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 int path_cat_out(char *out, const char *p1, const char *p2);
  51
  52 static u64 index_cnt = 2;
  53 static int verbose = 1;
  54
  55 struct directory_name_entry {
  56         const char *dir_name;
  57         const char *path;
  58         ino_t inum;
  59         struct list_head list;
  60 };
  61
  62 struct mkfs_allocation {
  63         u64 data;
  64         u64 metadata;
  65         u64 mixed;
  66         u64 system;
  67 };
  68
  69 static int create_metadata_block_groups(struct btrfs_root *root, int mixed,
  70                                 struct mkfs_allocation *allocation)
  71 {
  72         struct btrfs_fs_info *fs_info = root->fs_info;
  73         struct btrfs_trans_handle *trans;
  74         u64 bytes_used;
  75         u64 chunk_start = 0;
  76         u64 chunk_size = 0;
  77         int ret;
  78
  79         trans = btrfs_start_transaction(root, 1);
  80         BUG_ON(IS_ERR(trans));
  81         bytes_used = btrfs_super_bytes_used(fs_info->super_copy);
  82
  83         root->fs_info->system_allocs = 1;
  84         ret = btrfs_make_block_group(trans, fs_info, bytes_used,
  85                                      BTRFS_BLOCK_GROUP_SYSTEM,
  86                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
  87                                      0, BTRFS_MKFS_SYSTEM_GROUP_SIZE);
  88         allocation->system += BTRFS_MKFS_SYSTEM_GROUP_SIZE;
  89         if (ret)
  90                 return ret;
  91
  92         if (mixed) {
  93                 ret = btrfs_alloc_chunk(trans, fs_info,
  94                                         &chunk_start, &chunk_size,
  95                                         BTRFS_BLOCK_GROUP_METADATA |
  96                                         BTRFS_BLOCK_GROUP_DATA);
  97                 if (ret == -ENOSPC) {
  98                         error("no space to allocate data/metadata chunk");
  99                         goto err;
 100                 }
 101                 if (ret)
 102                         return ret;
 103                 ret = btrfs_make_block_group(trans, fs_info, 0,
 104                                              BTRFS_BLOCK_GROUP_METADATA |
 105                                              BTRFS_BLOCK_GROUP_DATA,
 106                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 107                                              chunk_start, chunk_size);
 108                 if (ret)
 109                         return ret;
 110                 allocation->mixed += chunk_size;
 111         } else {
 112                 ret = btrfs_alloc_chunk(trans, fs_info,
 113                                         &chunk_start, &chunk_size,
 114                                         BTRFS_BLOCK_GROUP_METADATA);
 115                 if (ret == -ENOSPC) {
 116                         error("no space to allocate metadata chunk");
 117                         goto err;
 118                 }
 119                 if (ret)
 120                         return ret;
 121                 ret = btrfs_make_block_group(trans, fs_info, 0,
 122                                              BTRFS_BLOCK_GROUP_METADATA,
 123                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 124                                              chunk_start, chunk_size);
 125                 allocation->metadata += chunk_size;
 126                 if (ret)
 127                         return ret;
 128         }
 129
 130         root->fs_info->system_allocs = 0;
 131         ret = btrfs_commit_transaction(trans, root);
 132
 133 err:
 134         return ret;
 135 }
 136
 137 static int create_data_block_groups(struct btrfs_trans_handle *trans,
 138                 struct btrfs_root *root, int mixed,
 139                 struct mkfs_allocation *allocation)
 140 {
 141         struct btrfs_fs_info *fs_info = root->fs_info;
 142         u64 chunk_start = 0;
 143         u64 chunk_size = 0;
 144         int ret = 0;
 145
 146         if (!mixed) {
 147                 ret = btrfs_alloc_chunk(trans, fs_info,
 148                                         &chunk_start, &chunk_size,
 149                                         BTRFS_BLOCK_GROUP_DATA);
 150                 if (ret == -ENOSPC) {
 151                         error("no space to allocate data chunk");
 152                         goto err;
 153                 }
 154                 if (ret)
 155                         return ret;
 156                 ret = btrfs_make_block_group(trans, fs_info, 0,
 157                                              BTRFS_BLOCK_GROUP_DATA,
 158                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 159                                              chunk_start, chunk_size);
 160                 allocation->data += chunk_size;
 161                 if (ret)
 162                         return ret;
 163         }
 164
 165 err:
 166         return ret;
 167 }
 168
 169 static int make_root_dir(struct btrfs_trans_handle *trans,
 170                 struct btrfs_root *root)
 171 {
 172         struct btrfs_key location;
 173         int ret;
 174
 175         ret = btrfs_make_root_dir(trans, root->fs_info->tree_root,
 176                               BTRFS_ROOT_TREE_DIR_OBJECTID);
 177         if (ret)
 178                 goto err;
 179         ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
 180         if (ret)
 181                 goto err;
 182         memcpy(&location, &root->fs_info->fs_root->root_key, sizeof(location));
 183         location.offset = (u64)-1;
 184         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
 185                         "default", 7,
 186                         btrfs_super_root_dir(root->fs_info->super_copy),
 187                         &location, BTRFS_FT_DIR, 0);
 188         if (ret)
 189                 goto err;
 190
 191         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 192                              "default", 7, location.objectid,
 193                              BTRFS_ROOT_TREE_DIR_OBJECTID, 0);
 194         if (ret)
 195                 goto err;
 196
 197 err:
 198         return ret;
 199 }
 200
 201 static int __recow_root(struct btrfs_trans_handle *trans,
 202                          struct btrfs_root *root)
 203 {
 204         struct extent_buffer *tmp;
 205         int ret;
 206
 207         if (trans->transid != btrfs_root_generation(&root->root_item)) {
 208                 extent_buffer_get(root->node);
 209                 ret = __btrfs_cow_block(trans, root, root->node,
 210                                         NULL, 0, &tmp, 0, 0);
 211                 if (ret)
 212                         return ret;
 213                 free_extent_buffer(tmp);
 214         }
 215
 216         return 0;
 217 }
 218
 219 static int recow_roots(struct btrfs_trans_handle *trans,
 220                        struct btrfs_root *root)
 221 {
 222         struct btrfs_fs_info *info = root->fs_info;
 223         int ret;
 224
 225         ret = __recow_root(trans, info->fs_root);
 226         if (ret)
 227                 return ret;
 228         ret = __recow_root(trans, info->tree_root);
 229         if (ret)
 230                 return ret;
 231         ret = __recow_root(trans, info->extent_root);
 232         if (ret)
 233                 return ret;
 234         ret = __recow_root(trans, info->chunk_root);
 235         if (ret)
 236                 return ret;
 237         ret = __recow_root(trans, info->dev_root);
 238         if (ret)
 239                 return ret;
 240         ret = __recow_root(trans, info->csum_root);
 241         if (ret)
 242                 return ret;
 243
 244         return 0;
 245 }
 246
 247 static int create_one_raid_group(struct btrfs_trans_handle *trans,
 248                               struct btrfs_root *root, u64 type,
 249                               struct mkfs_allocation *allocation)
 250
 251 {
 252         struct btrfs_fs_info *fs_info = root->fs_info;
 253         u64 chunk_start;
 254         u64 chunk_size;
 255         int ret;
 256
 257         ret = btrfs_alloc_chunk(trans, fs_info,
 258                                 &chunk_start, &chunk_size, type);
 259         if (ret == -ENOSPC) {
 260                 error("not enough free space to allocate chunk");
 261                 exit(1);
 262         }
 263         if (ret)
 264                 return ret;
 265
 266         ret = btrfs_make_block_group(trans, fs_info, 0,
 267                                      type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 268                                      chunk_start, chunk_size);
 269
 270         type &= BTRFS_BLOCK_GROUP_TYPE_MASK;
 271         if (type == BTRFS_BLOCK_GROUP_DATA) {
 272                 allocation->data += chunk_size;
 273         } else if (type == BTRFS_BLOCK_GROUP_METADATA) {
 274                 allocation->metadata += chunk_size;
 275         } else if (type == BTRFS_BLOCK_GROUP_SYSTEM) {
 276                 allocation->system += chunk_size;
 277         } else if (type ==
 278                         (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA)) {
 279                 allocation->mixed += chunk_size;
 280         } else {
 281                 error("unrecognized profile type: 0x%llx",
 282                                 (unsigned long long)type);
 283                 ret = -EINVAL;
 284         }
 285
 286         return ret;
 287 }
 288
 289 static int create_raid_groups(struct btrfs_trans_handle *trans,
 290                               struct btrfs_root *root, u64 data_profile,
 291                               u64 metadata_profile, int mixed,
 292                               struct mkfs_allocation *allocation)
 293 {
 294         int ret;
 295
 296         if (metadata_profile) {
 297                 u64 meta_flags = BTRFS_BLOCK_GROUP_METADATA;
 298
 299                 ret = create_one_raid_group(trans, root,
 300                                             BTRFS_BLOCK_GROUP_SYSTEM |
 301                                             metadata_profile, allocation);
 302                 if (ret)
 303                         return ret;
 304
 305                 if (mixed)
 306                         meta_flags |= BTRFS_BLOCK_GROUP_DATA;
 307
 308                 ret = create_one_raid_group(trans, root, meta_flags |
 309                                             metadata_profile, allocation);
 310                 if (ret)
 311                         return ret;
 312
 313         }
 314         if (!mixed && data_profile) {
 315                 ret = create_one_raid_group(trans, root,
 316                                             BTRFS_BLOCK_GROUP_DATA |
 317                                             data_profile, allocation);
 318                 if (ret)
 319                         return ret;
 320         }
 321         ret = recow_roots(trans, root);
 322
 323         return ret;
 324 }
 325
 326 static int create_tree(struct btrfs_trans_handle *trans,
 327                         struct btrfs_root *root, u64 objectid)
 328 {
 329         struct btrfs_key location;
 330         struct btrfs_root_item root_item;
 331         struct extent_buffer *tmp;
 332         int ret;
 333
 334         ret = btrfs_copy_root(trans, root, root->node, &tmp, objectid);
 335         if (ret)
 336                 return ret;
 337
 338         memcpy(&root_item, &root->root_item, sizeof(root_item));
 339         btrfs_set_root_bytenr(&root_item, tmp->start);
 340         btrfs_set_root_level(&root_item, btrfs_header_level(tmp));
 341         btrfs_set_root_generation(&root_item, trans->transid);
 342         free_extent_buffer(tmp);
 343
 344         location.objectid = objectid;
 345         location.type = BTRFS_ROOT_ITEM_KEY;
 346         location.offset = 0;
 347         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
 348                                 &location, &root_item);
 349
 350         return ret;
 351 }
 352
 353 static void print_usage(int ret)
 354 {
 355         printf("Usage: mkfs.btrfs [options] dev [ dev ... ]\n");
 356         printf("Options:\n");
 357         printf("  allocation profiles:\n");
 358         printf("\t-d|--data PROFILE       data profile, raid0, raid1, raid5, raid6, raid10, dup or single\n");
 359         printf("\t-m|--metadata PROFILE   metadata profile, values like for data profile\n");
 360         printf("\t-M|--mixed              mix metadata and data together\n");
 361         printf("  features:\n");
 362         printf("\t-n|--nodesize SIZE      size of btree nodes\n");
 363         printf("\t-s|--sectorsize SIZE    data block size (may not be mountable by current kernel)\n");
 364         printf("\t-O|--features LIST      comma separated list of filesystem features (use '-O list-all' to list features)\n");
 365         printf("\t-L|--label LABEL        set the filesystem label\n");
 366         printf("\t-U|--uuid UUID          specify the filesystem UUID (must be unique)\n");
 367         printf("  creation:\n");
 368         printf("\t-b|--byte-count SIZE    set filesystem size to SIZE (on the first device)\n");
 369         printf("\t-r|--rootdir DIR        copy files from DIR to the image root directory\n");
 370         printf("\t-K|--nodiscard          do not perform whole device TRIM\n");
 371         printf("\t-f|--force              force overwrite of existing filesystem\n");
 372         printf("  general:\n");
 373         printf("\t-q|--quiet              no messages except errors\n");
 374         printf("\t-V|--version            print the mkfs.btrfs version and exit\n");
 375         printf("\t--help                  print this help and exit\n");
 376         printf("  deprecated:\n");
 377         printf("\t-A|--alloc-start START  the offset to start the filesystem\n");
 378         printf("\t-l|--leafsize SIZE      deprecated, alias for nodesize\n");
 379         exit(ret);
 380 }
 381
 382 static u64 parse_profile(const char *s)
 383 {
 384         if (strcasecmp(s, "raid0") == 0) {
 385                 return BTRFS_BLOCK_GROUP_RAID0;
 386         } else if (strcasecmp(s, "raid1") == 0) {
 387                 return BTRFS_BLOCK_GROUP_RAID1;
 388         } else if (strcasecmp(s, "raid5") == 0) {
 389                 return BTRFS_BLOCK_GROUP_RAID5;
 390         } else if (strcasecmp(s, "raid6") == 0) {
 391                 return BTRFS_BLOCK_GROUP_RAID6;
 392         } else if (strcasecmp(s, "raid10") == 0) {
 393                 return BTRFS_BLOCK_GROUP_RAID10;
 394         } else if (strcasecmp(s, "dup") == 0) {
 395                 return BTRFS_BLOCK_GROUP_DUP;
 396         } else if (strcasecmp(s, "single") == 0) {
 397                 return 0;
 398         } else {
 399                 error("unknown profile %s", s);
 400                 exit(1);
 401         }
 402         /* not reached */
 403         return 0;
 404 }
 405
 406 static char *parse_label(const char *input)
 407 {
 408         int len = strlen(input);
 409
 410         if (len >= BTRFS_LABEL_SIZE) {
 411                 error("label %s is too long (max %d)", input,
 412                         BTRFS_LABEL_SIZE - 1);
 413                 exit(1);
 414         }
 415         return strdup(input);
 416 }
 417
 418 static int add_directory_items(struct btrfs_trans_handle *trans,
 419                                struct btrfs_root *root, u64 objectid,
 420                                ino_t parent_inum, const char *name,
 421                                struct stat *st, int *dir_index_cnt)
 422 {
 423         int ret;
 424         int name_len;
 425         struct btrfs_key location;
 426         u8 filetype = 0;
 427
 428         name_len = strlen(name);
 429
 430         location.objectid = objectid;
 431         location.offset = 0;
 432         location.type = BTRFS_INODE_ITEM_KEY;
 433
 434         if (S_ISDIR(st->st_mode))
 435                 filetype = BTRFS_FT_DIR;
 436         if (S_ISREG(st->st_mode))
 437                 filetype = BTRFS_FT_REG_FILE;
 438         if (S_ISLNK(st->st_mode))
 439                 filetype = BTRFS_FT_SYMLINK;
 440         if (S_ISSOCK(st->st_mode))
 441                 filetype = BTRFS_FT_SOCK;
 442         if (S_ISCHR(st->st_mode))
 443                 filetype = BTRFS_FT_CHRDEV;
 444         if (S_ISBLK(st->st_mode))
 445                 filetype = BTRFS_FT_BLKDEV;
 446         if (S_ISFIFO(st->st_mode))
 447                 filetype = BTRFS_FT_FIFO;
 448
 449         ret = btrfs_insert_dir_item(trans, root, name, name_len,
 450                                     parent_inum, &location,
 451                                     filetype, index_cnt);
 452         if (ret)
 453                 return ret;
 454         ret = btrfs_insert_inode_ref(trans, root, name, name_len,
 455                                      objectid, parent_inum, index_cnt);
 456         *dir_index_cnt = index_cnt;
 457         index_cnt++;
 458
 459         return ret;
 460 }
 461
 462 static int fill_inode_item(struct btrfs_trans_handle *trans,
 463                            struct btrfs_root *root,
 464                            struct btrfs_inode_item *dst, struct stat *src)
 465 {
 466         u64 blocks = 0;
 467         u64 sectorsize = root->fs_info->sectorsize;
 468
 469         /*
 470          * btrfs_inode_item has some reserved fields
 471          * and represents on-disk inode entry, so
 472          * zero everything to prevent information leak
 473          */
 474         memset(dst, 0, sizeof (*dst));
 475
 476         btrfs_set_stack_inode_generation(dst, trans->transid);
 477         btrfs_set_stack_inode_size(dst, src->st_size);
 478         btrfs_set_stack_inode_nbytes(dst, 0);
 479         btrfs_set_stack_inode_block_group(dst, 0);
 480         btrfs_set_stack_inode_nlink(dst, src->st_nlink);
 481         btrfs_set_stack_inode_uid(dst, src->st_uid);
 482         btrfs_set_stack_inode_gid(dst, src->st_gid);
 483         btrfs_set_stack_inode_mode(dst, src->st_mode);
 484         btrfs_set_stack_inode_rdev(dst, 0);
 485         btrfs_set_stack_inode_flags(dst, 0);
 486         btrfs_set_stack_timespec_sec(&dst->atime, src->st_atime);
 487         btrfs_set_stack_timespec_nsec(&dst->atime, 0);
 488         btrfs_set_stack_timespec_sec(&dst->ctime, src->st_ctime);
 489         btrfs_set_stack_timespec_nsec(&dst->ctime, 0);
 490         btrfs_set_stack_timespec_sec(&dst->mtime, src->st_mtime);
 491         btrfs_set_stack_timespec_nsec(&dst->mtime, 0);
 492         btrfs_set_stack_timespec_sec(&dst->otime, 0);
 493         btrfs_set_stack_timespec_nsec(&dst->otime, 0);
 494
 495         if (S_ISDIR(src->st_mode)) {
 496                 btrfs_set_stack_inode_size(dst, 0);
 497                 btrfs_set_stack_inode_nlink(dst, 1);
 498         }
 499         if (S_ISREG(src->st_mode)) {
 500                 btrfs_set_stack_inode_size(dst, (u64)src->st_size);
 501                 if (src->st_size <= BTRFS_MAX_INLINE_DATA_SIZE(root))
 502                         btrfs_set_stack_inode_nbytes(dst, src->st_size);
 503                 else {
 504                         blocks = src->st_size / sectorsize;
 505                         if (src->st_size % sectorsize)
 506                                 blocks += 1;
 507                         blocks *= sectorsize;
 508                         btrfs_set_stack_inode_nbytes(dst, blocks);
 509                 }
 510         }
 511         if (S_ISLNK(src->st_mode))
 512                 btrfs_set_stack_inode_nbytes(dst, src->st_size + 1);
 513
 514         return 0;
 515 }
 516
 517 static int directory_select(const struct direct *entry)
 518 {
 519         if (entry->d_name[0] == '.' &&
 520                 (entry->d_name[1] == 0 ||
 521                  (entry->d_name[1] == '.' && entry->d_name[2] == 0)))
 522                 return 0;
 523         return 1;
 524 }
 525
 526 static void free_namelist(struct direct **files, int count)
 527 {
 528         int i;
 529
 530         if (count < 0)
 531                 return;
 532
 533         for (i = 0; i < count; ++i)
 534                 free(files[i]);
 535         free(files);
 536 }
 537
 538 static u64 calculate_dir_inode_size(const char *dirname)
 539 {
 540         int count, i;
 541         struct direct **files, *cur_file;
 542         u64 dir_inode_size = 0;
 543
 544         count = scandir(dirname, &files, directory_select, NULL);
 545
 546         for (i = 0; i < count; i++) {
 547                 cur_file = files[i];
 548                 dir_inode_size += strlen(cur_file->d_name);
 549         }
 550
 551         free_namelist(files, count);
 552
 553         dir_inode_size *= 2;
 554         return dir_inode_size;
 555 }
 556
 557 static int add_inode_items(struct btrfs_trans_handle *trans,
 558                            struct btrfs_root *root,
 559                            struct stat *st, const char *name,
 560                            u64 self_objectid,
 561                            struct btrfs_inode_item *inode_ret)
 562 {
 563         int ret;
 564         struct btrfs_inode_item btrfs_inode;
 565         u64 objectid;
 566         u64 inode_size = 0;
 567
 568         fill_inode_item(trans, root, &btrfs_inode, st);
 569         objectid = self_objectid;
 570
 571         if (S_ISDIR(st->st_mode)) {
 572                 inode_size = calculate_dir_inode_size(name);
 573                 btrfs_set_stack_inode_size(&btrfs_inode, inode_size);
 574         }
 575
 576         ret = btrfs_insert_inode(trans, root, objectid, &btrfs_inode);
 577
 578         *inode_ret = btrfs_inode;
 579         return ret;
 580 }
 581
 582 static int add_xattr_item(struct btrfs_trans_handle *trans,
 583                           struct btrfs_root *root, u64 objectid,
 584                           const char *file_name)
 585 {
 586         int ret;
 587         int cur_name_len;
 588         char xattr_list[XATTR_LIST_MAX];
 589         char *cur_name;
 590         char cur_value[XATTR_SIZE_MAX];
 591         char delimiter = '\0';
 592         char *next_location = xattr_list;
 593
 594         ret = llistxattr(file_name, xattr_list, XATTR_LIST_MAX);
 595         if (ret < 0) {
 596                 if(errno == ENOTSUP)
 597                         return 0;
 598                 error("getting a list of xattr failed for %s: %s", file_name,
 599                                 strerror(errno));
 600                 return ret;
 601         }
 602         if (ret == 0)
 603                 return ret;
 604
 605         cur_name = strtok(xattr_list, &delimiter);
 606         while (cur_name != NULL) {
 607                 cur_name_len = strlen(cur_name);
 608                 next_location += cur_name_len + 1;
 609
 610                 ret = getxattr(file_name, cur_name, cur_value, XATTR_SIZE_MAX);
 611                 if (ret < 0) {
 612                         if(errno == ENOTSUP)
 613                                 return 0;
 614                         error("gettig a xattr value failed for %s attr %s: %s",
 615                                 file_name, cur_name, strerror(errno));
 616                         return ret;
 617                 }
 618
 619                 ret = btrfs_insert_xattr_item(trans, root, cur_name,
 620                                               cur_name_len, cur_value,
 621                                               ret, objectid);
 622                 if (ret) {
 623                         error("inserting a xattr item failed for %s: %s",
 624                                         file_name, strerror(-ret));
 625                 }
 626
 627                 cur_name = strtok(next_location, &delimiter);
 628         }
 629
 630         return ret;
 631 }
 632
 633 static int add_symbolic_link(struct btrfs_trans_handle *trans,
 634                              struct btrfs_root *root,
 635                              u64 objectid, const char *path_name)
 636 {
 637         int ret;
 638         char buf[PATH_MAX];
 639
 640         ret = readlink(path_name, buf, sizeof(buf));
 641         if (ret <= 0) {
 642                 error("readlink failed for %s: %s", path_name, strerror(errno));
 643                 goto fail;
 644         }
 645         if (ret >= sizeof(buf)) {
 646                 error("symlink too long for %s", path_name);
 647                 ret = -1;
 648                 goto fail;
 649         }
 650
 651         buf[ret] = '\0'; /* readlink does not do it for us */
 652         ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
 653                                          buf, ret + 1);
 654 fail:
 655         return ret;
 656 }
 657
 658 static int add_file_items(struct btrfs_trans_handle *trans,
 659                           struct btrfs_root *root,
 660                           struct btrfs_inode_item *btrfs_inode, u64 objectid,
 661                           struct stat *st, const char *path_name)
 662 {
 663         int ret = -1;
 664         ssize_t ret_read;
 665         u64 bytes_read = 0;
 666         struct btrfs_key key;
 667         int blocks;
 668         u32 sectorsize = root->fs_info->sectorsize;
 669         u64 first_block = 0;
 670         u64 file_pos = 0;
 671         u64 cur_bytes;
 672         u64 total_bytes;
 673         struct extent_buffer *eb = NULL;
 674         int fd;
 675
 676         if (st->st_size == 0)
 677                 return 0;
 678
 679         fd = open(path_name, O_RDONLY);
 680         if (fd == -1) {
 681                 error("cannot open %s: %s", path_name, strerror(errno));
 682                 return ret;
 683         }
 684
 685         blocks = st->st_size / sectorsize;
 686         if (st->st_size % sectorsize)
 687                 blocks += 1;
 688
 689         if (st->st_size <= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 690                 char *buffer = malloc(st->st_size);
 691
 692                 if (!buffer) {
 693                         ret = -ENOMEM;
 694                         goto end;
 695                 }
 696
 697                 ret_read = pread64(fd, buffer, st->st_size, bytes_read);
 698                 if (ret_read == -1) {
 699                         error("cannot read %s at offset %llu length %llu: %s",
 700                                 path_name, (unsigned long long)bytes_read,
 701                                 (unsigned long long)st->st_size,
 702                                 strerror(errno));
 703                         free(buffer);
 704                         goto end;
 705                 }
 706
 707                 ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
 708                                                  buffer, st->st_size);
 709                 free(buffer);
 710                 goto end;
 711         }
 712
 713         /* round up our st_size to the FS blocksize */
 714         total_bytes = (u64)blocks * sectorsize;
 715
 716         /*
 717          * do our IO in extent buffers so it can work
 718          * against any raid type
 719          */
 720         eb = calloc(1, sizeof(*eb) + sectorsize);
 721         if (!eb) {
 722                 ret = -ENOMEM;
 723                 goto end;
 724         }
 725
 726 again:
 727
 728         /*
 729          * keep our extent size at 1MB max, this makes it easier to work inside
 730          * the tiny block groups created during mkfs
 731          */
 732         cur_bytes = min(total_bytes, (u64)SZ_1M);
 733         ret = btrfs_reserve_extent(trans, root, cur_bytes, 0, 0, (u64)-1,
 734                                    &key, 1);
 735         if (ret)
 736                 goto end;
 737
 738         first_block = key.objectid;
 739         bytes_read = 0;
 740
 741         while (bytes_read < cur_bytes) {
 742
 743                 memset(eb->data, 0, sectorsize);
 744
 745                 ret_read = pread64(fd, eb->data, sectorsize, file_pos + bytes_read);
 746                 if (ret_read == -1) {
 747                         error("cannot read %s at offset %llu length %llu: %s",
 748                                 path_name,
 749                                 (unsigned long long)file_pos + bytes_read,
 750                                 (unsigned long long)sectorsize,
 751                                 strerror(errno));
 752                         goto end;
 753                 }
 754
 755                 eb->start = first_block + bytes_read;
 756                 eb->len = sectorsize;
 757
 758                 /*
 759                  * we're doing the csum before we record the extent, but
 760                  * that's ok
 761                  */
 762                 ret = btrfs_csum_file_block(trans, root->fs_info->csum_root,
 763                                             first_block + bytes_read + sectorsize,
 764                                             first_block + bytes_read,
 765                                             eb->data, sectorsize);
 766                 if (ret)
 767                         goto end;
 768
 769                 ret = write_and_map_eb(root->fs_info, eb);
 770                 if (ret) {
 771                         error("failed to write %s", path_name);
 772                         goto end;
 773                 }
 774
 775                 bytes_read += sectorsize;
 776         }
 777
 778         if (bytes_read) {
 779                 ret = btrfs_record_file_extent(trans, root, objectid, btrfs_inode,
 780                                                file_pos, first_block, cur_bytes);
 781                 if (ret)
 782                         goto end;
 783
 784         }
 785
 786         file_pos += cur_bytes;
 787         total_bytes -= cur_bytes;
 788
 789         if (total_bytes)
 790                 goto again;
 791
 792 end:
 793         free(eb);
 794         close(fd);
 795         return ret;
 796 }
 797
 798 static int traverse_directory(struct btrfs_trans_handle *trans,
 799                               struct btrfs_root *root, const char *dir_name,
 800                               struct directory_name_entry *dir_head)
 801 {
 802         int ret = 0;
 803
 804         struct btrfs_inode_item cur_inode;
 805         struct btrfs_inode_item *inode_item;
 806         int count, i, dir_index_cnt;
 807         struct direct **files;
 808         struct stat st;
 809         struct directory_name_entry *dir_entry, *parent_dir_entry;
 810         struct direct *cur_file;
 811         ino_t parent_inum, cur_inum;
 812         ino_t highest_inum = 0;
 813         const char *parent_dir_name;
 814         char real_path[PATH_MAX];
 815         struct btrfs_path path;
 816         struct extent_buffer *leaf;
 817         struct btrfs_key root_dir_key;
 818         u64 root_dir_inode_size = 0;
 819
 820         /* Add list for source directory */
 821         dir_entry = malloc(sizeof(struct directory_name_entry));
 822         if (!dir_entry)
 823                 return -ENOMEM;
 824         dir_entry->dir_name = dir_name;
 825         dir_entry->path = realpath(dir_name, real_path);
 826         if (!dir_entry->path) {
 827                 error("realpath  failed for %s: %s", dir_name, strerror(errno));
 828                 ret = -1;
 829                 goto fail_no_dir;
 830         }
 831
 832         parent_inum = highest_inum + BTRFS_FIRST_FREE_OBJECTID;
 833         dir_entry->inum = parent_inum;
 834         list_add_tail(&dir_entry->list, &dir_head->list);
 835
 836         btrfs_init_path(&path);
 837
 838         root_dir_key.objectid = btrfs_root_dirid(&root->root_item);
 839         root_dir_key.offset = 0;
 840         root_dir_key.type = BTRFS_INODE_ITEM_KEY;
 841         ret = btrfs_lookup_inode(trans, root, &path, &root_dir_key, 1);
 842         if (ret) {
 843                 error("failed to lookup root dir: %d", ret);
 844                 goto fail_no_dir;
 845         }
 846
 847         leaf = path.nodes[0];
 848         inode_item = btrfs_item_ptr(leaf, path.slots[0],
 849                                     struct btrfs_inode_item);
 850
 851         root_dir_inode_size = calculate_dir_inode_size(dir_name);
 852         btrfs_set_inode_size(leaf, inode_item, root_dir_inode_size);
 853         btrfs_mark_buffer_dirty(leaf);
 854
 855         btrfs_release_path(&path);
 856
 857         do {
 858                 parent_dir_entry = list_entry(dir_head->list.next,
 859                                               struct directory_name_entry,
 860                                               list);
 861                 list_del(&parent_dir_entry->list);
 862
 863                 parent_inum = parent_dir_entry->inum;
 864                 parent_dir_name = parent_dir_entry->dir_name;
 865                 if (chdir(parent_dir_entry->path)) {
 866                         error("chdir failed for %s: %s",
 867                                 parent_dir_name, strerror(errno));
 868                         ret = -1;
 869                         goto fail_no_files;
 870                 }
 871
 872                 count = scandir(parent_dir_entry->path, &files,
 873                                 directory_select, NULL);
 874                 if (count == -1)
 875                 {
 876                         error("scandir failed for %s: %s",
 877                                 parent_dir_name, strerror (errno));
 878                         ret = -1;
 879                         goto fail;
 880                 }
 881
 882                 for (i = 0; i < count; i++) {
 883                         cur_file = files[i];
 884
 885                         if (lstat(cur_file->d_name, &st) == -1) {
 886                                 error("lstat failed for %s: %s",
 887                                         cur_file->d_name, strerror(errno));
 888                                 ret = -1;
 889                                 goto fail;
 890                         }
 891
 892                         cur_inum = st.st_ino;
 893                         ret = add_directory_items(trans, root,
 894                                                   cur_inum, parent_inum,
 895                                                   cur_file->d_name,
 896                                                   &st, &dir_index_cnt);
 897                         if (ret) {
 898                                 error("unable to add directory items for %s: %d",
 899                                         cur_file->d_name, ret);
 900                                 goto fail;
 901                         }
 902
 903                         ret = add_inode_items(trans, root, &st,
 904                                               cur_file->d_name, cur_inum,
 905                                               &cur_inode);
 906                         if (ret == -EEXIST) {
 907                                 if (st.st_nlink <= 1) {
 908                                         error(
 909                         "item %s already exists but has wrong st_nlink %lu <= 1",
 910                                                 cur_file->d_name,
 911                                                 (unsigned long)st.st_nlink);
 912                                         goto fail;
 913                                 }
 914                                 continue;
 915                         }
 916                         if (ret) {
 917                                 error("unable to add inode items for %s: %d",
 918                                         cur_file->d_name, ret);
 919                                 goto fail;
 920                         }
 921
 922                         ret = add_xattr_item(trans, root,
 923                                              cur_inum, cur_file->d_name);
 924                         if (ret) {
 925                                 error("unable to add xattr items for %s: %d",
 926                                         cur_file->d_name, ret);
 927                                 if(ret != -ENOTSUP)
 928                                         goto fail;
 929                         }
 930
 931                         if (S_ISDIR(st.st_mode)) {
 932                                 char tmp[PATH_MAX];
 933
 934                                 dir_entry = malloc(sizeof(struct directory_name_entry));
 935                                 if (!dir_entry) {
 936                                         ret = -ENOMEM;
 937                                         goto fail;
 938                                 }
 939                                 dir_entry->dir_name = cur_file->d_name;
 940                                 if (path_cat_out(tmp, parent_dir_entry->path,
 941                                                         cur_file->d_name)) {
 942                                         error("invalid path: %s/%s",
 943                                                         parent_dir_entry->path,
 944                                                         cur_file->d_name);
 945                                         ret = -EINVAL;
 946                                         goto fail;
 947                                 }
 948                                 dir_entry->path = strdup(tmp);
 949                                 if (!dir_entry->path) {
 950                                         error("not enough memory to store path");
 951                                         ret = -ENOMEM;
 952                                         goto fail;
 953                                 }
 954                                 dir_entry->inum = cur_inum;
 955                                 list_add_tail(&dir_entry->list, &dir_head->list);
 956                         } else if (S_ISREG(st.st_mode)) {
 957                                 ret = add_file_items(trans, root, &cur_inode,
 958                                                      cur_inum, &st,
 959                                                      cur_file->d_name);
 960                                 if (ret) {
 961                                         error("unable to add file items for %s: %d",
 962                                                 cur_file->d_name, ret);
 963                                         goto fail;
 964                                 }
 965                         } else if (S_ISLNK(st.st_mode)) {
 966                                 ret = add_symbolic_link(trans, root,
 967                                                         cur_inum, cur_file->d_name);
 968                                 if (ret) {
 969                                         error("unable to add symlink for %s: %d",
 970                                                 cur_file->d_name, ret);
 971                                         goto fail;
 972                                 }
 973                         }
 974                 }
 975
 976                 free_namelist(files, count);
 977                 free(parent_dir_entry);
 978
 979                 index_cnt = 2;
 980
 981         } while (!list_empty(&dir_head->list));
 982
 983 out:
 984         return !!ret;
 985 fail:
 986         free_namelist(files, count);
 987 fail_no_files:
 988         free(parent_dir_entry);
 989         goto out;
 990 fail_no_dir:
 991         free(dir_entry);
 992         goto out;
 993 }
 994
 995 static int create_chunks(struct btrfs_trans_handle *trans,
 996                          struct btrfs_root *root, u64 num_of_meta_chunks,
 997                          u64 size_of_data,
 998                          struct mkfs_allocation *allocation)
 999 {
1000         struct btrfs_fs_info *fs_info = root->fs_info;
1001         u64 chunk_start;
1002         u64 chunk_size;
1003         u64 meta_type = BTRFS_BLOCK_GROUP_METADATA;
1004         u64 data_type = BTRFS_BLOCK_GROUP_DATA;
1005         u64 minimum_data_chunk_size = SZ_8M;
1006         u64 i;
1007         int ret;
1008
1009         for (i = 0; i < num_of_meta_chunks; i++) {
1010                 ret = btrfs_alloc_chunk(trans, fs_info,
1011                                         &chunk_start, &chunk_size, meta_type);
1012                 if (ret)
1013                         return ret;
1014                 ret = btrfs_make_block_group(trans, fs_info, 0,
1015                                              meta_type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1016                                              chunk_start, chunk_size);
1017                 allocation->metadata += chunk_size;
1018                 if (ret)
1019                         return ret;
1020                 set_extent_dirty(&root->fs_info->free_space_cache,
1021                                  chunk_start, chunk_start + chunk_size - 1);
1022         }
1023
1024         if (size_of_data < minimum_data_chunk_size)
1025                 size_of_data = minimum_data_chunk_size;
1026
1027         ret = btrfs_alloc_data_chunk(trans, fs_info,
1028                                      &chunk_start, size_of_data, data_type, 0);
1029         if (ret)
1030                 return ret;
1031         ret = btrfs_make_block_group(trans, fs_info, 0,
1032                                      data_type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1033                                      chunk_start, size_of_data);
1034         allocation->data += size_of_data;
1035         if (ret)
1036                 return ret;
1037         set_extent_dirty(&root->fs_info->free_space_cache,
1038                          chunk_start, chunk_start + size_of_data - 1);
1039         return ret;
1040 }
1041
1042 static int make_image(const char *source_dir, struct btrfs_root *root)
1043 {
1044         int ret;
1045         struct btrfs_trans_handle *trans;
1046         struct stat root_st;
1047         struct directory_name_entry dir_head;
1048         struct directory_name_entry *dir_entry = NULL;
1049
1050         ret = lstat(source_dir, &root_st);
1051         if (ret) {
1052                 error("unable to lstat %s: %s", source_dir, strerror(errno));
1053                 ret = -errno;
1054                 goto out;
1055         }
1056
1057         INIT_LIST_HEAD(&dir_head.list);
1058
1059         trans = btrfs_start_transaction(root, 1);
1060         BUG_ON(IS_ERR(trans));
1061         ret = traverse_directory(trans, root, source_dir, &dir_head);
1062         if (ret) {
1063                 error("unable to traverse directory %s: %d", source_dir, ret);
1064                 goto fail;
1065         }
1066         ret = btrfs_commit_transaction(trans, root);
1067         if (ret) {
1068                 error("transaction commit failed: %d", ret);
1069                 goto out;
1070         }
1071
1072         if (verbose)
1073                 printf("Making image is completed.\n");
1074         return 0;
1075 fail:
1076         while (!list_empty(&dir_head.list)) {
1077                 dir_entry = list_entry(dir_head.list.next,
1078                                        struct directory_name_entry, list);
1079                 list_del(&dir_entry->list);
1080                 free(dir_entry);
1081         }
1082 out:
1083         return ret;
1084 }
1085
1086 /*
1087  * This ignores symlinks with unreadable targets and subdirs that can't
1088  * be read.  It's a best-effort to give a rough estimate of the size of
1089  * a subdir.  It doesn't guarantee that prepopulating btrfs from this
1090  * tree won't still run out of space.
1091  */
1092 static u64 global_total_size;
1093 static u64 fs_block_size;
1094 static int ftw_add_entry_size(const char *fpath, const struct stat *st,
1095                               int type)
1096 {
1097         if (type == FTW_F || type == FTW_D)
1098                 global_total_size += round_up(st->st_size, fs_block_size);
1099
1100         return 0;
1101 }
1102
1103 static u64 size_sourcedir(const char *dir_name, u64 sectorsize,
1104                           u64 *num_of_meta_chunks_ret, u64 *size_of_data_ret)
1105 {
1106         u64 dir_size = 0;
1107         u64 total_size = 0;
1108         int ret;
1109         u64 default_chunk_size = SZ_8M;
1110         u64 allocated_meta_size = SZ_8M;
1111         u64 allocated_total_size = 20 * SZ_1M;  /* 20MB */
1112         u64 num_of_meta_chunks = 0;
1113         u64 num_of_data_chunks = 0;
1114         u64 num_of_allocated_meta_chunks =
1115                         allocated_meta_size / default_chunk_size;
1116
1117         global_total_size = 0;
1118         fs_block_size = sectorsize;
1119         ret = ftw(dir_name, ftw_add_entry_size, 10);
1120         dir_size = global_total_size;
1121         if (ret < 0) {
1122                 error("ftw subdir walk of %s failed: %s", dir_name,
1123                         strerror(errno));
1124                 exit(1);
1125         }
1126
1127         num_of_data_chunks = (dir_size + default_chunk_size - 1) /
1128                 default_chunk_size;
1129
1130         num_of_meta_chunks = (dir_size / 2) / default_chunk_size;
1131         if (((dir_size / 2) % default_chunk_size) != 0)
1132                 num_of_meta_chunks++;
1133         if (num_of_meta_chunks <= num_of_allocated_meta_chunks)
1134                 num_of_meta_chunks = 0;
1135         else
1136                 num_of_meta_chunks -= num_of_allocated_meta_chunks;
1137
1138         total_size = allocated_total_size +
1139                      (num_of_data_chunks * default_chunk_size) +
1140                      (num_of_meta_chunks * default_chunk_size);
1141
1142         *num_of_meta_chunks_ret = num_of_meta_chunks;
1143         *size_of_data_ret = num_of_data_chunks * default_chunk_size;
1144         return total_size;
1145 }
1146
1147 static int zero_output_file(int out_fd, u64 size)
1148 {
1149         int loop_num;
1150         u64 location = 0;
1151         char buf[4096];
1152         int ret = 0, i;
1153         ssize_t written;
1154
1155         memset(buf, 0, 4096);
1156         loop_num = size / 4096;
1157         for (i = 0; i < loop_num; i++) {
1158                 written = pwrite64(out_fd, buf, 4096, location);
1159                 if (written != 4096)
1160                         ret = -EIO;
1161                 location += 4096;
1162         }
1163         return ret;
1164 }
1165
1166 static int is_ssd(const char *file)
1167 {
1168         blkid_probe probe;
1169         char wholedisk[PATH_MAX];
1170         char sysfs_path[PATH_MAX];
1171         dev_t devno;
1172         int fd;
1173         char rotational;
1174         int ret;
1175
1176         probe = blkid_new_probe_from_filename(file);
1177         if (!probe)
1178                 return 0;
1179
1180         /* Device number of this disk (possibly a partition) */
1181         devno = blkid_probe_get_devno(probe);
1182         if (!devno) {
1183                 blkid_free_probe(probe);
1184                 return 0;
1185         }
1186
1187         /* Get whole disk name (not full path) for this devno */
1188         ret = blkid_devno_to_wholedisk(devno,
1189                         wholedisk, sizeof(wholedisk), NULL);
1190         if (ret) {
1191                 blkid_free_probe(probe);
1192                 return 0;
1193         }
1194
1195         snprintf(sysfs_path, PATH_MAX, "/sys/block/%s/queue/rotational",
1196                  wholedisk);
1197
1198         blkid_free_probe(probe);
1199
1200         fd = open(sysfs_path, O_RDONLY);
1201         if (fd < 0) {
1202                 return 0;
1203         }
1204
1205         if (read(fd, &rotational, 1) < 1) {
1206                 close(fd);
1207                 return 0;
1208         }
1209         close(fd);
1210
1211         return rotational == '0';
1212 }
1213
1214 static int _cmp_device_by_id(void *priv, struct list_head *a,
1215                              struct list_head *b)
1216 {
1217         return list_entry(a, struct btrfs_device, dev_list)->devid -
1218                list_entry(b, struct btrfs_device, dev_list)->devid;
1219 }
1220
1221 static void list_all_devices(struct btrfs_root *root)
1222 {
1223         struct btrfs_fs_devices *fs_devices;
1224         struct btrfs_device *device;
1225         int number_of_devices = 0;
1226         u64 total_block_count = 0;
1227
1228         fs_devices = root->fs_info->fs_devices;
1229
1230         list_for_each_entry(device, &fs_devices->devices, dev_list)
1231                 number_of_devices++;
1232
1233         list_sort(NULL, &fs_devices->devices, _cmp_device_by_id);
1234
1235         printf("Number of devices:  %d\n", number_of_devices);
1236         /* printf("Total devices size: %10s\n", */
1237                 /* pretty_size(total_block_count)); */
1238         printf("Devices:\n");
1239         printf("   ID        SIZE  PATH\n");
1240         list_for_each_entry(device, &fs_devices->devices, dev_list) {
1241                 printf("  %3llu  %10s  %s\n",
1242                         device->devid,
1243                         pretty_size(device->total_bytes),
1244                         device->name);
1245                 total_block_count += device->total_bytes;
1246         }
1247
1248         printf("\n");
1249 }
1250
1251 static int is_temp_block_group(struct extent_buffer *node,
1252                                struct btrfs_block_group_item *bgi,
1253                                u64 data_profile, u64 meta_profile,
1254                                u64 sys_profile)
1255 {
1256         u64 flag = btrfs_disk_block_group_flags(node, bgi);
1257         u64 flag_type = flag & BTRFS_BLOCK_GROUP_TYPE_MASK;
1258         u64 flag_profile = flag & BTRFS_BLOCK_GROUP_PROFILE_MASK;
1259         u64 used = btrfs_disk_block_group_used(node, bgi);
1260
1261         /*
1262          * Chunks meets all the following conditions is a temp chunk
1263          * 1) Empty chunk
1264          * Temp chunk is always empty.
1265          *
1266          * 2) profile mismatch with mkfs profile.
1267          * Temp chunk is always in SINGLE
1268          *
1269          * 3) Size differs with mkfs_alloc
1270          * Special case for SINGLE/SINGLE btrfs.
1271          * In that case, temp data chunk and real data chunk are always empty.
1272          * So we need to use mkfs_alloc to be sure which chunk is the newly
1273          * allocated.
1274          *
1275          * Normally, new chunk size is equal to mkfs one (One chunk)
1276          * If it has multiple chunks, we just refuse to delete any one.
1277          * As they are all single, so no real problem will happen.
1278          * So only use condition 1) and 2) to judge them.
1279          */
1280         if (used != 0)
1281                 return 0;
1282         switch (flag_type) {
1283         case BTRFS_BLOCK_GROUP_DATA:
1284         case BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA:
1285                 data_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1286                 if (flag_profile != data_profile)
1287                         return 1;
1288                 break;
1289         case BTRFS_BLOCK_GROUP_METADATA:
1290                 meta_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1291                 if (flag_profile != meta_profile)
1292                         return 1;
1293                 break;
1294         case BTRFS_BLOCK_GROUP_SYSTEM:
1295                 sys_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1296                 if (flag_profile != sys_profile)
1297                         return 1;
1298                 break;
1299         }
1300         return 0;
1301 }
1302
1303 /* Note: if current is a block group, it will skip it anyway */
1304 static int next_block_group(struct btrfs_root *root,
1305                             struct btrfs_path *path)
1306 {
1307         struct btrfs_key key;
1308         int ret = 0;
1309
1310         while (1) {
1311                 ret = btrfs_next_item(root, path);
1312                 if (ret)
1313                         goto out;
1314
1315                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1316                 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
1317                         goto out;
1318         }
1319 out:
1320         return ret;
1321 }
1322
1323 /* This function will cleanup  */
1324 static int cleanup_temp_chunks(struct btrfs_fs_info *fs_info,
1325                                struct mkfs_allocation *alloc,
1326                                u64 data_profile, u64 meta_profile,
1327                                u64 sys_profile)
1328 {
1329         struct btrfs_trans_handle *trans = NULL;
1330         struct btrfs_block_group_item *bgi;
1331         struct btrfs_root *root = fs_info->extent_root;
1332         struct btrfs_key key;
1333         struct btrfs_key found_key;
1334         struct btrfs_path path;
1335         int ret = 0;
1336
1337         btrfs_init_path(&path);
1338         trans = btrfs_start_transaction(root, 1);
1339         BUG_ON(IS_ERR(trans));
1340
1341         key.objectid = 0;
1342         key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
1343         key.offset = 0;
1344
1345         while (1) {
1346                 /*
1347                  * as the rest of the loop may modify the tree, we need to
1348                  * start a new search each time.
1349                  */
1350                 ret = btrfs_search_slot(trans, root, &key, &path, 0, 0);
1351                 if (ret < 0)
1352                         goto out;
1353                 /* Don't pollute ret for >0 case */
1354                 if (ret > 0)
1355                         ret = 0;
1356
1357                 btrfs_item_key_to_cpu(path.nodes[0], &found_key,
1358                                       path.slots[0]);
1359                 if (found_key.objectid < key.objectid)
1360                         goto out;
1361                 if (found_key.type != BTRFS_BLOCK_GROUP_ITEM_KEY) {
1362                         ret = next_block_group(root, &path);
1363                         if (ret < 0)
1364                                 goto out;
1365                         if (ret > 0) {
1366                                 ret = 0;
1367                                 goto out;
1368                         }
1369                         btrfs_item_key_to_cpu(path.nodes[0], &found_key,
1370                                               path.slots[0]);
1371                 }
1372
1373                 bgi = btrfs_item_ptr(path.nodes[0], path.slots[0],
1374                                      struct btrfs_block_group_item);
1375                 if (is_temp_block_group(path.nodes[0], bgi,
1376                                         data_profile, meta_profile,
1377                                         sys_profile)) {
1378                         u64 flags = btrfs_disk_block_group_flags(path.nodes[0],
1379                                                              bgi);
1380
1381                         ret = btrfs_free_block_group(trans, fs_info,
1382                                         found_key.objectid, found_key.offset);
1383                         if (ret < 0)
1384                                 goto out;
1385
1386                         if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
1387                             BTRFS_BLOCK_GROUP_DATA)
1388                                 alloc->data -= found_key.offset;
1389                         else if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
1390                                  BTRFS_BLOCK_GROUP_METADATA)
1391                                 alloc->metadata -= found_key.offset;
1392                         else if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
1393                                  BTRFS_BLOCK_GROUP_SYSTEM)
1394                                 alloc->system -= found_key.offset;
1395                         else if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
1396                                  (BTRFS_BLOCK_GROUP_METADATA |
1397                                   BTRFS_BLOCK_GROUP_DATA))
1398                                 alloc->mixed -= found_key.offset;
1399                 }
1400                 btrfs_release_path(&path);
1401                 key.objectid = found_key.objectid + found_key.offset;
1402         }
1403 out:
1404         if (trans)
1405                 btrfs_commit_transaction(trans, root);
1406         btrfs_release_path(&path);
1407         return ret;
1408 }
1409
1410 int main(int argc, char **argv)
1411 {
1412         char *file;
1413         struct btrfs_root *root;
1414         struct btrfs_fs_info *fs_info;
1415         struct btrfs_trans_handle *trans;
1416         char *label = NULL;
1417         u64 block_count = 0;
1418         u64 dev_block_count = 0;
1419         u64 alloc_start = 0;
1420         u64 metadata_profile = 0;
1421         u64 data_profile = 0;
1422         u32 nodesize = max_t(u32, sysconf(_SC_PAGESIZE),
1423                         BTRFS_MKFS_DEFAULT_NODE_SIZE);
1424         u32 sectorsize = 4096;
1425         u32 stripesize = 4096;
1426         int zero_end = 1;
1427         int fd = -1;
1428         int ret;
1429         int close_ret;
1430         int i;
1431         int mixed = 0;
1432         int nodesize_forced = 0;
1433         int data_profile_opt = 0;
1434         int metadata_profile_opt = 0;
1435         int discard = 1;
1436         int ssd = 0;
1437         int force_overwrite = 0;
1438         char *source_dir = NULL;
1439         int source_dir_set = 0;
1440         u64 num_of_meta_chunks = 0;
1441         u64 size_of_data = 0;
1442         u64 source_dir_size = 0;
1443         int dev_cnt = 0;
1444         int saved_optind;
1445         char fs_uuid[BTRFS_UUID_UNPARSED_SIZE] = { 0 };
1446         u64 features = BTRFS_MKFS_DEFAULT_FEATURES;
1447         struct mkfs_allocation allocation = { 0 };
1448         struct btrfs_mkfs_config mkfs_cfg;
1449
1450         while(1) {
1451                 int c;
1452                 static const struct option long_options[] = {
1453                         { "alloc-start", required_argument, NULL, 'A'},
1454                         { "byte-count", required_argument, NULL, 'b' },
1455                         { "force", no_argument, NULL, 'f' },
1456                         { "leafsize", required_argument, NULL, 'l' },
1457                         { "label", required_argument, NULL, 'L'},
1458                         { "metadata", required_argument, NULL, 'm' },
1459                         { "mixed", no_argument, NULL, 'M' },
1460                         { "nodesize", required_argument, NULL, 'n' },
1461                         { "sectorsize", required_argument, NULL, 's' },
1462                         { "data", required_argument, NULL, 'd' },
1463                         { "version", no_argument, NULL, 'V' },
1464                         { "rootdir", required_argument, NULL, 'r' },
1465                         { "nodiscard", no_argument, NULL, 'K' },
1466                         { "features", required_argument, NULL, 'O' },
1467                         { "uuid", required_argument, NULL, 'U' },
1468                         { "quiet", 0, NULL, 'q' },
1469                         { "help", no_argument, NULL, GETOPT_VAL_HELP },
1470                         { NULL, 0, NULL, 0}
1471                 };
1472
1473                 c = getopt_long(argc, argv, "A:b:fl:n:s:m:d:L:O:r:U:VMKq",
1474                                 long_options, NULL);
1475                 if (c < 0)
1476                         break;
1477                 switch(c) {
1478                         case 'A':
1479                                 alloc_start = parse_size(optarg);
1480                                 break;
1481                         case 'f':
1482                                 force_overwrite = 1;
1483                                 break;
1484                         case 'd':
1485                                 data_profile = parse_profile(optarg);
1486                                 data_profile_opt = 1;
1487                                 break;
1488                         case 'l':
1489                                 warning("--leafsize is deprecated, use --nodesize");
1490                                 /* fall through */
1491                         case 'n':
1492                                 nodesize = parse_size(optarg);
1493                                 nodesize_forced = 1;
1494                                 break;
1495                         case 'L':
1496                                 label = parse_label(optarg);
1497                                 break;
1498                         case 'm':
1499                                 metadata_profile = parse_profile(optarg);
1500                                 metadata_profile_opt = 1;
1501                                 break;
1502                         case 'M':
1503                                 mixed = 1;
1504                                 break;
1505                         case 'O': {
1506                                 char *orig = strdup(optarg);
1507                                 char *tmp = orig;
1508
1509                                 tmp = btrfs_parse_fs_features(tmp, &features);
1510                                 if (tmp) {
1511                                         error("unrecognized filesystem feature '%s'",
1512                                                         tmp);
1513                                         free(orig);
1514                                         goto error;
1515                                 }
1516                                 free(orig);
1517                                 if (features & BTRFS_FEATURE_LIST_ALL) {
1518                                         btrfs_list_all_fs_features(0);
1519                                         goto success;
1520                                 }
1521                                 break;
1522                                 }
1523                         case 's':
1524                                 sectorsize = parse_size(optarg);
1525                                 break;
1526                         case 'b':
1527                                 block_count = parse_size(optarg);
1528                                 zero_end = 0;
1529                                 break;
1530                         case 'V':
1531                                 printf("mkfs.btrfs, part of %s\n",
1532                                                 PACKAGE_STRING);
1533                                 goto success;
1534                         case 'r':
1535                                 source_dir = optarg;
1536                                 source_dir_set = 1;
1537                                 break;
1538                         case 'U':
1539                                 strncpy(fs_uuid, optarg,
1540                                         BTRFS_UUID_UNPARSED_SIZE - 1);
1541                                 break;
1542                         case 'K':
1543                                 discard = 0;
1544                                 break;
1545                         case 'q':
1546                                 verbose = 0;
1547                                 break;
1548                         case GETOPT_VAL_HELP:
1549                         default:
1550                                 print_usage(c != GETOPT_VAL_HELP);
1551                 }
1552         }
1553
1554         if (verbose) {
1555                 printf("%s\n", PACKAGE_STRING);
1556                 printf("See %s for more information.\n\n", PACKAGE_URL);
1557         }
1558
1559         sectorsize = max(sectorsize, (u32)sysconf(_SC_PAGESIZE));
1560         stripesize = sectorsize;
1561         saved_optind = optind;
1562         dev_cnt = argc - optind;
1563         if (dev_cnt == 0)
1564                 print_usage(1);
1565
1566         if (source_dir_set && dev_cnt > 1) {
1567                 error("the option -r is limited to a single device");
1568                 goto error;
1569         }
1570
1571         if (*fs_uuid) {
1572                 uuid_t dummy_uuid;
1573
1574                 if (uuid_parse(fs_uuid, dummy_uuid) != 0) {
1575                         error("could not parse UUID: %s", fs_uuid);
1576                         goto error;
1577                 }
1578                 if (!test_uuid_unique(fs_uuid)) {
1579                         error("non-unique UUID: %s", fs_uuid);
1580                         goto error;
1581                 }
1582         }
1583
1584         while (dev_cnt-- > 0) {
1585                 file = argv[optind++];
1586                 if (is_block_device(file) == 1)
1587                         if (test_dev_for_mkfs(file, force_overwrite))
1588                                 goto error;
1589         }
1590
1591         optind = saved_optind;
1592         dev_cnt = argc - optind;
1593
1594         file = argv[optind++];
1595         ssd = is_ssd(file);
1596
1597         /*
1598         * Set default profiles according to number of added devices.
1599         * For mixed groups defaults are single/single.
1600         */
1601         if (!mixed) {
1602                 if (!metadata_profile_opt) {
1603                         if (dev_cnt == 1 && ssd && verbose)
1604                                 printf("Detected a SSD, turning off metadata "
1605                                 "duplication.  Mkfs with -m dup if you want to "
1606                                 "force metadata duplication.\n");
1607
1608                         metadata_profile = (dev_cnt > 1) ?
1609                                         BTRFS_BLOCK_GROUP_RAID1 : (ssd) ?
1610                                         0: BTRFS_BLOCK_GROUP_DUP;
1611                 }
1612                 if (!data_profile_opt) {
1613                         data_profile = (dev_cnt > 1) ?
1614                                 BTRFS_BLOCK_GROUP_RAID0 : 0; /* raid0 or single */
1615                 }
1616         } else {
1617                 u32 best_nodesize = max_t(u32, sysconf(_SC_PAGESIZE), sectorsize);
1618
1619                 if (metadata_profile_opt || data_profile_opt) {
1620                         if (metadata_profile != data_profile) {
1621                                 error(
1622         "with mixed block groups data and metadata profiles must be the same");
1623                                 goto error;
1624                         }
1625                 }
1626
1627                 if (!nodesize_forced)
1628                         nodesize = best_nodesize;
1629         }
1630
1631         /*
1632          * FS features that can be set by other means than -O
1633          * just set the bit here
1634          */
1635         if (mixed)
1636                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS;
1637
1638         if ((data_profile | metadata_profile) &
1639             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
1640                 features |= BTRFS_FEATURE_INCOMPAT_RAID56;
1641         }
1642
1643         if (btrfs_check_nodesize(nodesize, sectorsize,
1644                                  features))
1645                 goto error;
1646
1647         if (sectorsize < sizeof(struct btrfs_super_block)) {
1648                 error("sectorsize smaller than superblock: %u < %zu",
1649                                 sectorsize, sizeof(struct btrfs_super_block));
1650                 goto error;
1651         }
1652
1653         /* Check device/block_count after the nodesize is determined */
1654         if (block_count && block_count < btrfs_min_dev_size(nodesize)) {
1655                 error("size %llu is too small to make a usable filesystem",
1656                         block_count);
1657                 error("minimum size for btrfs filesystem is %llu",
1658                         btrfs_min_dev_size(nodesize));
1659                 goto error;
1660         }
1661         for (i = saved_optind; i < saved_optind + dev_cnt; i++) {
1662                 char *path;
1663
1664                 path = argv[i];
1665                 ret = test_minimum_size(path, nodesize);
1666                 if (ret < 0) {
1667                         error("failed to check size for %s: %s",
1668                                 path, strerror(-ret));
1669                         goto error;
1670                 }
1671                 if (ret > 0) {
1672                         error("'%s' is too small to make a usable filesystem",
1673                                 path);
1674                         error("minimum size for each btrfs device is %llu",
1675                                 btrfs_min_dev_size(nodesize));
1676                         goto error;
1677                 }
1678         }
1679         ret = test_num_disk_vs_raid(metadata_profile, data_profile,
1680                         dev_cnt, mixed, ssd);
1681         if (ret)
1682                 goto error;
1683
1684         dev_cnt--;
1685
1686         if (!source_dir_set) {
1687                 /*
1688                  * open without O_EXCL so that the problem should not
1689                  * occur by the following processing.
1690                  * (btrfs_register_one_device() fails if O_EXCL is on)
1691                  */
1692                 fd = open(file, O_RDWR);
1693                 if (fd < 0) {
1694                         error("unable to open %s: %s", file, strerror(errno));
1695                         goto error;
1696                 }
1697                 ret = btrfs_prepare_device(fd, file, &dev_block_count,
1698                                 block_count,
1699                                 (zero_end ? PREP_DEVICE_ZERO_END : 0) |
1700                                 (discard ? PREP_DEVICE_DISCARD : 0) |
1701                                 (verbose ? PREP_DEVICE_VERBOSE : 0));
1702                 if (ret) {
1703                         goto error;
1704                 }
1705                 if (block_count && block_count > dev_block_count) {
1706                         error("%s is smaller than requested size, expected %llu, found %llu",
1707                                         file,
1708                                         (unsigned long long)block_count,
1709                                         (unsigned long long)dev_block_count);
1710                         goto error;
1711                 }
1712         } else {
1713                 fd = open(file, O_CREAT | O_RDWR,
1714                                 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH);
1715                 if (fd < 0) {
1716                         error("unable to open %s: %s", file, strerror(errno));
1717                         goto error;
1718                 }
1719
1720                 source_dir_size = size_sourcedir(source_dir, sectorsize,
1721                                              &num_of_meta_chunks, &size_of_data);
1722                 if(block_count < source_dir_size)
1723                         block_count = source_dir_size;
1724                 ret = zero_output_file(fd, block_count);
1725                 if (ret) {
1726                         error("unable to zero the output file");
1727                         goto error;
1728                 }
1729                 /* our "device" is the new image file */
1730                 dev_block_count = block_count;
1731         }
1732
1733         /* To create the first block group and chunk 0 in make_btrfs */
1734         if (dev_block_count < BTRFS_MKFS_SYSTEM_GROUP_SIZE) {
1735                 error("device is too small to make filesystem, must be at least %llu",
1736                                 (unsigned long long)BTRFS_MKFS_SYSTEM_GROUP_SIZE);
1737                 goto error;
1738         }
1739
1740         if (group_profile_max_safe_loss(metadata_profile) <
1741                 group_profile_max_safe_loss(data_profile)){
1742                 warning("metadata has lower redundancy than data!\n");
1743         }
1744
1745         mkfs_cfg.label = label;
1746         memcpy(mkfs_cfg.fs_uuid, fs_uuid, sizeof(mkfs_cfg.fs_uuid));
1747         mkfs_cfg.num_bytes = dev_block_count;
1748         mkfs_cfg.nodesize = nodesize;
1749         mkfs_cfg.sectorsize = sectorsize;
1750         mkfs_cfg.stripesize = stripesize;
1751         mkfs_cfg.features = features;
1752
1753         ret = make_btrfs(fd, &mkfs_cfg);
1754         if (ret) {
1755                 error("error during mkfs: %s", strerror(-ret));
1756                 goto error;
1757         }
1758
1759         fs_info = open_ctree_fs_info(file, 0, 0, 0,
1760                         OPEN_CTREE_WRITES | OPEN_CTREE_FS_PARTIAL);
1761         if (!fs_info) {
1762                 error("open ctree failed");
1763                 goto error;
1764         }
1765         close(fd);
1766         fd = -1;
1767         root = fs_info->fs_root;
1768         fs_info->alloc_start = alloc_start;
1769
1770         ret = create_metadata_block_groups(root, mixed, &allocation);
1771         if (ret) {
1772                 error("failed to create default block groups: %d", ret);
1773                 goto error;
1774         }
1775
1776         trans = btrfs_start_transaction(root, 1);
1777         if (IS_ERR(trans)) {
1778                 error("failed to start transaction");
1779                 goto error;
1780         }
1781
1782         ret = create_data_block_groups(trans, root, mixed, &allocation);
1783         if (ret) {
1784                 error("failed to create default data block groups: %d", ret);
1785                 goto error;
1786         }
1787
1788         ret = make_root_dir(trans, root);
1789         if (ret) {
1790                 error("failed to setup the root directory: %d", ret);
1791                 goto error;
1792         }
1793
1794         ret = btrfs_commit_transaction(trans, root);
1795         if (ret) {
1796                 error("unable to commit transaction: %d", ret);
1797                 goto out;
1798         }
1799
1800         trans = btrfs_start_transaction(root, 1);
1801         if (IS_ERR(trans)) {
1802                 error("failed to start transaction");
1803                 goto error;
1804         }
1805
1806         if (dev_cnt == 0)
1807                 goto raid_groups;
1808
1809         while (dev_cnt-- > 0) {
1810                 file = argv[optind++];
1811
1812                 /*
1813                  * open without O_EXCL so that the problem should not
1814                  * occur by the following processing.
1815                  * (btrfs_register_one_device() fails if O_EXCL is on)
1816                  */
1817                 fd = open(file, O_RDWR);
1818                 if (fd < 0) {
1819                         error("unable to open %s: %s", file, strerror(errno));
1820                         goto error;
1821                 }
1822                 ret = btrfs_device_already_in_root(root, fd,
1823                                                    BTRFS_SUPER_INFO_OFFSET);
1824                 if (ret) {
1825                         error("skipping duplicate device %s in the filesystem",
1826                                 file);
1827                         close(fd);
1828                         continue;
1829                 }
1830                 ret = btrfs_prepare_device(fd, file, &dev_block_count,
1831                                 block_count,
1832                                 (verbose ? PREP_DEVICE_VERBOSE : 0) |
1833                                 (zero_end ? PREP_DEVICE_ZERO_END : 0) |
1834                                 (discard ? PREP_DEVICE_DISCARD : 0));
1835                 if (ret) {
1836                         goto error;
1837                 }
1838
1839                 ret = btrfs_add_to_fsid(trans, root, fd, file, dev_block_count,
1840                                         sectorsize, sectorsize, sectorsize);
1841                 if (ret) {
1842                         error("unable to add %s to filesystem: %d", file, ret);
1843                         goto out;
1844                 }
1845                 if (verbose >= 2) {
1846                         struct btrfs_device *device;
1847
1848                         device = container_of(fs_info->fs_devices->devices.next,
1849                                         struct btrfs_device, dev_list);
1850                         printf("adding device %s id %llu\n", file,
1851                                 (unsigned long long)device->devid);
1852                 }
1853         }
1854
1855 raid_groups:
1856         if (!source_dir_set) {
1857                 ret = create_raid_groups(trans, root, data_profile,
1858                                  metadata_profile, mixed, &allocation);
1859                 if (ret) {
1860                         error("unable to create raid groups: %d", ret);
1861                         goto out;
1862                 }
1863         }
1864
1865         ret = create_tree(trans, root, BTRFS_DATA_RELOC_TREE_OBJECTID);
1866         if (ret) {
1867                 error("unable to create data reloc tree: %d", ret);
1868                 goto out;
1869         }
1870
1871         ret = btrfs_commit_transaction(trans, root);
1872         if (ret) {
1873                 error("unable to commit transaction: %d", ret);
1874                 goto out;
1875         }
1876
1877         if (source_dir_set) {
1878                 trans = btrfs_start_transaction(root, 1);
1879                 BUG_ON(IS_ERR(trans));
1880                 ret = create_chunks(trans, root,
1881                                     num_of_meta_chunks, size_of_data,
1882                                     &allocation);
1883                 if (ret) {
1884                         error("unable to create chunks: %d", ret);
1885                         goto out;
1886                 }
1887                 ret = btrfs_commit_transaction(trans, root);
1888                 if (ret) {
1889                         error("transaction commit failed: %d", ret);
1890                         goto out;
1891                 }
1892
1893                 ret = make_image(source_dir, root);
1894                 if (ret) {
1895                         error("error wihle filling filesystem: %d", ret);
1896                         goto out;
1897                 }
1898         }
1899         ret = cleanup_temp_chunks(fs_info, &allocation, data_profile,
1900                                   metadata_profile, metadata_profile);
1901         if (ret < 0) {
1902                 error("failed to cleanup temporary chunks: %d", ret);
1903                 goto out;
1904         }
1905
1906         if (verbose) {
1907                 char features_buf[64];
1908
1909                 printf("Label:              %s\n", label);
1910                 printf("UUID:               %s\n", mkfs_cfg.fs_uuid);
1911                 printf("Node size:          %u\n", nodesize);
1912                 printf("Sector size:        %u\n", sectorsize);
1913                 printf("Filesystem size:    %s\n",
1914                         pretty_size(btrfs_super_total_bytes(fs_info->super_copy)));
1915                 printf("Block group profiles:\n");
1916                 if (allocation.data)
1917                         printf("  Data:             %-8s %16s\n",
1918                                 btrfs_group_profile_str(data_profile),
1919                                 pretty_size(allocation.data));
1920                 if (allocation.metadata)
1921                         printf("  Metadata:         %-8s %16s\n",
1922                                 btrfs_group_profile_str(metadata_profile),
1923                                 pretty_size(allocation.metadata));
1924                 if (allocation.mixed)
1925                         printf("  Data+Metadata:    %-8s %16s\n",
1926                                 btrfs_group_profile_str(data_profile),
1927                                 pretty_size(allocation.mixed));
1928                 printf("  System:           %-8s %16s\n",
1929                         btrfs_group_profile_str(metadata_profile),
1930                         pretty_size(allocation.system));
1931                 printf("SSD detected:       %s\n", ssd ? "yes" : "no");
1932                 btrfs_parse_features_to_string(features_buf, features);
1933                 printf("Incompat features:  %s", features_buf);
1934                 printf("\n");
1935
1936                 list_all_devices(root);
1937         }
1938
1939         /*
1940          * The filesystem is now fully set up, commit the remaining changes and
1941          * fix the signature as the last step before closing the devices.
1942          */
1943         fs_info->finalize_on_close = 1;
1944 out:
1945         close_ret = close_ctree(root);
1946
1947         if (!close_ret) {
1948                 optind = saved_optind;
1949                 dev_cnt = argc - optind;
1950                 while (dev_cnt-- > 0) {
1951                         file = argv[optind++];
1952                         if (is_block_device(file) == 1)
1953                                 btrfs_register_one_device(file);
1954                 }
1955         }
1956
1957         btrfs_close_all_devices();
1958         free(label);
1959
1960         return !!ret;
1961 error:
1962         if (fd > 0)
1963                 close(fd);
1964
1965         free(label);
1966         exit(1);
1967 success:
1968         exit(0);
1969 }