mkfs/main.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 #include "help.h"
  47 #include "mkfs/common.h"
  48 #include "fsfeatures.h"
  49
  50 static u64 index_cnt = 2;
  51 static int verbose = 1;
  52
  53 struct directory_name_entry {
  54         const char *dir_name;
  55         const char *path;
  56         ino_t inum;
  57         struct list_head list;
  58 };
  59
  60 struct mkfs_allocation {
  61         u64 data;
  62         u64 metadata;
  63         u64 mixed;
  64         u64 system;
  65 };
  66
  67 static int create_metadata_block_groups(struct btrfs_root *root, int mixed,
  68                                 struct mkfs_allocation *allocation)
  69 {
  70         struct btrfs_trans_handle *trans;
  71         u64 bytes_used;
  72         u64 chunk_start = 0;
  73         u64 chunk_size = 0;
  74         int ret;
  75
  76         trans = btrfs_start_transaction(root, 1);
  77         bytes_used = btrfs_super_bytes_used(root->fs_info->super_copy);
  78
  79         root->fs_info->system_allocs = 1;
  80         ret = btrfs_make_block_group(trans, root, bytes_used,
  81                                      BTRFS_BLOCK_GROUP_SYSTEM,
  82                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
  83                                      0, BTRFS_MKFS_SYSTEM_GROUP_SIZE);
  84         allocation->system += BTRFS_MKFS_SYSTEM_GROUP_SIZE;
  85         if (ret)
  86                 return ret;
  87
  88         if (mixed) {
  89                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
  90                                         &chunk_start, &chunk_size,
  91                                         BTRFS_BLOCK_GROUP_METADATA |
  92                                         BTRFS_BLOCK_GROUP_DATA);
  93                 if (ret == -ENOSPC) {
  94                         error("no space to allocate data/metadata chunk");
  95                         goto err;
  96                 }
  97                 if (ret)
  98                         return ret;
  99                 ret = btrfs_make_block_group(trans, root, 0,
 100                                              BTRFS_BLOCK_GROUP_METADATA |
 101                                              BTRFS_BLOCK_GROUP_DATA,
 102                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 103                                              chunk_start, chunk_size);
 104                 if (ret)
 105                         return ret;
 106                 allocation->mixed += chunk_size;
 107         } else {
 108                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 109                                         &chunk_start, &chunk_size,
 110                                         BTRFS_BLOCK_GROUP_METADATA);
 111                 if (ret == -ENOSPC) {
 112                         error("no space to allocate metadata chunk");
 113                         goto err;
 114                 }
 115                 if (ret)
 116                         return ret;
 117                 ret = btrfs_make_block_group(trans, root, 0,
 118                                              BTRFS_BLOCK_GROUP_METADATA,
 119                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 120                                              chunk_start, chunk_size);
 121                 allocation->metadata += chunk_size;
 122                 if (ret)
 123                         return ret;
 124         }
 125
 126         root->fs_info->system_allocs = 0;
 127         ret = btrfs_commit_transaction(trans, root);
 128
 129 err:
 130         return ret;
 131 }
 132
 133 static int create_data_block_groups(struct btrfs_trans_handle *trans,
 134                 struct btrfs_root *root, int mixed,
 135                 struct mkfs_allocation *allocation)
 136 {
 137         u64 chunk_start = 0;
 138         u64 chunk_size = 0;
 139         int ret = 0;
 140
 141         if (!mixed) {
 142                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 143                                         &chunk_start, &chunk_size,
 144                                         BTRFS_BLOCK_GROUP_DATA);
 145                 if (ret == -ENOSPC) {
 146                         error("no space to allocate data chunk");
 147                         goto err;
 148                 }
 149                 if (ret)
 150                         return ret;
 151                 ret = btrfs_make_block_group(trans, root, 0,
 152                                              BTRFS_BLOCK_GROUP_DATA,
 153                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 154                                              chunk_start, chunk_size);
 155                 allocation->data += chunk_size;
 156                 if (ret)
 157                         return ret;
 158         }
 159
 160 err:
 161         return ret;
 162 }
 163
 164 static int make_root_dir(struct btrfs_trans_handle *trans,
 165                 struct btrfs_root *root)
 166 {
 167         struct btrfs_key location;
 168         int ret;
 169
 170         ret = btrfs_make_root_dir(trans, root->fs_info->tree_root,
 171                               BTRFS_ROOT_TREE_DIR_OBJECTID);
 172         if (ret)
 173                 goto err;
 174         ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
 175         if (ret)
 176                 goto err;
 177         memcpy(&location, &root->fs_info->fs_root->root_key, sizeof(location));
 178         location.offset = (u64)-1;
 179         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
 180                         "default", 7,
 181                         btrfs_super_root_dir(root->fs_info->super_copy),
 182                         &location, BTRFS_FT_DIR, 0);
 183         if (ret)
 184                 goto err;
 185
 186         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 187                              "default", 7, location.objectid,
 188                              BTRFS_ROOT_TREE_DIR_OBJECTID, 0);
 189         if (ret)
 190                 goto err;
 191
 192 err:
 193         return ret;
 194 }
 195
 196 static int __recow_root(struct btrfs_trans_handle *trans,
 197                          struct btrfs_root *root)
 198 {
 199         struct extent_buffer *tmp;
 200         int ret;
 201
 202         if (trans->transid != btrfs_root_generation(&root->root_item)) {
 203                 extent_buffer_get(root->node);
 204                 ret = __btrfs_cow_block(trans, root, root->node,
 205                                         NULL, 0, &tmp, 0, 0);
 206                 if (ret)
 207                         return ret;
 208                 free_extent_buffer(tmp);
 209         }
 210
 211         return 0;
 212 }
 213
 214 static int recow_roots(struct btrfs_trans_handle *trans,
 215                        struct btrfs_root *root)
 216 {
 217         struct btrfs_fs_info *info = root->fs_info;
 218         int ret;
 219
 220         ret = __recow_root(trans, info->fs_root);
 221         if (ret)
 222                 return ret;
 223         ret = __recow_root(trans, info->tree_root);
 224         if (ret)
 225                 return ret;
 226         ret = __recow_root(trans, info->extent_root);
 227         if (ret)
 228                 return ret;
 229         ret = __recow_root(trans, info->chunk_root);
 230         if (ret)
 231                 return ret;
 232         ret = __recow_root(trans, info->dev_root);
 233         if (ret)
 234                 return ret;
 235         ret = __recow_root(trans, info->csum_root);
 236         if (ret)
 237                 return ret;
 238
 239         return 0;
 240 }
 241
 242 static int create_one_raid_group(struct btrfs_trans_handle *trans,
 243                               struct btrfs_root *root, u64 type,
 244                               struct mkfs_allocation *allocation)
 245
 246 {
 247         u64 chunk_start;
 248         u64 chunk_size;
 249         int ret;
 250
 251         ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 252                                 &chunk_start, &chunk_size, type);
 253         if (ret == -ENOSPC) {
 254                 error("not enough free space to allocate chunk");
 255                 exit(1);
 256         }
 257         if (ret)
 258                 return ret;
 259
 260         ret = btrfs_make_block_group(trans, root->fs_info->extent_root, 0,
 261                                      type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 262                                      chunk_start, chunk_size);
 263
 264         type &= BTRFS_BLOCK_GROUP_TYPE_MASK;
 265         if (type == BTRFS_BLOCK_GROUP_DATA) {
 266                 allocation->data += chunk_size;
 267         } else if (type == BTRFS_BLOCK_GROUP_METADATA) {
 268                 allocation->metadata += chunk_size;
 269         } else if (type == BTRFS_BLOCK_GROUP_SYSTEM) {
 270                 allocation->system += chunk_size;
 271         } else if (type ==
 272                         (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA)) {
 273                 allocation->mixed += chunk_size;
 274         } else {
 275                 error("unrecognized profile type: 0x%llx",
 276                                 (unsigned long long)type);
 277                 ret = -EINVAL;
 278         }
 279
 280         return ret;
 281 }
 282
 283 static int create_raid_groups(struct btrfs_trans_handle *trans,
 284                               struct btrfs_root *root, u64 data_profile,
 285                               u64 metadata_profile, int mixed,
 286                               struct mkfs_allocation *allocation)
 287 {
 288         int ret;
 289
 290         if (metadata_profile) {
 291                 u64 meta_flags = BTRFS_BLOCK_GROUP_METADATA;
 292
 293                 ret = create_one_raid_group(trans, root,
 294                                             BTRFS_BLOCK_GROUP_SYSTEM |
 295                                             metadata_profile, allocation);
 296                 if (ret)
 297                         return ret;
 298
 299                 if (mixed)
 300                         meta_flags |= BTRFS_BLOCK_GROUP_DATA;
 301
 302                 ret = create_one_raid_group(trans, root, meta_flags |
 303                                             metadata_profile, allocation);
 304                 if (ret)
 305                         return ret;
 306
 307         }
 308         if (!mixed && data_profile) {
 309                 ret = create_one_raid_group(trans, root,
 310                                             BTRFS_BLOCK_GROUP_DATA |
 311                                             data_profile, allocation);
 312                 if (ret)
 313                         return ret;
 314         }
 315         ret = recow_roots(trans, root);
 316
 317         return ret;
 318 }
 319
 320 static int create_data_reloc_tree(struct btrfs_trans_handle *trans,
 321                                   struct btrfs_root *root)
 322 {
 323         struct btrfs_key location;
 324         struct btrfs_root_item root_item;
 325         struct extent_buffer *tmp;
 326         u64 objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
 327         int ret;
 328
 329         ret = btrfs_copy_root(trans, root, root->node, &tmp, objectid);
 330         if (ret)
 331                 return ret;
 332
 333         memcpy(&root_item, &root->root_item, sizeof(root_item));
 334         btrfs_set_root_bytenr(&root_item, tmp->start);
 335         btrfs_set_root_level(&root_item, btrfs_header_level(tmp));
 336         btrfs_set_root_generation(&root_item, trans->transid);
 337         free_extent_buffer(tmp);
 338
 339         location.objectid = objectid;
 340         location.type = BTRFS_ROOT_ITEM_KEY;
 341         location.offset = 0;
 342         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
 343                                 &location, &root_item);
 344
 345         return ret;
 346 }
 347
 348 static void print_usage(int ret)
 349 {
 350         printf("Usage: mkfs.btrfs [options] dev [ dev ... ]\n");
 351         printf("Options:\n");
 352         printf("  allocation profiles:\n");
 353         printf("\t-d|--data PROFILE       data profile, raid0, raid1, raid5, raid6, raid10, dup or single\n");
 354         printf("\t-m|--metadata PROFILE   metadata profile, values like for data profile\n");
 355         printf("\t-M|--mixed              mix metadata and data together\n");
 356         printf("  features:\n");
 357         printf("\t-n|--nodesize SIZE      size of btree nodes\n");
 358         printf("\t-s|--sectorsize SIZE    data block size (may not be mountable by current kernel)\n");
 359         printf("\t-O|--features LIST      comma separated list of filesystem features (use '-O list-all' to list features)\n");
 360         printf("\t-L|--label LABEL        set the filesystem label\n");
 361         printf("\t-U|--uuid UUID          specify the filesystem UUID (must be unique)\n");
 362         printf("  creation:\n");
 363         printf("\t-b|--byte-count SIZE    set filesystem size to SIZE (on the first device)\n");
 364         printf("\t-r|--rootdir DIR        copy files from DIR to the image root directory\n");
 365         printf("\t-K|--nodiscard          do not perform whole device TRIM\n");
 366         printf("\t-f|--force              force overwrite of existing filesystem\n");
 367         printf("  general:\n");
 368         printf("\t-q|--quiet              no messages except errors\n");
 369         printf("\t-V|--version            print the mkfs.btrfs version and exit\n");
 370         printf("\t--help                  print this help and exit\n");
 371         printf("  deprecated:\n");
 372         printf("\t-A|--alloc-start START  the offset to start the filesystem\n");
 373         printf("\t-l|--leafsize SIZE      deprecated, alias for nodesize\n");
 374         exit(ret);
 375 }
 376
 377 static u64 parse_profile(const char *s)
 378 {
 379         if (strcasecmp(s, "raid0") == 0) {
 380                 return BTRFS_BLOCK_GROUP_RAID0;
 381         } else if (strcasecmp(s, "raid1") == 0) {
 382                 return BTRFS_BLOCK_GROUP_RAID1;
 383         } else if (strcasecmp(s, "raid5") == 0) {
 384                 return BTRFS_BLOCK_GROUP_RAID5;
 385         } else if (strcasecmp(s, "raid6") == 0) {
 386                 return BTRFS_BLOCK_GROUP_RAID6;
 387         } else if (strcasecmp(s, "raid10") == 0) {
 388                 return BTRFS_BLOCK_GROUP_RAID10;
 389         } else if (strcasecmp(s, "dup") == 0) {
 390                 return BTRFS_BLOCK_GROUP_DUP;
 391         } else if (strcasecmp(s, "single") == 0) {
 392                 return 0;
 393         } else {
 394                 error("unknown profile %s", s);
 395                 exit(1);
 396         }
 397         /* not reached */
 398         return 0;
 399 }
 400
 401 static char *parse_label(const char *input)
 402 {
 403         int len = strlen(input);
 404
 405         if (len >= BTRFS_LABEL_SIZE) {
 406                 error("label %s is too long (max %d)", input,
 407                         BTRFS_LABEL_SIZE - 1);
 408                 exit(1);
 409         }
 410         return strdup(input);
 411 }
 412
 413 static int add_directory_items(struct btrfs_trans_handle *trans,
 414                                struct btrfs_root *root, u64 objectid,
 415                                ino_t parent_inum, const char *name,
 416                                struct stat *st, int *dir_index_cnt)
 417 {
 418         int ret;
 419         int name_len;
 420         struct btrfs_key location;
 421         u8 filetype = 0;
 422
 423         name_len = strlen(name);
 424
 425         location.objectid = objectid;
 426         location.offset = 0;
 427         location.type = BTRFS_INODE_ITEM_KEY;
 428
 429         if (S_ISDIR(st->st_mode))
 430                 filetype = BTRFS_FT_DIR;
 431         if (S_ISREG(st->st_mode))
 432                 filetype = BTRFS_FT_REG_FILE;
 433         if (S_ISLNK(st->st_mode))
 434                 filetype = BTRFS_FT_SYMLINK;
 435
 436         ret = btrfs_insert_dir_item(trans, root, name, name_len,
 437                                     parent_inum, &location,
 438                                     filetype, index_cnt);
 439         if (ret)
 440                 return ret;
 441         ret = btrfs_insert_inode_ref(trans, root, name, name_len,
 442                                      objectid, parent_inum, index_cnt);
 443         *dir_index_cnt = index_cnt;
 444         index_cnt++;
 445
 446         return ret;
 447 }
 448
 449 static int fill_inode_item(struct btrfs_trans_handle *trans,
 450                            struct btrfs_root *root,
 451                            struct btrfs_inode_item *dst, struct stat *src)
 452 {
 453         u64 blocks = 0;
 454         u64 sectorsize = root->sectorsize;
 455
 456         /*
 457          * btrfs_inode_item has some reserved fields
 458          * and represents on-disk inode entry, so
 459          * zero everything to prevent information leak
 460          */
 461         memset(dst, 0, sizeof (*dst));
 462
 463         btrfs_set_stack_inode_generation(dst, trans->transid);
 464         btrfs_set_stack_inode_size(dst, src->st_size);
 465         btrfs_set_stack_inode_nbytes(dst, 0);
 466         btrfs_set_stack_inode_block_group(dst, 0);
 467         btrfs_set_stack_inode_nlink(dst, src->st_nlink);
 468         btrfs_set_stack_inode_uid(dst, src->st_uid);
 469         btrfs_set_stack_inode_gid(dst, src->st_gid);
 470         btrfs_set_stack_inode_mode(dst, src->st_mode);
 471         btrfs_set_stack_inode_rdev(dst, 0);
 472         btrfs_set_stack_inode_flags(dst, 0);
 473         btrfs_set_stack_timespec_sec(&dst->atime, src->st_atime);
 474         btrfs_set_stack_timespec_nsec(&dst->atime, 0);
 475         btrfs_set_stack_timespec_sec(&dst->ctime, src->st_ctime);
 476         btrfs_set_stack_timespec_nsec(&dst->ctime, 0);
 477         btrfs_set_stack_timespec_sec(&dst->mtime, src->st_mtime);
 478         btrfs_set_stack_timespec_nsec(&dst->mtime, 0);
 479         btrfs_set_stack_timespec_sec(&dst->otime, 0);
 480         btrfs_set_stack_timespec_nsec(&dst->otime, 0);
 481
 482         if (S_ISDIR(src->st_mode)) {
 483                 btrfs_set_stack_inode_size(dst, 0);
 484                 btrfs_set_stack_inode_nlink(dst, 1);
 485         }
 486         if (S_ISREG(src->st_mode)) {
 487                 btrfs_set_stack_inode_size(dst, (u64)src->st_size);
 488                 if (src->st_size <= BTRFS_MAX_INLINE_DATA_SIZE(root))
 489                         btrfs_set_stack_inode_nbytes(dst, src->st_size);
 490                 else {
 491                         blocks = src->st_size / sectorsize;
 492                         if (src->st_size % sectorsize)
 493                                 blocks += 1;
 494                         blocks *= sectorsize;
 495                         btrfs_set_stack_inode_nbytes(dst, blocks);
 496                 }
 497         }
 498         if (S_ISLNK(src->st_mode))
 499                 btrfs_set_stack_inode_nbytes(dst, src->st_size + 1);
 500
 501         return 0;
 502 }
 503
 504 static int directory_select(const struct direct *entry)
 505 {
 506         if (entry->d_name[0] == '.' &&
 507                 (entry->d_name[1] == 0 ||
 508                  (entry->d_name[1] == '.' && entry->d_name[2] == 0)))
 509                 return 0;
 510         return 1;
 511 }
 512
 513 static void free_namelist(struct direct **files, int count)
 514 {
 515         int i;
 516
 517         if (count < 0)
 518                 return;
 519
 520         for (i = 0; i < count; ++i)
 521                 free(files[i]);
 522         free(files);
 523 }
 524
 525 static u64 calculate_dir_inode_size(const char *dirname)
 526 {
 527         int count, i;
 528         struct direct **files, *cur_file;
 529         u64 dir_inode_size = 0;
 530
 531         count = scandir(dirname, &files, directory_select, NULL);
 532
 533         for (i = 0; i < count; i++) {
 534                 cur_file = files[i];
 535                 dir_inode_size += strlen(cur_file->d_name);
 536         }
 537
 538         free_namelist(files, count);
 539
 540         dir_inode_size *= 2;
 541         return dir_inode_size;
 542 }
 543
 544 static int add_inode_items(struct btrfs_trans_handle *trans,
 545                            struct btrfs_root *root,
 546                            struct stat *st, const char *name,
 547                            u64 self_objectid,
 548                            struct btrfs_inode_item *inode_ret)
 549 {
 550         int ret;
 551         struct btrfs_inode_item btrfs_inode;
 552         u64 objectid;
 553         u64 inode_size = 0;
 554
 555         fill_inode_item(trans, root, &btrfs_inode, st);
 556         objectid = self_objectid;
 557
 558         if (S_ISDIR(st->st_mode)) {
 559                 inode_size = calculate_dir_inode_size(name);
 560                 btrfs_set_stack_inode_size(&btrfs_inode, inode_size);
 561         }
 562
 563         ret = btrfs_insert_inode(trans, root, objectid, &btrfs_inode);
 564
 565         *inode_ret = btrfs_inode;
 566         return ret;
 567 }
 568
 569 static int add_xattr_item(struct btrfs_trans_handle *trans,
 570                           struct btrfs_root *root, u64 objectid,
 571                           const char *file_name)
 572 {
 573         int ret;
 574         int cur_name_len;
 575         char xattr_list[XATTR_LIST_MAX];
 576         char *cur_name;
 577         char cur_value[XATTR_SIZE_MAX];
 578         char delimiter = '\0';
 579         char *next_location = xattr_list;
 580
 581         ret = llistxattr(file_name, xattr_list, XATTR_LIST_MAX);
 582         if (ret < 0) {
 583                 if(errno == ENOTSUP)
 584                         return 0;
 585                 error("getting a list of xattr failed for %s: %s", file_name,
 586                                 strerror(errno));
 587                 return ret;
 588         }
 589         if (ret == 0)
 590                 return ret;
 591
 592         cur_name = strtok(xattr_list, &delimiter);
 593         while (cur_name != NULL) {
 594                 cur_name_len = strlen(cur_name);
 595                 next_location += cur_name_len + 1;
 596
 597                 ret = getxattr(file_name, cur_name, cur_value, XATTR_SIZE_MAX);
 598                 if (ret < 0) {
 599                         if(errno == ENOTSUP)
 600                                 return 0;
 601                         error("gettig a xattr value failed for %s attr %s: %s",
 602                                 file_name, cur_name, strerror(errno));
 603                         return ret;
 604                 }
 605
 606                 ret = btrfs_insert_xattr_item(trans, root, cur_name,
 607                                               cur_name_len, cur_value,
 608                                               ret, objectid);
 609                 if (ret) {
 610                         error("inserting a xattr item failed for %s: %s",
 611                                         file_name, strerror(-ret));
 612                 }
 613
 614                 cur_name = strtok(next_location, &delimiter);
 615         }
 616
 617         return ret;
 618 }
 619
 620 static int add_symbolic_link(struct btrfs_trans_handle *trans,
 621                              struct btrfs_root *root,
 622                              u64 objectid, const char *path_name)
 623 {
 624         int ret;
 625         char buf[PATH_MAX];
 626
 627         ret = readlink(path_name, buf, sizeof(buf));
 628         if (ret <= 0) {
 629                 error("readlink failed for %s: %s", path_name, strerror(errno));
 630                 goto fail;
 631         }
 632         if (ret >= sizeof(buf)) {
 633                 error("symlink too long for %s", path_name);
 634                 ret = -1;
 635                 goto fail;
 636         }
 637
 638         buf[ret] = '\0'; /* readlink does not do it for us */
 639         ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
 640                                          buf, ret + 1);
 641 fail:
 642         return ret;
 643 }
 644
 645 static int add_file_items(struct btrfs_trans_handle *trans,
 646                           struct btrfs_root *root,
 647                           struct btrfs_inode_item *btrfs_inode, u64 objectid,
 648                           struct stat *st, const char *path_name)
 649 {
 650         int ret = -1;
 651         ssize_t ret_read;
 652         u64 bytes_read = 0;
 653         struct btrfs_key key;
 654         int blocks;
 655         u32 sectorsize = root->sectorsize;
 656         u64 first_block = 0;
 657         u64 file_pos = 0;
 658         u64 cur_bytes;
 659         u64 total_bytes;
 660         struct extent_buffer *eb = NULL;
 661         int fd;
 662
 663         if (st->st_size == 0)
 664                 return 0;
 665
 666         fd = open(path_name, O_RDONLY);
 667         if (fd == -1) {
 668                 error("cannot open %s: %s", path_name, strerror(errno));
 669                 return ret;
 670         }
 671
 672         blocks = st->st_size / sectorsize;
 673         if (st->st_size % sectorsize)
 674                 blocks += 1;
 675
 676         if (st->st_size <= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 677                 char *buffer = malloc(st->st_size);
 678
 679                 if (!buffer) {
 680                         ret = -ENOMEM;
 681                         goto end;
 682                 }
 683
 684                 ret_read = pread64(fd, buffer, st->st_size, bytes_read);
 685                 if (ret_read == -1) {
 686                         error("cannot read %s at offset %llu length %llu: %s",
 687                                 path_name, (unsigned long long)bytes_read,
 688                                 (unsigned long long)st->st_size,
 689                                 strerror(errno));
 690                         free(buffer);
 691                         goto end;
 692                 }
 693
 694                 ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
 695                                                  buffer, st->st_size);
 696                 free(buffer);
 697                 goto end;
 698         }
 699
 700         /* round up our st_size to the FS blocksize */
 701         total_bytes = (u64)blocks * sectorsize;
 702
 703         /*
 704          * do our IO in extent buffers so it can work
 705          * against any raid type
 706          */
 707         eb = calloc(1, sizeof(*eb) + sectorsize);
 708         if (!eb) {
 709                 ret = -ENOMEM;
 710                 goto end;
 711         }
 712
 713 again:
 714
 715         /*
 716          * keep our extent size at 1MB max, this makes it easier to work inside
 717          * the tiny block groups created during mkfs
 718          */
 719         cur_bytes = min(total_bytes, 1024ULL * 1024);
 720         ret = btrfs_reserve_extent(trans, root, cur_bytes, 0, 0, (u64)-1,
 721                                    &key, 1);
 722         if (ret)
 723                 goto end;
 724
 725         first_block = key.objectid;
 726         bytes_read = 0;
 727
 728         while (bytes_read < cur_bytes) {
 729
 730                 memset(eb->data, 0, sectorsize);
 731
 732                 ret_read = pread64(fd, eb->data, sectorsize, file_pos + bytes_read);
 733                 if (ret_read == -1) {
 734                         error("cannot read %s at offset %llu length %llu: %s",
 735                                 path_name,
 736                                 (unsigned long long)file_pos + bytes_read,
 737                                 (unsigned long long)sectorsize,
 738                                 strerror(errno));
 739                         goto end;
 740                 }
 741
 742                 eb->start = first_block + bytes_read;
 743                 eb->len = sectorsize;
 744
 745                 /*
 746                  * we're doing the csum before we record the extent, but
 747                  * that's ok
 748                  */
 749                 ret = btrfs_csum_file_block(trans, root->fs_info->csum_root,
 750                                             first_block + bytes_read + sectorsize,
 751                                             first_block + bytes_read,
 752                                             eb->data, sectorsize);
 753                 if (ret)
 754                         goto end;
 755
 756                 ret = write_and_map_eb(root, eb);
 757                 if (ret) {
 758                         error("failed to write %s", path_name);
 759                         goto end;
 760                 }
 761
 762                 bytes_read += sectorsize;
 763         }
 764
 765         if (bytes_read) {
 766                 ret = btrfs_record_file_extent(trans, root, objectid, btrfs_inode,
 767                                                file_pos, first_block, cur_bytes);
 768                 if (ret)
 769                         goto end;
 770
 771         }
 772
 773         file_pos += cur_bytes;
 774         total_bytes -= cur_bytes;
 775
 776         if (total_bytes)
 777                 goto again;
 778
 779 end:
 780         free(eb);
 781         close(fd);
 782         return ret;
 783 }
 784
 785 static char *make_path(const char *dir, const char *name)
 786 {
 787         char *path;
 788
 789         path = malloc(strlen(dir) + strlen(name) + 2);
 790         if (!path)
 791                 return NULL;
 792         strcpy(path, dir);
 793         if (dir[strlen(dir) - 1] != '/')
 794                 strcat(path, "/");
 795         strcat(path, name);
 796         return path;
 797 }
 798
 799 static int traverse_directory(struct btrfs_trans_handle *trans,
 800                               struct btrfs_root *root, const char *dir_name,
 801                               struct directory_name_entry *dir_head, int out_fd)
 802 {
 803         int ret = 0;
 804
 805         struct btrfs_inode_item cur_inode;
 806         struct btrfs_inode_item *inode_item;
 807         int count, i, dir_index_cnt;
 808         struct direct **files;
 809         struct stat st;
 810         struct directory_name_entry *dir_entry, *parent_dir_entry;
 811         struct direct *cur_file;
 812         ino_t parent_inum, cur_inum;
 813         ino_t highest_inum = 0;
 814         const char *parent_dir_name;
 815         char real_path[PATH_MAX];
 816         struct btrfs_path path;
 817         struct extent_buffer *leaf;
 818         struct btrfs_key root_dir_key;
 819         u64 root_dir_inode_size = 0;
 820
 821         /* Add list for source directory */
 822         dir_entry = malloc(sizeof(struct directory_name_entry));
 823         if (!dir_entry)
 824                 return -ENOMEM;
 825         dir_entry->dir_name = dir_name;
 826         dir_entry->path = realpath(dir_name, real_path);
 827         if (!dir_entry->path) {
 828                 error("realpath  failed for %s: %s", dir_name, strerror(errno));
 829                 ret = -1;
 830                 goto fail_no_dir;
 831         }
 832
 833         parent_inum = highest_inum + BTRFS_FIRST_FREE_OBJECTID;
 834         dir_entry->inum = parent_inum;
 835         list_add_tail(&dir_entry->list, &dir_head->list);
 836
 837         btrfs_init_path(&path);
 838
 839         root_dir_key.objectid = btrfs_root_dirid(&root->root_item);
 840         root_dir_key.offset = 0;
 841         root_dir_key.type = BTRFS_INODE_ITEM_KEY;
 842         ret = btrfs_lookup_inode(trans, root, &path, &root_dir_key, 1);
 843         if (ret) {
 844                 error("failed to lookup root dir: %d", ret);
 845                 goto fail_no_dir;
 846         }
 847
 848         leaf = path.nodes[0];
 849         inode_item = btrfs_item_ptr(leaf, path.slots[0],
 850                                     struct btrfs_inode_item);
 851
 852         root_dir_inode_size = calculate_dir_inode_size(dir_name);
 853         btrfs_set_inode_size(leaf, inode_item, root_dir_inode_size);
 854         btrfs_mark_buffer_dirty(leaf);
 855
 856         btrfs_release_path(&path);
 857
 858         do {
 859                 parent_dir_entry = list_entry(dir_head->list.next,
 860                                               struct directory_name_entry,
 861                                               list);
 862                 list_del(&parent_dir_entry->list);
 863
 864                 parent_inum = parent_dir_entry->inum;
 865                 parent_dir_name = parent_dir_entry->dir_name;
 866                 if (chdir(parent_dir_entry->path)) {
 867                         error("chdir failed for %s: %s",
 868                                 parent_dir_name, strerror(errno));
 869                         ret = -1;
 870                         goto fail_no_files;
 871                 }
 872
 873                 count = scandir(parent_dir_entry->path, &files,
 874                                 directory_select, NULL);
 875                 if (count == -1)
 876                 {
 877                         error("scandir failed for %s: %s",
 878                                 parent_dir_name, strerror (errno));
 879                         ret = -1;
 880                         goto fail;
 881                 }
 882
 883                 for (i = 0; i < count; i++) {
 884                         cur_file = files[i];
 885
 886                         if (lstat(cur_file->d_name, &st) == -1) {
 887                                 error("lstat failed for %s: %s",
 888                                         cur_file->d_name, strerror(errno));
 889                                 ret = -1;
 890                                 goto fail;
 891                         }
 892
 893                         cur_inum = st.st_ino;
 894                         ret = add_directory_items(trans, root,
 895                                                   cur_inum, parent_inum,
 896                                                   cur_file->d_name,
 897                                                   &st, &dir_index_cnt);
 898                         if (ret) {
 899                                 error("unable to add directory items for %s: %d",
 900                                         cur_file->d_name, ret);
 901                                 goto fail;
 902                         }
 903
 904                         ret = add_inode_items(trans, root, &st,
 905                                               cur_file->d_name, cur_inum,
 906                                               &cur_inode);
 907                         if (ret == -EEXIST) {
 908                                 if (st.st_nlink <= 1) {
 909                                         error(
 910                         "item %s already exists but has wrong st_nlink %lu <= 1",
 911                                                 cur_file->d_name,
 912                                                 (unsigned long)st.st_nlink);
 913                                         goto fail;
 914                                 }
 915                                 continue;
 916                         }
 917                         if (ret) {
 918                                 error("unable to add inode items for %s: %d",
 919                                         cur_file->d_name, ret);
 920                                 goto fail;
 921                         }
 922
 923                         ret = add_xattr_item(trans, root,
 924                                              cur_inum, cur_file->d_name);
 925                         if (ret) {
 926                                 error("unable to add xattr items for %s: %d",
 927                                         cur_file->d_name, ret);
 928                                 if(ret != -ENOTSUP)
 929                                         goto fail;
 930                         }
 931
 932                         if (S_ISDIR(st.st_mode)) {
 933                                 dir_entry = malloc(sizeof(struct directory_name_entry));
 934                                 if (!dir_entry) {
 935                                         ret = -ENOMEM;
 936                                         goto fail;
 937                                 }
 938                                 dir_entry->dir_name = cur_file->d_name;
 939                                 dir_entry->path = make_path(parent_dir_entry->path,
 940                                                             cur_file->d_name);
 941                                 dir_entry->inum = cur_inum;
 942                                 list_add_tail(&dir_entry->list, &dir_head->list);
 943                         } else if (S_ISREG(st.st_mode)) {
 944                                 ret = add_file_items(trans, root, &cur_inode,
 945                                                      cur_inum, &st,
 946                                                      cur_file->d_name);
 947                                 if (ret) {
 948                                         error("unable to add file items for %s: %d",
 949                                                 cur_file->d_name, ret);
 950                                         goto fail;
 951                                 }
 952                         } else if (S_ISLNK(st.st_mode)) {
 953                                 ret = add_symbolic_link(trans, root,
 954                                                         cur_inum, cur_file->d_name);
 955                                 if (ret) {
 956                                         error("unable to add symlink for %s: %d",
 957                                                 cur_file->d_name, ret);
 958                                         goto fail;
 959                                 }
 960                         }
 961                 }
 962
 963                 free_namelist(files, count);
 964                 free(parent_dir_entry);
 965
 966                 index_cnt = 2;
 967
 968         } while (!list_empty(&dir_head->list));
 969
 970 out:
 971         return !!ret;
 972 fail:
 973         free_namelist(files, count);
 974 fail_no_files:
 975         free(parent_dir_entry);
 976         goto out;
 977 fail_no_dir:
 978         free(dir_entry);
 979         goto out;
 980 }
 981
 982 static int create_chunks(struct btrfs_trans_handle *trans,
 983                          struct btrfs_root *root, u64 num_of_meta_chunks,
 984                          u64 size_of_data,
 985                          struct mkfs_allocation *allocation)
 986 {
 987         u64 chunk_start;
 988         u64 chunk_size;
 989         u64 meta_type = BTRFS_BLOCK_GROUP_METADATA;
 990         u64 data_type = BTRFS_BLOCK_GROUP_DATA;
 991         u64 minimum_data_chunk_size = 8 * 1024 * 1024;
 992         u64 i;
 993         int ret;
 994
 995         for (i = 0; i < num_of_meta_chunks; i++) {
 996                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 997                                         &chunk_start, &chunk_size, meta_type);
 998                 if (ret)
 999                         return ret;
1000                 ret = btrfs_make_block_group(trans, root->fs_info->extent_root, 0,
1001                                              meta_type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1002                                              chunk_start, chunk_size);
1003                 allocation->metadata += chunk_size;
1004                 if (ret)
1005                         return ret;
1006                 set_extent_dirty(&root->fs_info->free_space_cache,
1007                                  chunk_start, chunk_start + chunk_size - 1);
1008         }
1009
1010         if (size_of_data < minimum_data_chunk_size)
1011                 size_of_data = minimum_data_chunk_size;
1012
1013         ret = btrfs_alloc_data_chunk(trans, root->fs_info->extent_root,
1014                                      &chunk_start, size_of_data, data_type, 0);
1015         if (ret)
1016                 return ret;
1017         ret = btrfs_make_block_group(trans, root->fs_info->extent_root, 0,
1018                                      data_type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1019                                      chunk_start, size_of_data);
1020         allocation->data += size_of_data;
1021         if (ret)
1022                 return ret;
1023         set_extent_dirty(&root->fs_info->free_space_cache,
1024                          chunk_start, chunk_start + size_of_data - 1);
1025         return ret;
1026 }
1027
1028 static int make_image(const char *source_dir, struct btrfs_root *root,
1029                 int out_fd)
1030 {
1031         int ret;
1032         struct btrfs_trans_handle *trans;
1033         struct stat root_st;
1034         struct directory_name_entry dir_head;
1035         struct directory_name_entry *dir_entry = NULL;
1036
1037         ret = lstat(source_dir, &root_st);
1038         if (ret) {
1039                 error("unable to lstat %s: %s", source_dir, strerror(errno));
1040                 ret = -errno;
1041                 goto out;
1042         }
1043
1044         INIT_LIST_HEAD(&dir_head.list);
1045
1046         trans = btrfs_start_transaction(root, 1);
1047         ret = traverse_directory(trans, root, source_dir, &dir_head, out_fd);
1048         if (ret) {
1049                 error("unable to traverse directory %s: %d", source_dir, ret);
1050                 goto fail;
1051         }
1052         ret = btrfs_commit_transaction(trans, root);
1053         if (ret) {
1054                 error("transaction commit failed: %d", ret);
1055                 goto out;
1056         }
1057
1058         if (verbose)
1059                 printf("Making image is completed.\n");
1060         return 0;
1061 fail:
1062         while (!list_empty(&dir_head.list)) {
1063                 dir_entry = list_entry(dir_head.list.next,
1064                                        struct directory_name_entry, list);
1065                 list_del(&dir_entry->list);
1066                 free(dir_entry);
1067         }
1068 out:
1069         return ret;
1070 }
1071
1072 /*
1073  * This ignores symlinks with unreadable targets and subdirs that can't
1074  * be read.  It's a best-effort to give a rough estimate of the size of
1075  * a subdir.  It doesn't guarantee that prepopulating btrfs from this
1076  * tree won't still run out of space.
1077  */
1078 static u64 global_total_size;
1079 static u64 fs_block_size;
1080 static int ftw_add_entry_size(const char *fpath, const struct stat *st,
1081                               int type)
1082 {
1083         if (type == FTW_F || type == FTW_D)
1084                 global_total_size += round_up(st->st_size, fs_block_size);
1085
1086         return 0;
1087 }
1088
1089 static u64 size_sourcedir(const char *dir_name, u64 sectorsize,
1090                           u64 *num_of_meta_chunks_ret, u64 *size_of_data_ret)
1091 {
1092         u64 dir_size = 0;
1093         u64 total_size = 0;
1094         int ret;
1095         u64 default_chunk_size = 8 * 1024 * 1024;       /* 8MB */
1096         u64 allocated_meta_size = 8 * 1024 * 1024;      /* 8MB */
1097         u64 allocated_total_size = 20 * 1024 * 1024;    /* 20MB */
1098         u64 num_of_meta_chunks = 0;
1099         u64 num_of_data_chunks = 0;
1100         u64 num_of_allocated_meta_chunks =
1101                         allocated_meta_size / default_chunk_size;
1102
1103         global_total_size = 0;
1104         fs_block_size = sectorsize;
1105         ret = ftw(dir_name, ftw_add_entry_size, 10);
1106         dir_size = global_total_size;
1107         if (ret < 0) {
1108                 error("ftw subdir walk of %s failed: %s", dir_name,
1109                         strerror(errno));
1110                 exit(1);
1111         }
1112
1113         num_of_data_chunks = (dir_size + default_chunk_size - 1) /
1114                 default_chunk_size;
1115
1116         num_of_meta_chunks = (dir_size / 2) / default_chunk_size;
1117         if (((dir_size / 2) % default_chunk_size) != 0)
1118                 num_of_meta_chunks++;
1119         if (num_of_meta_chunks <= num_of_allocated_meta_chunks)
1120                 num_of_meta_chunks = 0;
1121         else
1122                 num_of_meta_chunks -= num_of_allocated_meta_chunks;
1123
1124         total_size = allocated_total_size +
1125                      (num_of_data_chunks * default_chunk_size) +
1126                      (num_of_meta_chunks * default_chunk_size);
1127
1128         *num_of_meta_chunks_ret = num_of_meta_chunks;
1129         *size_of_data_ret = num_of_data_chunks * default_chunk_size;
1130         return total_size;
1131 }
1132
1133 static int zero_output_file(int out_fd, u64 size)
1134 {
1135         int loop_num;
1136         u64 location = 0;
1137         char buf[4096];
1138         int ret = 0, i;
1139         ssize_t written;
1140
1141         memset(buf, 0, 4096);
1142         loop_num = size / 4096;
1143         for (i = 0; i < loop_num; i++) {
1144                 written = pwrite64(out_fd, buf, 4096, location);
1145                 if (written != 4096)
1146                         ret = -EIO;
1147                 location += 4096;
1148         }
1149         return ret;
1150 }
1151
1152 static int is_ssd(const char *file)
1153 {
1154         blkid_probe probe;
1155         char wholedisk[PATH_MAX];
1156         char sysfs_path[PATH_MAX];
1157         dev_t devno;
1158         int fd;
1159         char rotational;
1160         int ret;
1161
1162         probe = blkid_new_probe_from_filename(file);
1163         if (!probe)
1164                 return 0;
1165
1166         /* Device number of this disk (possibly a partition) */
1167         devno = blkid_probe_get_devno(probe);
1168         if (!devno) {
1169                 blkid_free_probe(probe);
1170                 return 0;
1171         }
1172
1173         /* Get whole disk name (not full path) for this devno */
1174         ret = blkid_devno_to_wholedisk(devno,
1175                         wholedisk, sizeof(wholedisk), NULL);
1176         if (ret) {
1177                 blkid_free_probe(probe);
1178                 return 0;
1179         }
1180
1181         snprintf(sysfs_path, PATH_MAX, "/sys/block/%s/queue/rotational",
1182                  wholedisk);
1183
1184         blkid_free_probe(probe);
1185
1186         fd = open(sysfs_path, O_RDONLY);
1187         if (fd < 0) {
1188                 return 0;
1189         }
1190
1191         if (read(fd, &rotational, 1) < 1) {
1192                 close(fd);
1193                 return 0;
1194         }
1195         close(fd);
1196
1197         return rotational == '0';
1198 }
1199
1200 static int _cmp_device_by_id(void *priv, struct list_head *a,
1201                              struct list_head *b)
1202 {
1203         return list_entry(a, struct btrfs_device, dev_list)->devid -
1204                list_entry(b, struct btrfs_device, dev_list)->devid;
1205 }
1206
1207 static void list_all_devices(struct btrfs_root *root)
1208 {
1209         struct btrfs_fs_devices *fs_devices;
1210         struct btrfs_device *device;
1211         int number_of_devices = 0;
1212         u64 total_block_count = 0;
1213
1214         fs_devices = root->fs_info->fs_devices;
1215
1216         list_for_each_entry(device, &fs_devices->devices, dev_list)
1217                 number_of_devices++;
1218
1219         list_sort(NULL, &fs_devices->devices, _cmp_device_by_id);
1220
1221         printf("Number of devices:  %d\n", number_of_devices);
1222         /* printf("Total devices size: %10s\n", */
1223                 /* pretty_size(total_block_count)); */
1224         printf("Devices:\n");
1225         printf("   ID        SIZE  PATH\n");
1226         list_for_each_entry(device, &fs_devices->devices, dev_list) {
1227                 printf("  %3llu  %10s  %s\n",
1228                         device->devid,
1229                         pretty_size(device->total_bytes),
1230                         device->name);
1231                 total_block_count += device->total_bytes;
1232         }
1233
1234         printf("\n");
1235 }
1236
1237 static int is_temp_block_group(struct extent_buffer *node,
1238                                struct btrfs_block_group_item *bgi,
1239                                u64 data_profile, u64 meta_profile,
1240                                u64 sys_profile)
1241 {
1242         u64 flag = btrfs_disk_block_group_flags(node, bgi);
1243         u64 flag_type = flag & BTRFS_BLOCK_GROUP_TYPE_MASK;
1244         u64 flag_profile = flag & BTRFS_BLOCK_GROUP_PROFILE_MASK;
1245         u64 used = btrfs_disk_block_group_used(node, bgi);
1246
1247         /*
1248          * Chunks meets all the following conditions is a temp chunk
1249          * 1) Empty chunk
1250          * Temp chunk is always empty.
1251          *
1252          * 2) profile mismatch with mkfs profile.
1253          * Temp chunk is always in SINGLE
1254          *
1255          * 3) Size differs with mkfs_alloc
1256          * Special case for SINGLE/SINGLE btrfs.
1257          * In that case, temp data chunk and real data chunk are always empty.
1258          * So we need to use mkfs_alloc to be sure which chunk is the newly
1259          * allocated.
1260          *
1261          * Normally, new chunk size is equal to mkfs one (One chunk)
1262          * If it has multiple chunks, we just refuse to delete any one.
1263          * As they are all single, so no real problem will happen.
1264          * So only use condition 1) and 2) to judge them.
1265          */
1266         if (used != 0)
1267                 return 0;
1268         switch (flag_type) {
1269         case BTRFS_BLOCK_GROUP_DATA:
1270         case BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA:
1271                 data_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1272                 if (flag_profile != data_profile)
1273                         return 1;
1274                 break;
1275         case BTRFS_BLOCK_GROUP_METADATA:
1276                 meta_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1277                 if (flag_profile != meta_profile)
1278                         return 1;
1279                 break;
1280         case BTRFS_BLOCK_GROUP_SYSTEM:
1281                 sys_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1282                 if (flag_profile != sys_profile)
1283                         return 1;
1284                 break;
1285         }
1286         return 0;
1287 }
1288
1289 /* Note: if current is a block group, it will skip it anyway */
1290 static int next_block_group(struct btrfs_root *root,
1291                             struct btrfs_path *path)
1292 {
1293         struct btrfs_key key;
1294         int ret = 0;
1295
1296         while (1) {
1297                 ret = btrfs_next_item(root, path);
1298                 if (ret)
1299                         goto out;
1300
1301                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1302                 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
1303                         goto out;
1304         }
1305 out:
1306         return ret;
1307 }
1308
1309 /* This function will cleanup  */
1310 static int cleanup_temp_chunks(struct btrfs_fs_info *fs_info,
1311                                struct mkfs_allocation *alloc,
1312                                u64 data_profile, u64 meta_profile,
1313                                u64 sys_profile)
1314 {
1315         struct btrfs_trans_handle *trans = NULL;
1316         struct btrfs_block_group_item *bgi;
1317         struct btrfs_root *root = fs_info->extent_root;
1318         struct btrfs_key key;
1319         struct btrfs_key found_key;
1320         struct btrfs_path path;
1321         int ret = 0;
1322
1323         btrfs_init_path(&path);
1324         trans = btrfs_start_transaction(root, 1);
1325
1326         key.objectid = 0;
1327         key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
1328         key.offset = 0;
1329
1330         while (1) {
1331                 /*
1332                  * as the rest of the loop may modify the tree, we need to
1333                  * start a new search each time.
1334                  */
1335                 ret = btrfs_search_slot(trans, root, &key, &path, 0, 0);
1336                 if (ret < 0)
1337                         goto out;
1338
1339                 btrfs_item_key_to_cpu(path.nodes[0], &found_key,
1340                                       path.slots[0]);
1341                 if (found_key.objectid < key.objectid)
1342                         goto out;
1343                 if (found_key.type != BTRFS_BLOCK_GROUP_ITEM_KEY) {
1344                         ret = next_block_group(root, &path);
1345                         if (ret < 0)
1346                                 goto out;
1347                         if (ret > 0) {
1348                                 ret = 0;
1349                                 goto out;
1350                         }
1351                         btrfs_item_key_to_cpu(path.nodes[0], &found_key,
1352                                               path.slots[0]);
1353                 }
1354
1355                 bgi = btrfs_item_ptr(path.nodes[0], path.slots[0],
1356                                      struct btrfs_block_group_item);
1357                 if (is_temp_block_group(path.nodes[0], bgi,
1358                                         data_profile, meta_profile,
1359                                         sys_profile)) {
1360                         u64 flags = btrfs_disk_block_group_flags(path.nodes[0],
1361                                                              bgi);
1362
1363                         ret = btrfs_free_block_group(trans, fs_info,
1364                                         found_key.objectid, found_key.offset);
1365                         if (ret < 0)
1366                                 goto out;
1367
1368                         if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
1369                             BTRFS_BLOCK_GROUP_DATA)
1370                                 alloc->data -= found_key.offset;
1371                         else if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
1372                                  BTRFS_BLOCK_GROUP_METADATA)
1373                                 alloc->metadata -= found_key.offset;
1374                         else if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
1375                                  BTRFS_BLOCK_GROUP_SYSTEM)
1376                                 alloc->system -= found_key.offset;
1377                         else if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
1378                                  (BTRFS_BLOCK_GROUP_METADATA |
1379                                   BTRFS_BLOCK_GROUP_DATA))
1380                                 alloc->mixed -= found_key.offset;
1381                 }
1382                 btrfs_release_path(&path);
1383                 key.objectid = found_key.objectid + found_key.offset;
1384         }
1385 out:
1386         if (trans)
1387                 btrfs_commit_transaction(trans, root);
1388         btrfs_release_path(&path);
1389         return ret;
1390 }
1391
1392 int main(int argc, char **argv)
1393 {
1394         char *file;
1395         struct btrfs_root *root;
1396         struct btrfs_fs_info *fs_info;
1397         struct btrfs_trans_handle *trans;
1398         char *label = NULL;
1399         u64 block_count = 0;
1400         u64 dev_block_count = 0;
1401         u64 alloc_start = 0;
1402         u64 metadata_profile = 0;
1403         u64 data_profile = 0;
1404         u32 nodesize = max_t(u32, sysconf(_SC_PAGESIZE),
1405                         BTRFS_MKFS_DEFAULT_NODE_SIZE);
1406         u32 sectorsize = 4096;
1407         u32 stripesize = 4096;
1408         int zero_end = 1;
1409         int fd;
1410         int ret;
1411         int i;
1412         int mixed = 0;
1413         int nodesize_forced = 0;
1414         int data_profile_opt = 0;
1415         int metadata_profile_opt = 0;
1416         int discard = 1;
1417         int ssd = 0;
1418         int force_overwrite = 0;
1419         char *source_dir = NULL;
1420         int source_dir_set = 0;
1421         u64 num_of_meta_chunks = 0;
1422         u64 size_of_data = 0;
1423         u64 source_dir_size = 0;
1424         int dev_cnt = 0;
1425         int saved_optind;
1426         char fs_uuid[BTRFS_UUID_UNPARSED_SIZE] = { 0 };
1427         u64 features = BTRFS_MKFS_DEFAULT_FEATURES;
1428         struct mkfs_allocation allocation = { 0 };
1429         struct btrfs_mkfs_config mkfs_cfg;
1430
1431         while(1) {
1432                 int c;
1433                 static const struct option long_options[] = {
1434                         { "alloc-start", required_argument, NULL, 'A'},
1435                         { "byte-count", required_argument, NULL, 'b' },
1436                         { "force", no_argument, NULL, 'f' },
1437                         { "leafsize", required_argument, NULL, 'l' },
1438                         { "label", required_argument, NULL, 'L'},
1439                         { "metadata", required_argument, NULL, 'm' },
1440                         { "mixed", no_argument, NULL, 'M' },
1441                         { "nodesize", required_argument, NULL, 'n' },
1442                         { "sectorsize", required_argument, NULL, 's' },
1443                         { "data", required_argument, NULL, 'd' },
1444                         { "version", no_argument, NULL, 'V' },
1445                         { "rootdir", required_argument, NULL, 'r' },
1446                         { "nodiscard", no_argument, NULL, 'K' },
1447                         { "features", required_argument, NULL, 'O' },
1448                         { "uuid", required_argument, NULL, 'U' },
1449                         { "quiet", 0, NULL, 'q' },
1450                         { "help", no_argument, NULL, GETOPT_VAL_HELP },
1451                         { NULL, 0, NULL, 0}
1452                 };
1453
1454                 c = getopt_long(argc, argv, "A:b:fl:n:s:m:d:L:O:r:U:VMKq",
1455                                 long_options, NULL);
1456                 if (c < 0)
1457                         break;
1458                 switch(c) {
1459                         case 'A':
1460                                 alloc_start = parse_size(optarg);
1461                                 break;
1462                         case 'f':
1463                                 force_overwrite = 1;
1464                                 break;
1465                         case 'd':
1466                                 data_profile = parse_profile(optarg);
1467                                 data_profile_opt = 1;
1468                                 break;
1469                         case 'l':
1470                                 warning("--leafsize is deprecated, use --nodesize");
1471                                 /* fall through */
1472                         case 'n':
1473                                 nodesize = parse_size(optarg);
1474                                 nodesize_forced = 1;
1475                                 break;
1476                         case 'L':
1477                                 label = parse_label(optarg);
1478                                 break;
1479                         case 'm':
1480                                 metadata_profile = parse_profile(optarg);
1481                                 metadata_profile_opt = 1;
1482                                 break;
1483                         case 'M':
1484                                 mixed = 1;
1485                                 break;
1486                         case 'O': {
1487                                 char *orig = strdup(optarg);
1488                                 char *tmp = orig;
1489
1490                                 tmp = btrfs_parse_fs_features(tmp, &features);
1491                                 if (tmp) {
1492                                         error("unrecognized filesystem feature '%s'",
1493                                                         tmp);
1494                                         free(orig);
1495                                         exit(1);
1496                                 }
1497                                 free(orig);
1498                                 if (features & BTRFS_FEATURE_LIST_ALL) {
1499                                         btrfs_list_all_fs_features(0);
1500                                         exit(0);
1501                                 }
1502                                 break;
1503                                 }
1504                         case 's':
1505                                 sectorsize = parse_size(optarg);
1506                                 break;
1507                         case 'b':
1508                                 block_count = parse_size(optarg);
1509                                 zero_end = 0;
1510                                 break;
1511                         case 'V':
1512                                 printf("mkfs.btrfs, part of %s\n",
1513                                                 PACKAGE_STRING);
1514                                 exit(0);
1515                                 break;
1516                         case 'r':
1517                                 source_dir = optarg;
1518                                 source_dir_set = 1;
1519                                 break;
1520                         case 'U':
1521                                 strncpy(fs_uuid, optarg,
1522                                         BTRFS_UUID_UNPARSED_SIZE - 1);
1523                                 break;
1524                         case 'K':
1525                                 discard = 0;
1526                                 break;
1527                         case 'q':
1528                                 verbose = 0;
1529                                 break;
1530                         case GETOPT_VAL_HELP:
1531                         default:
1532                                 print_usage(c != GETOPT_VAL_HELP);
1533                 }
1534         }
1535
1536         if (verbose) {
1537                 printf("%s\n", PACKAGE_STRING);
1538                 printf("See %s for more information.\n\n", PACKAGE_URL);
1539         }
1540
1541         sectorsize = max(sectorsize, (u32)sysconf(_SC_PAGESIZE));
1542         stripesize = sectorsize;
1543         saved_optind = optind;
1544         dev_cnt = argc - optind;
1545         if (dev_cnt == 0)
1546                 print_usage(1);
1547
1548         if (source_dir_set && dev_cnt > 1) {
1549                 error("the option -r is limited to a single device");
1550                 exit(1);
1551         }
1552
1553         if (*fs_uuid) {
1554                 uuid_t dummy_uuid;
1555
1556                 if (uuid_parse(fs_uuid, dummy_uuid) != 0) {
1557                         error("could not parse UUID: %s", fs_uuid);
1558                         exit(1);
1559                 }
1560                 if (!test_uuid_unique(fs_uuid)) {
1561                         error("non-unique UUID: %s", fs_uuid);
1562                         exit(1);
1563                 }
1564         }
1565
1566         while (dev_cnt-- > 0) {
1567                 file = argv[optind++];
1568                 if (is_block_device(file) == 1)
1569                         if (test_dev_for_mkfs(file, force_overwrite))
1570                                 exit(1);
1571         }
1572
1573         optind = saved_optind;
1574         dev_cnt = argc - optind;
1575
1576         file = argv[optind++];
1577         ssd = is_ssd(file);
1578
1579         /*
1580         * Set default profiles according to number of added devices.
1581         * For mixed groups defaults are single/single.
1582         */
1583         if (!mixed) {
1584                 if (!metadata_profile_opt) {
1585                         if (dev_cnt == 1 && ssd && verbose)
1586                                 printf("Detected a SSD, turning off metadata "
1587                                 "duplication.  Mkfs with -m dup if you want to "
1588                                 "force metadata duplication.\n");
1589
1590                         metadata_profile = (dev_cnt > 1) ?
1591                                         BTRFS_BLOCK_GROUP_RAID1 : (ssd) ?
1592                                         0: BTRFS_BLOCK_GROUP_DUP;
1593                 }
1594                 if (!data_profile_opt) {
1595                         data_profile = (dev_cnt > 1) ?
1596                                 BTRFS_BLOCK_GROUP_RAID0 : 0; /* raid0 or single */
1597                 }
1598         } else {
1599                 u32 best_nodesize = max_t(u32, sysconf(_SC_PAGESIZE), sectorsize);
1600
1601                 if (metadata_profile_opt || data_profile_opt) {
1602                         if (metadata_profile != data_profile) {
1603                                 error(
1604         "with mixed block groups data and metadata profiles must be the same");
1605                                 exit(1);
1606                         }
1607                 }
1608
1609                 if (!nodesize_forced)
1610                         nodesize = best_nodesize;
1611         }
1612
1613         /*
1614          * FS features that can be set by other means than -O
1615          * just set the bit here
1616          */
1617         if (mixed)
1618                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS;
1619
1620         if ((data_profile | metadata_profile) &
1621             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
1622                 features |= BTRFS_FEATURE_INCOMPAT_RAID56;
1623         }
1624
1625         if (btrfs_check_nodesize(nodesize, sectorsize,
1626                                  features))
1627                 exit(1);
1628
1629         if (sectorsize < sizeof(struct btrfs_super_block)) {
1630                 error("sectorsize smaller than superblock: %u < %zu",
1631                                 sectorsize, sizeof(struct btrfs_super_block));
1632                 exit(1);
1633         }
1634
1635         /* Check device/block_count after the nodesize is determined */
1636         if (block_count && block_count < btrfs_min_dev_size(nodesize)) {
1637                 error("size %llu is too small to make a usable filesystem",
1638                         block_count);
1639                 error("minimum size for btrfs filesystem is %llu",
1640                         btrfs_min_dev_size(nodesize));
1641                 exit(1);
1642         }
1643         for (i = saved_optind; i < saved_optind + dev_cnt; i++) {
1644                 char *path;
1645
1646                 path = argv[i];
1647                 ret = test_minimum_size(path, nodesize);
1648                 if (ret < 0) {
1649                         error("failed to check size for %s: %s",
1650                                 path, strerror(-ret));
1651                         exit (1);
1652                 }
1653                 if (ret > 0) {
1654                         error("'%s' is too small to make a usable filesystem",
1655                                 path);
1656                         error("minimum size for each btrfs device is %llu",
1657                                 btrfs_min_dev_size(nodesize));
1658                         exit(1);
1659                 }
1660         }
1661         ret = test_num_disk_vs_raid(metadata_profile, data_profile,
1662                         dev_cnt, mixed, ssd);
1663         if (ret)
1664                 exit(1);
1665
1666         dev_cnt--;
1667
1668         if (!source_dir_set) {
1669                 /*
1670                  * open without O_EXCL so that the problem should not
1671                  * occur by the following processing.
1672                  * (btrfs_register_one_device() fails if O_EXCL is on)
1673                  */
1674                 fd = open(file, O_RDWR);
1675                 if (fd < 0) {
1676                         error("unable to open %s: %s", file, strerror(errno));
1677                         exit(1);
1678                 }
1679                 ret = btrfs_prepare_device(fd, file, &dev_block_count,
1680                                 block_count,
1681                                 (zero_end ? PREP_DEVICE_ZERO_END : 0) |
1682                                 (discard ? PREP_DEVICE_DISCARD : 0) |
1683                                 (verbose ? PREP_DEVICE_VERBOSE : 0));
1684                 if (ret) {
1685                         close(fd);
1686                         exit(1);
1687                 }
1688                 if (block_count && block_count > dev_block_count) {
1689                         error("%s is smaller than requested size, expected %llu, found %llu",
1690                                         file,
1691                                         (unsigned long long)block_count,
1692                                         (unsigned long long)dev_block_count);
1693                         exit(1);
1694                 }
1695         } else {
1696                 fd = open(file, O_CREAT | O_RDWR,
1697                                 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH);
1698                 if (fd < 0) {
1699                         error("unable to open %s: %s", file, strerror(errno));
1700                         exit(1);
1701                 }
1702
1703                 source_dir_size = size_sourcedir(source_dir, sectorsize,
1704                                              &num_of_meta_chunks, &size_of_data);
1705                 if(block_count < source_dir_size)
1706                         block_count = source_dir_size;
1707                 ret = zero_output_file(fd, block_count);
1708                 if (ret) {
1709                         error("unable to zero the output file");
1710                         exit(1);
1711                 }
1712                 /* our "device" is the new image file */
1713                 dev_block_count = block_count;
1714         }
1715
1716         /* To create the first block group and chunk 0 in make_btrfs */
1717         if (dev_block_count < BTRFS_MKFS_SYSTEM_GROUP_SIZE) {
1718                 error("device is too small to make filesystem, must be at least %llu",
1719                                 (unsigned long long)BTRFS_MKFS_SYSTEM_GROUP_SIZE);
1720                 exit(1);
1721         }
1722
1723         if (group_profile_max_safe_loss(metadata_profile) <
1724                 group_profile_max_safe_loss(data_profile)){
1725                 warning("metadata has lower redundancy than data!\n");
1726         }
1727
1728         mkfs_cfg.label = label;
1729         memcpy(mkfs_cfg.fs_uuid, fs_uuid, sizeof(mkfs_cfg.fs_uuid));
1730         mkfs_cfg.num_bytes = dev_block_count;
1731         mkfs_cfg.nodesize = nodesize;
1732         mkfs_cfg.sectorsize = sectorsize;
1733         mkfs_cfg.stripesize = stripesize;
1734         mkfs_cfg.features = features;
1735
1736         ret = make_btrfs(fd, &mkfs_cfg);
1737         if (ret) {
1738                 error("error during mkfs: %s", strerror(-ret));
1739                 exit(1);
1740         }
1741
1742         fs_info = open_ctree_fs_info(file, 0, 0, 0,
1743                         OPEN_CTREE_WRITES | OPEN_CTREE_FS_PARTIAL);
1744         if (!fs_info) {
1745                 error("open ctree failed");
1746                 close(fd);
1747                 exit(1);
1748         }
1749         root = fs_info->fs_root;
1750         fs_info->alloc_start = alloc_start;
1751
1752         ret = create_metadata_block_groups(root, mixed, &allocation);
1753         if (ret) {
1754                 error("failed to create default block groups: %d", ret);
1755                 exit(1);
1756         }
1757
1758         trans = btrfs_start_transaction(root, 1);
1759         if (!trans) {
1760                 error("failed to start transaction");
1761                 exit(1);
1762         }
1763
1764         ret = create_data_block_groups(trans, root, mixed, &allocation);
1765         if (ret) {
1766                 error("failed to create default data block groups: %d", ret);
1767                 exit(1);
1768         }
1769
1770         ret = make_root_dir(trans, root);
1771         if (ret) {
1772                 error("failed to setup the root directory: %d", ret);
1773                 exit(1);
1774         }
1775
1776         ret = btrfs_commit_transaction(trans, root);
1777         if (ret) {
1778                 error("unable to commit transaction: %d", ret);
1779                 goto out;
1780         }
1781
1782         trans = btrfs_start_transaction(root, 1);
1783         if (!trans) {
1784                 error("failed to start transaction");
1785                 exit(1);
1786         }
1787
1788         if (dev_cnt == 0)
1789                 goto raid_groups;
1790
1791         while (dev_cnt-- > 0) {
1792                 file = argv[optind++];
1793
1794                 /*
1795                  * open without O_EXCL so that the problem should not
1796                  * occur by the following processing.
1797                  * (btrfs_register_one_device() fails if O_EXCL is on)
1798                  */
1799                 fd = open(file, O_RDWR);
1800                 if (fd < 0) {
1801                         error("unable to open %s: %s", file, strerror(errno));
1802                         exit(1);
1803                 }
1804                 ret = btrfs_device_already_in_root(root, fd,
1805                                                    BTRFS_SUPER_INFO_OFFSET);
1806                 if (ret) {
1807                         error("skipping duplicate device %s in the filesystem",
1808                                 file);
1809                         close(fd);
1810                         continue;
1811                 }
1812                 ret = btrfs_prepare_device(fd, file, &dev_block_count,
1813                                 block_count,
1814                                 (verbose ? PREP_DEVICE_VERBOSE : 0) |
1815                                 (zero_end ? PREP_DEVICE_ZERO_END : 0) |
1816                                 (discard ? PREP_DEVICE_DISCARD : 0));
1817                 if (ret) {
1818                         close(fd);
1819                         exit(1);
1820                 }
1821
1822                 ret = btrfs_add_to_fsid(trans, root, fd, file, dev_block_count,
1823                                         sectorsize, sectorsize, sectorsize);
1824                 if (ret) {
1825                         error("unable to add %s to filesystem: %d", file, ret);
1826                         goto out;
1827                 }
1828                 if (verbose >= 2) {
1829                         struct btrfs_device *device;
1830
1831                         device = container_of(fs_info->fs_devices->devices.next,
1832                                         struct btrfs_device, dev_list);
1833                         printf("adding device %s id %llu\n", file,
1834                                 (unsigned long long)device->devid);
1835                 }
1836         }
1837
1838 raid_groups:
1839         if (!source_dir_set) {
1840                 ret = create_raid_groups(trans, root, data_profile,
1841                                  metadata_profile, mixed, &allocation);
1842                 if (ret) {
1843                         error("unable to create raid groups: %d", ret);
1844                         goto out;
1845                 }
1846         }
1847
1848         ret = create_data_reloc_tree(trans, root);
1849         if (ret) {
1850                 error("unable to create data reloc tree: %d", ret);
1851                 goto out;
1852         }
1853
1854         ret = btrfs_commit_transaction(trans, root);
1855         if (ret) {
1856                 error("unable to commit transaction: %d", ret);
1857                 goto out;
1858         }
1859
1860         if (source_dir_set) {
1861                 trans = btrfs_start_transaction(root, 1);
1862                 ret = create_chunks(trans, root,
1863                                     num_of_meta_chunks, size_of_data,
1864                                     &allocation);
1865                 if (ret) {
1866                         error("unable to create chunks: %d", ret);
1867                         goto out;
1868                 }
1869                 ret = btrfs_commit_transaction(trans, root);
1870                 if (ret) {
1871                         error("transaction commit failed: %d", ret);
1872                         goto out;
1873                 }
1874
1875                 ret = make_image(source_dir, root, fd);
1876                 if (ret) {
1877                         error("error wihle filling filesystem: %d", ret);
1878                         goto out;
1879                 }
1880         }
1881         ret = cleanup_temp_chunks(fs_info, &allocation, data_profile,
1882                                   metadata_profile, metadata_profile);
1883         if (ret < 0) {
1884                 error("failed to cleanup temporary chunks: %d", ret);
1885                 goto out;
1886         }
1887
1888         if (verbose) {
1889                 char features_buf[64];
1890
1891                 printf("Label:              %s\n", label);
1892                 printf("UUID:               %s\n", mkfs_cfg.fs_uuid);
1893                 printf("Node size:          %u\n", nodesize);
1894                 printf("Sector size:        %u\n", sectorsize);
1895                 printf("Filesystem size:    %s\n",
1896                         pretty_size(btrfs_super_total_bytes(fs_info->super_copy)));
1897                 printf("Block group profiles:\n");
1898                 if (allocation.data)
1899                         printf("  Data:             %-8s %16s\n",
1900                                 btrfs_group_profile_str(data_profile),
1901                                 pretty_size(allocation.data));
1902                 if (allocation.metadata)
1903                         printf("  Metadata:         %-8s %16s\n",
1904                                 btrfs_group_profile_str(metadata_profile),
1905                                 pretty_size(allocation.metadata));
1906                 if (allocation.mixed)
1907                         printf("  Data+Metadata:    %-8s %16s\n",
1908                                 btrfs_group_profile_str(data_profile),
1909                                 pretty_size(allocation.mixed));
1910                 printf("  System:           %-8s %16s\n",
1911                         btrfs_group_profile_str(metadata_profile),
1912                         pretty_size(allocation.system));
1913                 printf("SSD detected:       %s\n", ssd ? "yes" : "no");
1914                 btrfs_parse_features_to_string(features_buf, features);
1915                 printf("Incompat features:  %s", features_buf);
1916                 printf("\n");
1917
1918                 list_all_devices(root);
1919         }
1920
1921         /*
1922          * The filesystem is now fully set up, commit the remaining changes and
1923          * fix the signature as the last step before closing the devices.
1924          */
1925         fs_info->finalize_on_close = 1;
1926 out:
1927         ret = close_ctree(root);
1928
1929         if (!ret) {
1930                 optind = saved_optind;
1931                 dev_cnt = argc - optind;
1932                 while (dev_cnt-- > 0) {
1933                         file = argv[optind++];
1934                         if (is_block_device(file) == 1)
1935                                 btrfs_register_one_device(file);
1936                 }
1937         }
1938
1939         btrfs_close_all_devices();
1940         free(label);
1941
1942         return !!ret;
1943 }