         }
 606
 607         buf[ret] = '\0'; /* readlink does not do it for us */
 608         ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
 609                                          buf, ret + 1);
 610 fail:
 611         return ret;
 612 }
 613
 614 static int add_file_items(struct btrfs_trans_handle *trans,
 615                           struct btrfs_root *root,
 616                           struct btrfs_inode_item *btrfs_inode, u64 objectid,
 617                           ino_t parent_inum, struct stat *st,
 618                           const char *path_name, int out_fd)
 619 {
 620         int ret = -1;
 621         ssize_t ret_read;
 622         u64 bytes_read = 0;
 623         struct btrfs_key key;
 624         int blocks;
 625         u32 sectorsize = root->sectorsize;
 626         u64 first_block = 0;
 627         u64 file_pos = 0;
 628         u64 cur_bytes;
 629         u64 total_bytes;
 630         struct extent_buffer *eb = NULL;
 631         int fd;
 632
 633         if (st->st_size == 0)
 634                 return 0;
 635
 636         fd = open(path_name, O_RDONLY);
 637         if (fd == -1) {
 638                 fprintf(stderr, "%s open failed\n", path_name);
 639                 return ret;
 640         }
 641
 642         blocks = st->st_size / sectorsize;
 643         if (st->st_size % sectorsize)
 644                 blocks += 1;
 645
 646         if (st->st_size <= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 647                 char *buffer = malloc(st->st_size);
 648
 649                 if (!buffer) {
 650                         ret = -ENOMEM;
 651                         goto end;
 652                 }
 653
 654                 ret_read = pread64(fd, buffer, st->st_size, bytes_read);
 655                 if (ret_read == -1) {
 656                         fprintf(stderr, "%s read failed\n", path_name);
 657                         free(buffer);
 658                         goto end;
 659                 }
 660
 661                 ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
 662                                                  buffer, st->st_size);
 663                 free(buffer);
 664                 goto end;
 665         }
 666
 667         /* round up our st_size to the FS blocksize */
 668         total_bytes = (u64)blocks * sectorsize;
 669
 670         /*
 671          * do our IO in extent buffers so it can work
 672          * against any raid type
 673          */
 674         eb = calloc(1, sizeof(*eb) + sectorsize);
 675         if (!eb) {
 676                 ret = -ENOMEM;
 677                 goto end;
 678         }
 679
 680 again:
 681
 682         /*
 683          * keep our extent size at 1MB max, this makes it easier to work inside
 684          * the tiny block groups created during mkfs
 685          */
 686         cur_bytes = min(total_bytes, 1024ULL * 1024);
 687         ret = btrfs_reserve_extent(trans, root, cur_bytes, 0, 0, (u64)-1,
 688                                    &key, 1);
 689         if (ret)
 690                 goto end;
 691
 692         first_block = key.objectid;
 693         bytes_read = 0;
 694
 695         while (bytes_read < cur_bytes) {
 696
 697                 memset(eb->data, 0, sectorsize);
 698
 699                 ret_read = pread64(fd, eb->data, sectorsize, file_pos + bytes_read);
 700                 if (ret_read == -1) {
 701                         fprintf(stderr, "%s read failed\n", path_name);
 702                         goto end;
 703                 }
 704
 705                 eb->start = first_block + bytes_read;
 706                 eb->len = sectorsize;
 707
 708                 /*
 709                  * we're doing the csum before we record the extent, but
 710                  * that's ok
 711                  */
 712                 ret = btrfs_csum_file_block(trans, root->fs_info->csum_root,
 713                                             first_block + bytes_read + sectorsize,
 714                                             first_block + bytes_read,
 715                                             eb->data, sectorsize);
 716                 if (ret)
 717                         goto end;
 718
 719                 ret = write_and_map_eb(trans, root, eb);
 720                 if (ret) {
 721                         fprintf(stderr, "output file write failed\n");
 722                         goto end;
 723                 }
 724
 725                 bytes_read += sectorsize;
 726         }
 727
 728         if (bytes_read) {
 729                 ret = btrfs_record_file_extent(trans, root, objectid, btrfs_inode,
 730                                                file_pos, first_block, cur_bytes);
 731                 if (ret)
 732                         goto end;
 733
 734         }
 735
 736         file_pos += cur_bytes;
 737         total_bytes -= cur_bytes;
 738
 739         if (total_bytes)
 740                 goto again;
 741
 742 end:
 743         free(eb);
 744         close(fd);
 745         return ret;
 746 }
 747
 748 static char *make_path(char *dir, char *name)
 749 {
 750         char *path;
 751
 752         path = malloc(strlen(dir) + strlen(name) + 2);
 753         if (!path)
 754                 return NULL;
 755         strcpy(path, dir);
 756         if (dir[strlen(dir) - 1] != '/')
 757                 strcat(path, "/");
 758         strcat(path, name);
 759         return path;
 760 }
 761
 762 static int traverse_directory(struct btrfs_trans_handle *trans,
 763                               struct btrfs_root *root, char *dir_name,
 764                               struct directory_name_entry *dir_head, int out_fd)
 765 {
 766         int ret = 0;
 767
 768         struct btrfs_inode_item cur_inode;
 769         struct btrfs_inode_item *inode_item;
 770         int count, i, dir_index_cnt;
 771         struct direct **files;
 772         struct stat st;
 773         struct directory_name_entry *dir_entry, *parent_dir_entry;
 774         struct direct *cur_file;
 775         ino_t parent_inum, cur_inum;
 776         ino_t highest_inum = 0;
 777         char *parent_dir_name;
 778         char real_path[PATH_MAX];
 779         struct btrfs_path path;
 780         struct extent_buffer *leaf;
 781         struct btrfs_key root_dir_key;
 782         u64 root_dir_inode_size = 0;
 783
 784         /* Add list for source directory */
 785         dir_entry = malloc(sizeof(struct directory_name_entry));
 786         if (!dir_entry)
 787                 return -ENOMEM;
 788         dir_entry->dir_name = dir_name;
 789         dir_entry->path = realpath(dir_name, real_path);
 790         if (!dir_entry->path) {
 791                 fprintf(stderr, "get directory real path error\n");
 792                 ret = -1;
 793                 goto fail_no_dir;
 794         }
 795
 796         parent_inum = highest_inum + BTRFS_FIRST_FREE_OBJECTID;
 797         dir_entry->inum = parent_inum;
 798         list_add_tail(&dir_entry->list, &dir_head->list);
 799
 800         btrfs_init_path(&path);
 801
 802         root_dir_key.objectid = btrfs_root_dirid(&root->root_item);
 803         root_dir_key.offset = 0;
 804         btrfs_set_key_type(&root_dir_key, BTRFS_INODE_ITEM_KEY);
 805         ret = btrfs_lookup_inode(trans, root, &path, &root_dir_key, 1);
 806         if (ret) {
 807                 fprintf(stderr, "root dir lookup error\n");
 808                 goto fail_no_dir;
 809         }
 810
 811         leaf = path.nodes[0];
 812         inode_item = btrfs_item_ptr(leaf, path.slots[0],
 813                                     struct btrfs_inode_item);
 814
 815         root_dir_inode_size = calculate_dir_inode_size(dir_name);
 816         btrfs_set_inode_size(leaf, inode_item, root_dir_inode_size);
 817         btrfs_mark_buffer_dirty(leaf);
 818
 819         btrfs_release_path(&path);
 820
 821         do {
 822                 parent_dir_entry = list_entry(dir_head->list.next,
 823                                               struct directory_name_entry,
 824                                               list);
 825                 list_del(&parent_dir_entry->list);
 826
 827                 parent_inum = parent_dir_entry->inum;
 828                 parent_dir_name = parent_dir_entry->dir_name;
 829                 if (chdir(parent_dir_entry->path)) {
 830                         fprintf(stderr, "chdir error for %s\n",
 831                                 parent_dir_name);
 832                         ret = -1;
 833                         goto fail_no_files;
 834                 }
 835
 836                 count = scandir(parent_dir_entry->path, &files,
 837                                 directory_select, NULL);
 838                 if (count == -1)
 839                 {
 840                         fprintf(stderr, "scandir for %s failed: %s\n",
 841                                 parent_dir_name, strerror (errno));
 842                         ret = -1;
 843                         goto fail;
 844                 }
 845
 846                 for (i = 0; i < count; i++) {
 847                         cur_file = files[i];
 848
 849                         if (lstat(cur_file->d_name, &st) == -1) {
 850                                 fprintf(stderr, "lstat failed for file %s\n",
 851                                         cur_file->d_name);
 852                                 ret = -1;
 853                                 goto fail;
 854                         }
 855
 856                         cur_inum = st.st_ino;
 857                         ret = add_directory_items(trans, root,
 858                                                   cur_inum, parent_inum,
 859                                                   cur_file->d_name,
 860                                                   &st, &dir_index_cnt);
 861                         if (ret) {
 862                                 fprintf(stderr, "add_directory_items failed\n");
 863                                 goto fail;
 864                         }
 865
 866                         ret = add_inode_items(trans, root, &st,
 867                                               cur_file->d_name, cur_inum,
 868                                               parent_inum, dir_index_cnt,
 869                                               &cur_inode);
 870                         if (ret == -EEXIST) {
 871                                 BUG_ON(st.st_nlink <= 1);
 872                                 continue;
 873                         }
 874                         if (ret) {
 875                                 fprintf(stderr, "add_inode_items failed\n");
 876                                 goto fail;
 877                         }
 878
 879                         ret = add_xattr_item(trans, root,
 880                                              cur_inum, cur_file->d_name);
 881                         if (ret) {
 882                                 fprintf(stderr, "add_xattr_item failed\n");
 883                                 if(ret != -ENOTSUP)
 884                                         goto fail;
 885                         }
 886
 887                         if (S_ISDIR(st.st_mode)) {
 888                                 dir_entry = malloc(sizeof(struct directory_name_entry));
 889                                 if (!dir_entry) {
 890                                         ret = -ENOMEM;
 891                                         goto fail;
 892                                 }
 893                                 dir_entry->dir_name = cur_file->d_name;
 894                                 dir_entry->path = make_path(parent_dir_entry->path,
 895                                                             cur_file->d_name);
 896                                 dir_entry->inum = cur_inum;
 897                                 list_add_tail(&dir_entry->list, &dir_head->list);
 898                         } else if (S_ISREG(st.st_mode)) {
 899                                 ret = add_file_items(trans, root, &cur_inode,
 900                                                      cur_inum, parent_inum, &st,
 901                                                      cur_file->d_name, out_fd);
 902                                 if (ret) {
 903                                         fprintf(stderr, "add_file_items failed\n");
 904                                         goto fail;
 905                                 }
 906                         } else if (S_ISLNK(st.st_mode)) {
 907                                 ret = add_symbolic_link(trans, root,
 908                                                         cur_inum, cur_file->d_name);
 909                                 if (ret) {
 910                                         fprintf(stderr, "add_symbolic_link failed\n");
 911                                         goto fail;
 912                                 }
 913                         }
 914                 }
 915
 916                 free_namelist(files, count);
 917                 free(parent_dir_entry);
 918
 919                 index_cnt = 2;
 920
 921         } while (!list_empty(&dir_head->list));
 922
 923 out:
 924         return !!ret;
 925 fail:
 926         free_namelist(files, count);
 927 fail_no_files:
 928         free(parent_dir_entry);
 929         goto out;
 930 fail_no_dir:
 931         free(dir_entry);
 932         goto out;
 933 }
 934
 935 static int open_target(char *output_name)
 936 {
 937         int output_fd;
 938         output_fd = open(output_name, O_CREAT | O_RDWR,
 939                          S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH);
 940
 941         return output_fd;
 942 }
 943
 944 static int create_chunks(struct btrfs_trans_handle *trans,
 945                          struct btrfs_root *root, u64 num_of_meta_chunks,
 946                          u64 size_of_data,
 947                          struct mkfs_allocation *allocation)
 948 {
 949         u64 chunk_start;
 950         u64 chunk_size;
 951         u64 meta_type = BTRFS_BLOCK_GROUP_METADATA;
 952         u64 data_type = BTRFS_BLOCK_GROUP_DATA;
 953         u64 minimum_data_chunk_size = 8 * 1024 * 1024;
 954         u64 i;
 955         int ret;
 956
 957         for (i = 0; i < num_of_meta_chunks; i++) {
 958                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 959                                         &chunk_start, &chunk_size, meta_type);
 960                 BUG_ON(ret);
 961                 ret = btrfs_make_block_group(trans, root->fs_info->extent_root, 0,
 962                                              meta_type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 963                                              chunk_start, chunk_size);
 964                 allocation->metadata += chunk_size;
 965                 BUG_ON(ret);
 966                 set_extent_dirty(&root->fs_info->free_space_cache,
 967                                  chunk_start, chunk_start + chunk_size - 1, 0);
 968         }
 969
 970         if (size_of_data < minimum_data_chunk_size)
 971                 size_of_data = minimum_data_chunk_size;
 972
 973         ret = btrfs_alloc_data_chunk(trans, root->fs_info->extent_root,
 974                                      &chunk_start, size_of_data, data_type);
 975         BUG_ON(ret);
 976         ret = btrfs_make_block_group(trans, root->fs_info->extent_root, 0,
 977                                      data_type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 978                                      chunk_start, size_of_data);
 979         allocation->data += size_of_data;
 980         BUG_ON(ret);
 981         set_extent_dirty(&root->fs_info->free_space_cache,
 982                          chunk_start, chunk_start + size_of_data - 1, 0);
 983         return ret;
 984 }
 985
 986 static int make_image(char *source_dir, struct btrfs_root *root, int out_fd)
 987 {
 988         int ret;
 989         struct btrfs_trans_handle *trans;
 990
 991         struct stat root_st;
 992
 993         struct directory_name_entry dir_head;
 994
 995         struct directory_name_entry *dir_entry = NULL;
 996
 997         ret = lstat(source_dir, &root_st);
 998         if (ret) {
 999                 fprintf(stderr, "unable to lstat the %s\n", source_dir);
1000                 goto out;
1001         }
1002
1003         INIT_LIST_HEAD(&dir_head.list);
1004
1005         trans = btrfs_start_transaction(root, 1);
1006         ret = traverse_directory(trans, root, source_dir, &dir_head, out_fd);
1007         if (ret) {
1008                 fprintf(stderr, "unable to traverse_directory\n");
1009                 goto fail;
1010         }
1011         btrfs_commit_transaction(trans, root);
1012
1013         if (verbose)
1014                 printf("Making image is completed.\n");
1015         return 0;
1016 fail:
1017         while (!list_empty(&dir_head.list)) {
1018                 dir_entry = list_entry(dir_head.list.next,
1019                                        struct directory_name_entry, list);
1020                 list_del(&dir_entry->list);
1021                 free(dir_entry);
1022         }
1023 out:
1024         fprintf(stderr, "Making image is aborted.\n");
1025         return -1;
1026 }
1027
1028 /*
1029  * This ignores symlinks with unreadable targets and subdirs that can't
1030  * be read.  It's a best-effort to give a rough estimate of the size of
1031  * a subdir.  It doesn't guarantee that prepopulating btrfs from this
1032  * tree won't still run out of space.
1033  */
1034 static u64 global_total_size;
1035 static u64 fs_block_size;
1036 static int ftw_add_entry_size(const char *fpath, const struct stat *st,
1037                               int type)
1038 {
1039         if (type == FTW_F || type == FTW_D)
1040                 global_total_size += round_up(st->st_size, fs_block_size);
1041
1042         return 0;
1043 }
1044
1045 static u64 size_sourcedir(char *dir_name, u64 sectorsize,
1046                           u64 *num_of_meta_chunks_ret, u64 *size_of_data_ret)
1047 {
1048         u64 dir_size = 0;
1049         u64 total_size = 0;
1050         int ret;
1051         u64 default_chunk_size = 8 * 1024 * 1024;       /* 8MB */
1052         u64 allocated_meta_size = 8 * 1024 * 1024;      /* 8MB */
1053         u64 allocated_total_size = 20 * 1024 * 1024;    /* 20MB */
1054         u64 num_of_meta_chunks = 0;
1055         u64 num_of_data_chunks = 0;
1056         u64 num_of_allocated_meta_chunks =
1057                         allocated_meta_size / default_chunk_size;
1058
1059         global_total_size = 0;
1060         fs_block_size = sectorsize;
1061         ret = ftw(dir_name, ftw_add_entry_size, 10);
1062         dir_size = global_total_size;
1063         if (ret < 0) {
1064                 fprintf(stderr, "ftw subdir walk of '%s' failed: %s\n",
1065                         dir_name, strerror(errno));
1066                 exit(1);
1067         }
1068
1069         num_of_data_chunks = (dir_size + default_chunk_size - 1) /
1070                 default_chunk_size;
1071
1072         num_of_meta_chunks = (dir_size / 2) / default_chunk_size;
1073         if (((dir_size / 2) % default_chunk_size) != 0)
1074                 num_of_meta_chunks++;
1075         if (num_of_meta_chunks <= num_of_allocated_meta_chunks)
1076                 num_of_meta_chunks = 0;
1077         else
1078                 num_of_meta_chunks -= num_of_allocated_meta_chunks;
1079
1080         total_size = allocated_total_size +
1081                      (num_of_data_chunks * default_chunk_size) +
1082                      (num_of_meta_chunks * default_chunk_size);
1083
1084         *num_of_meta_chunks_ret = num_of_meta_chunks;
1085         *size_of_data_ret = num_of_data_chunks * default_chunk_size;
1086         return total_size;
1087 }
1088
1089 static int zero_output_file(int out_fd, u64 size)
1090 {
1091         int loop_num;
1092         u64 location = 0;
1093         char buf[4096];
1094         int ret = 0, i;
1095         ssize_t written;
1096
1097         memset(buf, 0, 4096);
1098         loop_num = size / 4096;
1099         for (i = 0; i < loop_num; i++) {
1100                 written = pwrite64(out_fd, buf, 4096, location);
1101                 if (written != 4096)
1102                         ret = -EIO;
1103                 location += 4096;
1104         }
1105         return ret;
1106 }
1107
1108 static int is_ssd(const char *file)
1109 {
1110         blkid_probe probe;
1111         char wholedisk[PATH_MAX];
1112         char sysfs_path[PATH_MAX];
1113         dev_t devno;
1114         int fd;
1115         char rotational;
1116         int ret;
1117
1118         probe = blkid_new_probe_from_filename(file);
1119         if (!probe)
1120                 return 0;
1121
1122         /* Device number of this disk (possibly a partition) */
1123         devno = blkid_probe_get_devno(probe);
1124         if (!devno) {
1125                 blkid_free_probe(probe);
1126                 return 0;
1127         }
1128
1129         /* Get whole disk name (not full path) for this devno */
1130         ret = blkid_devno_to_wholedisk(devno,
1131                         wholedisk, sizeof(wholedisk), NULL);
1132         if (ret) {
1133                 blkid_free_probe(probe);
1134                 return 0;
1135         }
1136
1137         snprintf(sysfs_path, PATH_MAX, "/sys/block/%s/queue/rotational",
1138                  wholedisk);
1139
1140         blkid_free_probe(probe);
1141
1142         fd = open(sysfs_path, O_RDONLY);
1143         if (fd < 0) {
1144                 return 0;
1145         }
1146
1147         if (read(fd, &rotational, sizeof(char)) < sizeof(char)) {
1148                 close(fd);
1149                 return 0;
1150         }
1151         close(fd);
1152
1153         return !atoi((const char *)&rotational);
1154 }
1155
1156 static int _cmp_device_by_id(void *priv, struct list_head *a,
1157                              struct list_head *b)
1158 {
1159         return list_entry(a, struct btrfs_device, dev_list)->devid -
1160                list_entry(b, struct btrfs_device, dev_list)->devid;
1161 }
1162
1163 static void list_all_devices(struct btrfs_root *root)
1164 {
1165         struct btrfs_fs_devices *fs_devices;
1166         struct btrfs_device *device;
1167         int number_of_devices = 0;
1168         u64 total_block_count = 0;
1169
1170         fs_devices = root->fs_info->fs_devices;
1171
1172         list_for_each_entry(device, &fs_devices->devices, dev_list)
1173                 number_of_devices++;
1174
1175         list_sort(NULL, &fs_devices->devices, _cmp_device_by_id);
1176
1177         printf("Number of devices:  %d\n", number_of_devices);
1178         /* printf("Total devices size: %10s\n", */
1179                 /* pretty_size(total_block_count)); */
1180         printf("Devices:\n");
1181         printf("   ID        SIZE  PATH\n");
1182         list_for_each_entry(device, &fs_devices->devices, dev_list) {
1183                 printf("  %3llu  %10s  %s\n",
1184                         device->devid,
1185                         pretty_size(device->total_bytes),
1186                         device->name);
1187                 total_block_count += device->total_bytes;
1188         }
1189
1190         printf("\n");
1191 }
1192
1193 static int is_temp_block_group(struct extent_buffer *node,
1194                                struct btrfs_block_group_item *bgi,
1195                                u64 data_profile, u64 meta_profile,
1196                                u64 sys_profile)
1197 {
1198         u64 flag = btrfs_disk_block_group_flags(node, bgi);
1199         u64 flag_type = flag & BTRFS_BLOCK_GROUP_TYPE_MASK;
1200         u64 flag_profile = flag & BTRFS_BLOCK_GROUP_PROFILE_MASK;
1201         u64 used = btrfs_disk_block_group_used(node, bgi);
1202
1203         /*
1204          * Chunks meets all the following conditions is a temp chunk
1205          * 1) Empty chunk
1206          * Temp chunk is always empty.
1207          *
1208          * 2) profile dismatch with mkfs profile.
1209          * Temp chunk is always in SINGLE
1210          *
1211          * 3) Size differs with mkfs_alloc
1212          * Special case for SINGLE/SINGLE btrfs.
1213          * In that case, temp data chunk and real data chunk are always empty.
1214          * So we need to use mkfs_alloc to be sure which chunk is the newly
1215          * allocated.
1216          *
1217          * Normally, new chunk size is equal to mkfs one (One chunk)
1218          * If it has multiple chunks, we just refuse to delete any one.
1219          * As they are all single, so no real problem will happen.
1220          * So only use condition 1) and 2) to judge them.
1221          */
1222         if (used != 0)
1223                 return 0;
1224         switch (flag_type) {
1225         case BTRFS_BLOCK_GROUP_DATA:
1226         case BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA:
1227                 data_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1228                 if (flag_profile != data_profile)
1229                         return 1;
1230                 break;
1231         case BTRFS_BLOCK_GROUP_METADATA:
1232                 meta_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1233                 if (flag_profile != meta_profile)
1234                         return 1;
1235                 break;
1236         case BTRFS_BLOCK_GROUP_SYSTEM:
1237                 sys_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1238                 if (flag_profile != sys_profile)
1239                         return 1;
1240                 break;
1241         }
1242         return 0;
1243 }
1244
1245 /* Note: if current is a block group, it will skip it anyway */
1246 static int next_block_group(struct btrfs_root *root,
1247                             struct btrfs_path *path)
1248 {
1249         struct btrfs_key key;
1250         int ret = 0;
1251
1252         while (1) {
1253                 ret = btrfs_next_item(root, path);
1254                 if (ret)
1255                         goto out;
1256
1257                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1258                 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
1259                         goto out;
1260         }
1261 out:
1262         return ret;
1263 }
1264
1265 /* This function will cleanup  */
1266 static int cleanup_temp_chunks(struct btrfs_fs_info *fs_info,
1267                                struct mkfs_allocation *alloc,
1268                                u64 data_profile, u64 meta_profile,
1269                                u64 sys_profile)
1270 {
1271         struct btrfs_trans_handle *trans = NULL;
1272         struct btrfs_block_group_item *bgi;
1273         struct btrfs_root *root = fs_info->extent_root;
1274         struct btrfs_key key;
1275         struct btrfs_key found_key;
1276         struct btrfs_path *path;
1277         int ret = 0;
1278
1279         path = btrfs_alloc_path();
1280         if (!path) {
1281                 ret = -ENOMEM;
1282                 goto out;
1283         }
1284
1285         trans = btrfs_start_transaction(root, 1);
1286
1287         key.objectid = 0;
1288         key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
1289         key.offset = 0;
1290
1291         while (1) {
1292                 /*
1293                  * as the rest of the loop may modify the tree, we need to
1294                  * start a new search each time.
1295                  */
1296                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
1297                 if (ret < 0)
1298                         goto out;
1299
1300                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1301                                       path->slots[0]);
1302                 if (found_key.objectid < key.objectid)
1303                         goto out;
1304                 if (found_key.type != BTRFS_BLOCK_GROUP_ITEM_KEY) {
1305                         ret = next_block_group(root, path);
1306                         if (ret < 0)
1307                                 goto out;
1308                         if (ret > 0) {
1309                                 ret = 0;
1310                                 goto out;
1311                         }
1312                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1313                                               path->slots[0]);
1314                 }
1315
1316                 bgi = btrfs_item_ptr(path->nodes[0], path->slots[0],
1317                                      struct btrfs_block_group_item);
1318                 if (is_temp_block_group(path->nodes[0], bgi,
1319                                         data_profile, meta_profile,
1320                                         sys_profile)) {
1321                         ret = btrfs_free_block_group(trans, fs_info,
1322                                         found_key.objectid, found_key.offset);
1323                         if (ret < 0)
1324                                 goto out;
1325                 }
1326                 btrfs_release_path(path);
1327                 key.objectid = found_key.objectid + found_key.offset;
1328         }
1329 out:
1330         if (trans)
1331                 btrfs_commit_transaction(trans, root);
1332         btrfs_free_path(path);
1333         return ret;
1334 }
1335
1336 int main(int ac, char **av)
1337 {
1338         char *file;
1339         struct btrfs_root *root;
1340         struct btrfs_trans_handle *trans;
1341         char *label = NULL;
1342         u64 block_count = 0;
1343         u64 dev_block_count = 0;
1344         u64 blocks[7];
1345         u64 alloc_start = 0;
1346         u64 metadata_profile = 0;
1347         u64 data_profile = 0;
1348         u32 nodesize = max_t(u32, sysconf(_SC_PAGESIZE),
1349                         BTRFS_MKFS_DEFAULT_NODE_SIZE);
1350         u32 sectorsize = 4096;
1351         u32 stripesize = 4096;
1352         int zero_end = 1;
1353         int fd;
1354         int ret;
1355         int i;
1356         int mixed = 0;
1357         int nodesize_forced = 0;
1358         int data_profile_opt = 0;
1359         int metadata_profile_opt = 0;
1360         int discard = 1;
1361         int ssd = 0;
1362         int force_overwrite = 0;
1363         char *source_dir = NULL;
1364         int source_dir_set = 0;
1365         u64 num_of_meta_chunks = 0;
1366         u64 size_of_data = 0;
1367         u64 source_dir_size = 0;
1368         int dev_cnt = 0;
1369         int saved_optind;
1370         char fs_uuid[BTRFS_UUID_UNPARSED_SIZE] = { 0 };
1371         u64 features = BTRFS_MKFS_DEFAULT_FEATURES;
1372         struct mkfs_allocation allocation = { 0 };
1373         struct btrfs_mkfs_config mkfs_cfg;
1374
1375         while(1) {
1376                 int c;
1377                 static const struct option long_options[] = {
1378                         { "alloc-start", required_argument, NULL, 'A'},
1379                         { "byte-count", required_argument, NULL, 'b' },
1380                         { "force", no_argument, NULL, 'f' },
1381                         { "leafsize", required_argument, NULL, 'l' },
1382                         { "label", required_argument, NULL, 'L'},
1383                         { "metadata", required_argument, NULL, 'm' },
1384                         { "mixed", no_argument, NULL, 'M' },
1385                         { "nodesize", required_argument, NULL, 'n' },
1386                         { "sectorsize", required_argument, NULL, 's' },
1387                         { "data", required_argument, NULL, 'd' },
1388                         { "version", no_argument, NULL, 'V' },
1389                         { "rootdir", required_argument, NULL, 'r' },
1390                         { "nodiscard", no_argument, NULL, 'K' },
1391                         { "features", required_argument, NULL, 'O' },
1392                         { "uuid", required_argument, NULL, 'U' },
1393                         { "quiet", 0, NULL, 'q' },
1394                         { "help", no_argument, NULL, GETOPT_VAL_HELP },
1395                         { NULL, 0, NULL, 0}
1396                 };
1397
1398                 c = getopt_long(ac, av, "A:b:fl:n:s:m:d:L:O:r:U:VMKq",
1399                                 long_options, NULL);
1400                 if (c < 0)
1401                         break;
1402                 switch(c) {
1403                         case 'A':
1404                                 alloc_start = parse_size(optarg);
1405                                 break;
1406                         case 'f':
1407                                 force_overwrite = 1;
1408                                 break;
1409                         case 'd':
1410                                 data_profile = parse_profile(optarg);
1411                                 data_profile_opt = 1;
1412                                 break;
1413                         case 'l':
1414                                 fprintf(stderr,
1415                         "WARNING: --leafsize is deprecated, use --nodesize\n");
1416                         case 'n':
1417                                 nodesize = parse_size(optarg);
1418                                 nodesize_forced = 1;
1419                                 break;
1420                         case 'L':
1421                                 label = parse_label(optarg);
1422                                 break;
1423                         case 'm':
1424                                 metadata_profile = parse_profile(optarg);
1425                                 metadata_profile_opt = 1;
1426                                 break;
1427                         case 'M':
1428                                 mixed = 1;
1429                                 break;
1430                         case 'O': {
1431                                 char *orig = strdup(optarg);
1432                                 char *tmp = orig;
1433
1434                                 tmp = btrfs_parse_fs_features(tmp, &features);
1435                                 if (tmp) {
1436                                         fprintf(stderr,
1437                                                 "Unrecognized filesystem feature '%s'\n",
1438                                                         tmp);
1439                                         free(orig);
1440                                         exit(1);
1441                                 }
1442                                 free(orig);
1443                                 if (features & BTRFS_FEATURE_LIST_ALL) {
1444                                         btrfs_list_all_fs_features(0);
1445                                         exit(0);
1446                                 }
1447                                 break;
1448                                 }
1449                         case 's':
1450                                 sectorsize = parse_size(optarg);
1451                                 break;
1452                         case 'b':
1453                                 block_count = parse_size(optarg);
1454                                 zero_end = 0;
1455                                 break;
1456                         case 'V':
1457                                 print_version();
1458                                 break;
1459                         case 'r':
1460                                 source_dir = optarg;
1461                                 source_dir_set = 1;
1462                                 break;
1463                         case 'U':
1464                                 strncpy(fs_uuid, optarg,
1465                                         BTRFS_UUID_UNPARSED_SIZE - 1);
1466                                 break;
1467                         case 'K':
1468                                 discard = 0;
1469                                 break;
1470                         case 'q':
1471                                 verbose = 0;
1472                                 break;
1473                         case GETOPT_VAL_HELP:
1474                         default:
1475                                 print_usage(c != GETOPT_VAL_HELP);
1476                 }
1477         }
1478
1479         if (verbose) {
1480                 printf("%s\n", PACKAGE_STRING);
1481                 printf("See %s for more information.\n\n", PACKAGE_URL);
1482         }
1483
1484         sectorsize = max(sectorsize, (u32)sysconf(_SC_PAGESIZE));
1485         saved_optind = optind;
1486         dev_cnt = ac - optind;
1487         if (dev_cnt == 0)
1488                 print_usage(1);
1489
1490         if (source_dir_set && dev_cnt > 1) {
1491                 fprintf(stderr,
1492                         "The -r option is limited to a single device\n");
1493                 exit(1);
1494         }
1495
1496         if (*fs_uuid) {
1497                 uuid_t dummy_uuid;
1498
1499                 if (uuid_parse(fs_uuid, dummy_uuid) != 0) {
1500                         fprintf(stderr, "could not parse UUID: %s\n", fs_uuid);
1501                         exit(1);
1502                 }
1503                 if (!test_uuid_unique(fs_uuid)) {
1504                         fprintf(stderr, "non-unique UUID: %s\n", fs_uuid);
1505                         exit(1);
1506                 }
1507         }
1508
1509         while (dev_cnt-- > 0) {
1510                 file = av[optind++];
1511                 if (is_block_device(file) == 1)
1512                         if (test_dev_for_mkfs(file, force_overwrite))
1513                                 exit(1);
1514         }
1515
1516         optind = saved_optind;
1517         dev_cnt = ac - optind;
1518
1519         file = av[optind++];
1520         ssd = is_ssd(file);
1521
1522         /*
1523         * Set default profiles according to number of added devices.
1524         * For mixed groups defaults are single/single.
1525         */
1526         if (!mixed) {
1527                 if (!metadata_profile_opt) {
1528                         if (dev_cnt == 1 && ssd && verbose)
1529                                 printf("Detected a SSD, turning off metadata "
1530                                 "duplication.  Mkfs with -m dup if you want to "
1531                                 "force metadata duplication.\n");
1532
1533                         metadata_profile = (dev_cnt > 1) ?
1534                                         BTRFS_BLOCK_GROUP_RAID1 : (ssd) ?
1535                                         0: BTRFS_BLOCK_GROUP_DUP;
1536                 }
1537                 if (!data_profile_opt) {
1538                         data_profile = (dev_cnt > 1) ?
1539                                 BTRFS_BLOCK_GROUP_RAID0 : 0; /* raid0 or single */
1540                 }
1541         } else {
1542                 u32 best_nodesize = max_t(u32, sysconf(_SC_PAGESIZE), sectorsize);
1543
1544                 if (metadata_profile_opt || data_profile_opt) {
1545                         if (metadata_profile != data_profile) {
1546                                 fprintf(stderr,
1547         "ERROR: With mixed block groups data and metadata profiles must be the same\n");
1548                                 exit(1);
1549                         }
1550                 }
1551
1552                 if (!nodesize_forced)
1553                         nodesize = best_nodesize;
1554         }
1555
1556         /*
1557          * FS features that can be set by other means than -O
1558          * just set the bit here
1559          */
1560         if (mixed)
1561                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS;
1562
1563         if ((data_profile | metadata_profile) &
1564             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
1565                 features |= BTRFS_FEATURE_INCOMPAT_RAID56;
1566         }
1567
1568         if (btrfs_check_nodesize(nodesize, sectorsize,
1569                                  features))
1570                 exit(1);
1571
1572         /* Check device/block_count after the nodesize is determined */
1573         if (block_count && block_count < btrfs_min_dev_size(nodesize)) {
1574                 fprintf(stderr,
1575                         "Size '%llu' is too small to make a usable filesystem\n",
1576                         block_count);
1577                 fprintf(stderr,
1578                         "Minimum size for btrfs filesystem is %llu\n",
1579                         btrfs_min_dev_size(nodesize));
1580                 exit(1);
1581         }
1582         for (i = saved_optind; i < saved_optind + dev_cnt; i++) {
1583                 char *path;
1584
1585                 path = av[i];
1586                 ret = test_minimum_size(path, nodesize);
1587                 if (ret < 0) {
1588                         fprintf(stderr, "Failed to check size for '%s': %s\n",
1589                                 path, strerror(-ret));
1590                         exit (1);
1591                 }
1592                 if (ret > 0) {
1593                         fprintf(stderr,
1594                                 "'%s' is too small to make a usable filesystem\n",
1595                                 path);
1596                         fprintf(stderr,
1597                                 "Minimum size for each btrfs device is %llu.\n",
1598                                 btrfs_min_dev_size(nodesize));
1599                         exit(1);
1600                 }
1601         }
1602         ret = test_num_disk_vs_raid(metadata_profile, data_profile,
1603                         dev_cnt, mixed, ssd);
1604         if (ret)
1605                 exit(1);
1606
1607         dev_cnt--;
1608
1609         if (!source_dir_set) {
1610                 /*
1611                  * open without O_EXCL so that the problem should not
1612                  * occur by the following processing.
1613                  * (btrfs_register_one_device() fails if O_EXCL is on)
1614                  */
1615                 fd = open(file, O_RDWR);
1616                 if (fd < 0) {
1617                         fprintf(stderr, "unable to open %s: %s\n", file,
1618                                 strerror(errno));
1619                         exit(1);
1620                 }
1621                 ret = btrfs_prepare_device(fd, file, zero_end, &dev_block_count,
1622                                            block_count, discard);
1623                 if (ret) {
1624                         close(fd);
1625                         exit(1);
1626                 }
1627                 if (block_count && block_count > dev_block_count) {
1628                         fprintf(stderr, "%s is smaller than requested size\n", file);
1629                         exit(1);
1630                 }
1631         } else {
1632                 fd = open_target(file);
1633                 if (fd < 0) {
1634                         fprintf(stderr, "unable to open the %s\n", file);
1635                         exit(1);
1636                 }
1637
1638                 source_dir_size = size_sourcedir(source_dir, sectorsize,
1639                                              &num_of_meta_chunks, &size_of_data);
1640                 if(block_count < source_dir_size)
1641                         block_count = source_dir_size;
1642                 ret = zero_output_file(fd, block_count);
1643                 if (ret) {
1644                         fprintf(stderr, "unable to zero the output file\n");
1645                         exit(1);
1646                 }
1647                 /* our "device" is the new image file */
1648                 dev_block_count = block_count;
1649         }
1650
1651         /* To create the first block group and chunk 0 in make_btrfs */
1652         if (dev_block_count < BTRFS_MKFS_SYSTEM_GROUP_SIZE) {
1653                 fprintf(stderr, "device is too small to make filesystem\n");
1654                 exit(1);
1655         }
1656
1657         blocks[0] = BTRFS_SUPER_INFO_OFFSET;
1658         for (i = 1; i < 7; i++) {
1659                 blocks[i] = BTRFS_SUPER_INFO_OFFSET + 1024 * 1024 +
1660                         nodesize * i;
1661         }
1662
1663         if (group_profile_max_safe_loss(metadata_profile) <
1664                 group_profile_max_safe_loss(data_profile)){
1665                 fprintf(stderr,
1666                         "WARNING: metatdata has lower redundancy than data!\n\n");
1667         }
1668
1669         mkfs_cfg.label = label;
1670         mkfs_cfg.fs_uuid = fs_uuid;
1671         memcpy(mkfs_cfg.blocks, blocks, sizeof(blocks));
1672         mkfs_cfg.num_bytes = dev_block_count;
1673         mkfs_cfg.nodesize = nodesize;
1674         mkfs_cfg.sectorsize = sectorsize;
1675         mkfs_cfg.stripesize = stripesize;
1676         mkfs_cfg.features = features;
1677
1678         ret = make_btrfs(fd, &mkfs_cfg);
1679         if (ret) {
1680                 fprintf(stderr, "error during mkfs: %s\n", strerror(-ret));
1681                 exit(1);
1682         }
1683
1684         root = open_ctree(file, 0, OPEN_CTREE_WRITES);
1685         if (!root) {
1686                 fprintf(stderr, "Open ctree failed\n");
1687                 close(fd);
1688                 exit(1);
1689         }
1690         root->fs_info->alloc_start = alloc_start;
1691
1692         ret = create_metadata_block_groups(root, mixed, &allocation);
1693         if (ret) {
1694                 fprintf(stderr, "failed to create default block groups\n");
1695                 exit(1);
1696         }
1697
1698         trans = btrfs_start_transaction(root, 1);
1699         if (!trans) {
1700                 fprintf(stderr, "failed to start transaction\n");
1701                 exit(1);
1702         }
1703
1704         ret = create_data_block_groups(trans, root, mixed, &allocation);
1705         if (ret) {
1706                 fprintf(stderr, "failed to create default data block groups\n");
1707                 exit(1);
1708         }
1709
1710         ret = make_root_dir(trans, root, &allocation);
1711         if (ret) {
1712                 fprintf(stderr, "failed to setup the root directory\n");
1713                 exit(1);
1714         }
1715
1716         btrfs_commit_transaction(trans, root);
1717
1718         trans = btrfs_start_transaction(root, 1);
1719         if (!trans) {
1720                 fprintf(stderr, "failed to start transaction\n");
1721                 exit(1);
1722         }
1723
1724         if (is_block_device(file) == 1)
1725                 btrfs_register_one_device(file);
1726
1727         if (dev_cnt == 0)
1728                 goto raid_groups;
1729
1730         while (dev_cnt-- > 0) {
1731                 file = av[optind++];
1732
1733                 /*
1734                  * open without O_EXCL so that the problem should not
1735                  * occur by the following processing.
1736                  * (btrfs_register_one_device() fails if O_EXCL is on)
1737                  */
1738                 fd = open(file, O_RDWR);
1739                 if (fd < 0) {
1740                         fprintf(stderr, "unable to open %s: %s\n", file,
1741                                 strerror(errno));
1742                         exit(1);
1743                 }
1744                 ret = btrfs_device_already_in_root(root, fd,
1745                                                    BTRFS_SUPER_INFO_OFFSET);
1746                 if (ret) {
1747                         fprintf(stderr, "skipping duplicate device %s in FS\n",
1748                                 file);
1749                         close(fd);
1750                         continue;
1751                 }
1752                 ret = btrfs_prepare_device(fd, file, zero_end, &dev_block_count,
1753                                            block_count, discard);
1754                 if (ret) {
1755                         close(fd);
1756                         exit(1);
1757                 }
1758
1759                 ret = btrfs_add_to_fsid(trans, root, fd, file, dev_block_count,
1760                                         sectorsize, sectorsize, sectorsize);
1761                 BUG_ON(ret);
1762                 if (verbose >= 2) {
1763                         struct btrfs_device *device;
1764
1765                         device = container_of(root->fs_info->fs_devices->devices.next,
1766                                         struct btrfs_device, dev_list);
1767                         printf("adding device %s id %llu\n", file,
1768                                 (unsigned long long)device->devid);
1769                 }
1770
1771                 if (is_block_device(file) == 1)
1772                         btrfs_register_one_device(file);
1773         }
1774
1775 raid_groups:
1776         if (!source_dir_set) {
1777                 ret = create_raid_groups(trans, root, data_profile,
1778                                  metadata_profile, mixed, &allocation);
1779                 BUG_ON(ret);
1780         }
1781
1782         ret = create_data_reloc_tree(trans, root);
1783         BUG_ON(ret);
1784
1785         btrfs_commit_transaction(trans, root);
1786
1787         if (source_dir_set) {
1788                 trans = btrfs_start_transaction(root, 1);
1789                 ret = create_chunks(trans, root,
1790                                     num_of_meta_chunks, size_of_data,
1791                                     &allocation);
1792                 BUG_ON(ret);
1793                 btrfs_commit_transaction(trans, root);
1794
1795                 ret = make_image(source_dir, root, fd);
1796                 BUG_ON(ret);
1797         }
1798         ret = cleanup_temp_chunks(root->fs_info, &allocation, data_profile,
1799                                   metadata_profile, metadata_profile);
1800         if (ret < 0) {
1801                 fprintf(stderr, "Failed to cleanup temporary chunks\n");
1802                 goto out;
1803         }
1804
1805         if (verbose) {
1806                 char features_buf[64];
1807
1808                 printf("Label:              %s\n", label);
1809                 printf("UUID:               %s\n", fs_uuid);
1810                 printf("Node size:          %u\n", nodesize);
1811                 printf("Sector size:        %u\n", sectorsize);
1812                 printf("Filesystem size:    %s\n",
1813                         pretty_size(btrfs_super_total_bytes(root->fs_info->super_copy)));
1814                 printf("Block group profiles:\n");
1815                 if (allocation.data)
1816                         printf("  Data:             %-8s %16s\n",
1817                                 btrfs_group_profile_str(data_profile),
1818                                 pretty_size(allocation.data));
1819                 if (allocation.metadata)
1820                         printf("  Metadata:         %-8s %16s\n",
1821                                 btrfs_group_profile_str(metadata_profile),
1822                                 pretty_size(allocation.metadata));
1823                 if (allocation.mixed)
1824                         printf("  Data+Metadata:    %-8s %16s\n",
1825                                 btrfs_group_profile_str(data_profile),
1826                                 pretty_size(allocation.mixed));
1827                 printf("  System:           %-8s %16s\n",
1828                         btrfs_group_profile_str(metadata_profile),
1829                         pretty_size(allocation.system));
1830                 printf("SSD detected:       %s\n", ssd ? "yes" : "no");
1831                 btrfs_parse_features_to_string(features_buf, features);
1832                 printf("Incompat features:  %s", features_buf);
1833                 printf("\n");
1834
1835                 list_all_devices(root);
1836         }
1837
1838 out:
1839         ret = close_ctree(root);
1840         BUG_ON(ret);
1841         btrfs_close_all_devices();
1842         free(label);
1843         return 0;
1844 }