mkfs.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include "kerncompat.h"
  20 #include "androidcompat.h"
  21
  22 #include <sys/ioctl.h>
  23 #include <sys/mount.h>
  24 #include "ioctl.h"
  25 #include <stdio.h>
  26 #include <stdlib.h>
  27 #include <sys/types.h>
  28 #include <sys/stat.h>
  29 /* #include <sys/dir.h> included via androidcompat.h */
  30 #include <fcntl.h>
  31 #include <unistd.h>
  32 #include <getopt.h>
  33 #include <uuid/uuid.h>
  34 #include <ctype.h>
  35 #include <sys/xattr.h>
  36 #include <limits.h>
  37 #include <linux/limits.h>
  38 #include <blkid/blkid.h>
  39 #include <ftw.h>
  40 #include "ctree.h"
  41 #include "disk-io.h"
  42 #include "volumes.h"
  43 #include "transaction.h"
  44 #include "utils.h"
  45 #include "list_sort.h"
  46
  47 static u64 index_cnt = 2;
  48 static int verbose = 1;
  49
  50 struct directory_name_entry {
  51         char *dir_name;
  52         char *path;
  53         ino_t inum;
  54         struct list_head list;
  55 };
  56
  57 struct mkfs_allocation {
  58         u64 data;
  59         u64 metadata;
  60         u64 mixed;
  61         u64 system;
  62 };
  63
  64 static int create_metadata_block_groups(struct btrfs_root *root, int mixed,
  65                                 struct mkfs_allocation *allocation)
  66 {
  67         struct btrfs_trans_handle *trans;
  68         u64 bytes_used;
  69         u64 chunk_start = 0;
  70         u64 chunk_size = 0;
  71         int ret;
  72
  73         trans = btrfs_start_transaction(root, 1);
  74         bytes_used = btrfs_super_bytes_used(root->fs_info->super_copy);
  75
  76         root->fs_info->system_allocs = 1;
  77         ret = btrfs_make_block_group(trans, root, bytes_used,
  78                                      BTRFS_BLOCK_GROUP_SYSTEM,
  79                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
  80                                      0, BTRFS_MKFS_SYSTEM_GROUP_SIZE);
  81         allocation->system += BTRFS_MKFS_SYSTEM_GROUP_SIZE;
  82         BUG_ON(ret);
  83
  84         if (mixed) {
  85                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
  86                                         &chunk_start, &chunk_size,
  87                                         BTRFS_BLOCK_GROUP_METADATA |
  88                                         BTRFS_BLOCK_GROUP_DATA);
  89                 if (ret == -ENOSPC) {
  90                         fprintf(stderr,
  91                                 "no space to alloc data/metadata chunk\n");
  92                         goto err;
  93                 }
  94                 BUG_ON(ret);
  95                 ret = btrfs_make_block_group(trans, root, 0,
  96                                              BTRFS_BLOCK_GROUP_METADATA |
  97                                              BTRFS_BLOCK_GROUP_DATA,
  98                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
  99                                              chunk_start, chunk_size);
 100                 BUG_ON(ret);
 101                 allocation->mixed += chunk_size;
 102         } else {
 103                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 104                                         &chunk_start, &chunk_size,
 105                                         BTRFS_BLOCK_GROUP_METADATA);
 106                 if (ret == -ENOSPC) {
 107                         fprintf(stderr, "no space to alloc metadata chunk\n");
 108                         goto err;
 109                 }
 110                 BUG_ON(ret);
 111                 ret = btrfs_make_block_group(trans, root, 0,
 112                                              BTRFS_BLOCK_GROUP_METADATA,
 113                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 114                                              chunk_start, chunk_size);
 115                 allocation->metadata += chunk_size;
 116                 BUG_ON(ret);
 117         }
 118
 119         root->fs_info->system_allocs = 0;
 120         btrfs_commit_transaction(trans, root);
 121
 122 err:
 123         return ret;
 124 }
 125
 126 static int create_data_block_groups(struct btrfs_trans_handle *trans,
 127                 struct btrfs_root *root, int mixed,
 128                 struct mkfs_allocation *allocation)
 129 {
 130         u64 chunk_start = 0;
 131         u64 chunk_size = 0;
 132         int ret = 0;
 133
 134         if (!mixed) {
 135                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 136                                         &chunk_start, &chunk_size,
 137                                         BTRFS_BLOCK_GROUP_DATA);
 138                 if (ret == -ENOSPC) {
 139                         fprintf(stderr, "no space to alloc data chunk\n");
 140                         goto err;
 141                 }
 142                 BUG_ON(ret);
 143                 ret = btrfs_make_block_group(trans, root, 0,
 144                                              BTRFS_BLOCK_GROUP_DATA,
 145                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 146                                              chunk_start, chunk_size);
 147                 allocation->data += chunk_size;
 148                 BUG_ON(ret);
 149         }
 150
 151 err:
 152         return ret;
 153 }
 154
 155 static int make_root_dir(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 156                 struct mkfs_allocation *allocation)
 157 {
 158         struct btrfs_key location;
 159         int ret;
 160
 161         ret = btrfs_make_root_dir(trans, root->fs_info->tree_root,
 162                               BTRFS_ROOT_TREE_DIR_OBJECTID);
 163         if (ret)
 164                 goto err;
 165         ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
 166         if (ret)
 167                 goto err;
 168         memcpy(&location, &root->fs_info->fs_root->root_key, sizeof(location));
 169         location.offset = (u64)-1;
 170         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
 171                         "default", 7,
 172                         btrfs_super_root_dir(root->fs_info->super_copy),
 173                         &location, BTRFS_FT_DIR, 0);
 174         if (ret)
 175                 goto err;
 176
 177         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 178                              "default", 7, location.objectid,
 179                              BTRFS_ROOT_TREE_DIR_OBJECTID, 0);
 180         if (ret)
 181                 goto err;
 182
 183 err:
 184         return ret;
 185 }
 186
 187 static void __recow_root(struct btrfs_trans_handle *trans,
 188                          struct btrfs_root *root)
 189 {
 190         int ret;
 191         struct extent_buffer *tmp;
 192
 193         if (trans->transid != btrfs_root_generation(&root->root_item)) {
 194                 extent_buffer_get(root->node);
 195                 ret = __btrfs_cow_block(trans, root, root->node,
 196                                         NULL, 0, &tmp, 0, 0);
 197                 BUG_ON(ret);
 198                 free_extent_buffer(tmp);
 199         }
 200 }
 201
 202 static void recow_roots(struct btrfs_trans_handle *trans,
 203                        struct btrfs_root *root)
 204 {
 205         struct btrfs_fs_info *info = root->fs_info;
 206
 207         __recow_root(trans, info->fs_root);
 208         __recow_root(trans, info->tree_root);
 209         __recow_root(trans, info->extent_root);
 210         __recow_root(trans, info->chunk_root);
 211         __recow_root(trans, info->dev_root);
 212         __recow_root(trans, info->csum_root);
 213 }
 214
 215 static int create_one_raid_group(struct btrfs_trans_handle *trans,
 216                               struct btrfs_root *root, u64 type,
 217                               struct mkfs_allocation *allocation)
 218
 219 {
 220         u64 chunk_start;
 221         u64 chunk_size;
 222         int ret;
 223
 224         ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 225                                 &chunk_start, &chunk_size, type);
 226         if (ret == -ENOSPC) {
 227                 fprintf(stderr, "not enough free space\n");
 228                 exit(1);
 229         }
 230         BUG_ON(ret);
 231         ret = btrfs_make_block_group(trans, root->fs_info->extent_root, 0,
 232                                      type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 233                                      chunk_start, chunk_size);
 234         if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == BTRFS_BLOCK_GROUP_DATA)
 235                 allocation->data += chunk_size;
 236         else if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == BTRFS_BLOCK_GROUP_METADATA)
 237                 allocation->metadata += chunk_size;
 238         else if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == BTRFS_BLOCK_GROUP_SYSTEM)
 239                 allocation->system += chunk_size;
 240         else if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
 241                         (BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA))
 242                 allocation->mixed += chunk_size;
 243         else
 244                 BUG_ON(1);
 245
 246         BUG_ON(ret);
 247         return ret;
 248 }
 249
 250 static int create_raid_groups(struct btrfs_trans_handle *trans,
 251                               struct btrfs_root *root, u64 data_profile,
 252                               u64 metadata_profile, int mixed,
 253                               struct mkfs_allocation *allocation)
 254 {
 255         u64 num_devices = btrfs_super_num_devices(root->fs_info->super_copy);
 256         int ret;
 257
 258         if (metadata_profile) {
 259                 u64 meta_flags = BTRFS_BLOCK_GROUP_METADATA;
 260
 261                 ret = create_one_raid_group(trans, root,
 262                                             BTRFS_BLOCK_GROUP_SYSTEM |
 263                                             metadata_profile, allocation);
 264                 BUG_ON(ret);
 265
 266                 if (mixed)
 267                         meta_flags |= BTRFS_BLOCK_GROUP_DATA;
 268
 269                 ret = create_one_raid_group(trans, root, meta_flags |
 270                                             metadata_profile, allocation);
 271                 BUG_ON(ret);
 272
 273         }
 274         if (!mixed && num_devices > 1 && data_profile) {
 275                 ret = create_one_raid_group(trans, root,
 276                                             BTRFS_BLOCK_GROUP_DATA |
 277                                             data_profile, allocation);
 278                 BUG_ON(ret);
 279         }
 280         recow_roots(trans, root);
 281
 282         return 0;
 283 }
 284
 285 static int create_data_reloc_tree(struct btrfs_trans_handle *trans,
 286                                   struct btrfs_root *root)
 287 {
 288         struct btrfs_key location;
 289         struct btrfs_root_item root_item;
 290         struct extent_buffer *tmp;
 291         u64 objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
 292         int ret;
 293
 294         ret = btrfs_copy_root(trans, root, root->node, &tmp, objectid);
 295         BUG_ON(ret);
 296
 297         memcpy(&root_item, &root->root_item, sizeof(root_item));
 298         btrfs_set_root_bytenr(&root_item, tmp->start);
 299         btrfs_set_root_level(&root_item, btrfs_header_level(tmp));
 300         btrfs_set_root_generation(&root_item, trans->transid);
 301         free_extent_buffer(tmp);
 302
 303         location.objectid = objectid;
 304         location.type = BTRFS_ROOT_ITEM_KEY;
 305         location.offset = 0;
 306         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
 307                                 &location, &root_item);
 308         BUG_ON(ret);
 309         return 0;
 310 }
 311
 312 static void print_usage(int ret)
 313 {
 314         fprintf(stderr, "usage: mkfs.btrfs [options] dev [ dev ... ]\n");
 315         fprintf(stderr, "options:\n");
 316         fprintf(stderr, "\t-A|--alloc-start START  the offset to start the FS\n");
 317         fprintf(stderr, "\t-b|--byte-count SIZE    total number of bytes in the FS\n");
 318         fprintf(stderr, "\t-d|--data PROFILE       data profile, raid0, raid1, raid5, raid6, raid10, dup or single\n");
 319         fprintf(stderr, "\t-f|--force              force overwrite of existing filesystem\n");
 320         fprintf(stderr, "\t-l|--leafsize SIZE      deprecated, alias for nodesize\n");
 321         fprintf(stderr, "\t-L|--label LABEL        set a label\n");
 322         fprintf(stderr, "\t-m|--metadata PROFILE   metadata profile, values like data profile\n");
 323         fprintf(stderr, "\t-M|--mixed              mix metadata and data together\n");
 324         fprintf(stderr, "\t-n|--nodesize SIZE      size of btree nodes\n");
 325         fprintf(stderr, "\t-s|--sectorsize SIZE    min block allocation (may not mountable by current kernel)\n");
 326         fprintf(stderr, "\t-r|--rootdir DIR        the source directory\n");
 327         fprintf(stderr, "\t-K|--nodiscard          do not perform whole device TRIM\n");
 328         fprintf(stderr, "\t-O|--features LIST      comma separated list of filesystem features, use '-O list-all' to list features\n");
 329         fprintf(stderr, "\t-U|--uuid UUID          specify the filesystem UUID\n");
 330         fprintf(stderr, "\t-q|--quiet              no messages except errors\n");
 331         fprintf(stderr, "\t-V|--version            print the mkfs.btrfs version and exit\n");
 332         exit(ret);
 333 }
 334
 335 static void print_version(void) __attribute__((noreturn));
 336 static void print_version(void)
 337 {
 338         fprintf(stderr, "mkfs.btrfs, part of %s\n", PACKAGE_STRING);
 339         exit(0);
 340 }
 341
 342 static u64 parse_profile(char *s)
 343 {
 344         if (strcmp(s, "raid0") == 0) {
 345                 return BTRFS_BLOCK_GROUP_RAID0;
 346         } else if (strcasecmp(s, "raid1") == 0) {
 347                 return BTRFS_BLOCK_GROUP_RAID1;
 348         } else if (strcasecmp(s, "raid5") == 0) {
 349                 return BTRFS_BLOCK_GROUP_RAID5;
 350         } else if (strcasecmp(s, "raid6") == 0) {
 351                 return BTRFS_BLOCK_GROUP_RAID6;
 352         } else if (strcasecmp(s, "raid10") == 0) {
 353                 return BTRFS_BLOCK_GROUP_RAID10;
 354         } else if (strcasecmp(s, "dup") == 0) {
 355                 return BTRFS_BLOCK_GROUP_DUP;
 356         } else if (strcasecmp(s, "single") == 0) {
 357                 return 0;
 358         } else {
 359                 fprintf(stderr, "Unknown profile %s\n", s);
 360                 exit(1);
 361         }
 362         /* not reached */
 363         return 0;
 364 }
 365
 366 static char *parse_label(char *input)
 367 {
 368         int len = strlen(input);
 369
 370         if (len >= BTRFS_LABEL_SIZE) {
 371                 fprintf(stderr, "Label %s is too long (max %d)\n", input,
 372                         BTRFS_LABEL_SIZE - 1);
 373                 exit(1);
 374         }
 375         return strdup(input);
 376 }
 377
 378 static int add_directory_items(struct btrfs_trans_handle *trans,
 379                                struct btrfs_root *root, u64 objectid,
 380                                ino_t parent_inum, const char *name,
 381                                struct stat *st, int *dir_index_cnt)
 382 {
 383         int ret;
 384         int name_len;
 385         struct btrfs_key location;
 386         u8 filetype = 0;
 387
 388         name_len = strlen(name);
 389
 390         location.objectid = objectid;
 391         location.offset = 0;
 392         btrfs_set_key_type(&location, BTRFS_INODE_ITEM_KEY);
 393
 394         if (S_ISDIR(st->st_mode))
 395                 filetype = BTRFS_FT_DIR;
 396         if (S_ISREG(st->st_mode))
 397                 filetype = BTRFS_FT_REG_FILE;
 398         if (S_ISLNK(st->st_mode))
 399                 filetype = BTRFS_FT_SYMLINK;
 400
 401         ret = btrfs_insert_dir_item(trans, root, name, name_len,
 402                                     parent_inum, &location,
 403                                     filetype, index_cnt);
 404         if (ret)
 405                 return ret;
 406         ret = btrfs_insert_inode_ref(trans, root, name, name_len,
 407                                      objectid, parent_inum, index_cnt);
 408         *dir_index_cnt = index_cnt;
 409         index_cnt++;
 410
 411         return ret;
 412 }
 413
 414 static int fill_inode_item(struct btrfs_trans_handle *trans,
 415                            struct btrfs_root *root,
 416                            struct btrfs_inode_item *dst, struct stat *src)
 417 {
 418         u64 blocks = 0;
 419         u64 sectorsize = root->sectorsize;
 420
 421         /*
 422          * btrfs_inode_item has some reserved fields
 423          * and represents on-disk inode entry, so
 424          * zero everything to prevent information leak
 425          */
 426         memset(dst, 0, sizeof (*dst));
 427
 428         btrfs_set_stack_inode_generation(dst, trans->transid);
 429         btrfs_set_stack_inode_size(dst, src->st_size);
 430         btrfs_set_stack_inode_nbytes(dst, 0);
 431         btrfs_set_stack_inode_block_group(dst, 0);
 432         btrfs_set_stack_inode_nlink(dst, src->st_nlink);
 433         btrfs_set_stack_inode_uid(dst, src->st_uid);
 434         btrfs_set_stack_inode_gid(dst, src->st_gid);
 435         btrfs_set_stack_inode_mode(dst, src->st_mode);
 436         btrfs_set_stack_inode_rdev(dst, 0);
 437         btrfs_set_stack_inode_flags(dst, 0);
 438         btrfs_set_stack_timespec_sec(&dst->atime, src->st_atime);
 439         btrfs_set_stack_timespec_nsec(&dst->atime, 0);
 440         btrfs_set_stack_timespec_sec(&dst->ctime, src->st_ctime);
 441         btrfs_set_stack_timespec_nsec(&dst->ctime, 0);
 442         btrfs_set_stack_timespec_sec(&dst->mtime, src->st_mtime);
 443         btrfs_set_stack_timespec_nsec(&dst->mtime, 0);
 444         btrfs_set_stack_timespec_sec(&dst->otime, 0);
 445         btrfs_set_stack_timespec_nsec(&dst->otime, 0);
 446
 447         if (S_ISDIR(src->st_mode)) {
 448                 btrfs_set_stack_inode_size(dst, 0);
 449                 btrfs_set_stack_inode_nlink(dst, 1);
 450         }
 451         if (S_ISREG(src->st_mode)) {
 452                 btrfs_set_stack_inode_size(dst, (u64)src->st_size);
 453                 if (src->st_size <= BTRFS_MAX_INLINE_DATA_SIZE(root))
 454                         btrfs_set_stack_inode_nbytes(dst, src->st_size);
 455                 else {
 456                         blocks = src->st_size / sectorsize;
 457                         if (src->st_size % sectorsize)
 458                                 blocks += 1;
 459                         blocks *= sectorsize;
 460                         btrfs_set_stack_inode_nbytes(dst, blocks);
 461                 }
 462         }
 463         if (S_ISLNK(src->st_mode))
 464                 btrfs_set_stack_inode_nbytes(dst, src->st_size + 1);
 465
 466         return 0;
 467 }
 468
 469 static int directory_select(const struct direct *entry)
 470 {
 471         if ((strncmp(entry->d_name, ".", entry->d_reclen) == 0) ||
 472                 (strncmp(entry->d_name, "..", entry->d_reclen) == 0))
 473                 return 0;
 474         else
 475                 return 1;
 476 }
 477
 478 static void free_namelist(struct direct **files, int count)
 479 {
 480         int i;
 481
 482         if (count < 0)
 483                 return;
 484
 485         for (i = 0; i < count; ++i)
 486                 free(files[i]);
 487         free(files);
 488 }
 489
 490 static u64 calculate_dir_inode_size(char *dirname)
 491 {
 492         int count, i;
 493         struct direct **files, *cur_file;
 494         u64 dir_inode_size = 0;
 495
 496         count = scandir(dirname, &files, directory_select, NULL);
 497
 498         for (i = 0; i < count; i++) {
 499                 cur_file = files[i];
 500                 dir_inode_size += strlen(cur_file->d_name);
 501         }
 502
 503         free_namelist(files, count);
 504
 505         dir_inode_size *= 2;
 506         return dir_inode_size;
 507 }
 508
 509 static int add_inode_items(struct btrfs_trans_handle *trans,
 510                            struct btrfs_root *root,
 511                            struct stat *st, char *name,
 512                            u64 self_objectid, ino_t parent_inum,
 513                            int dir_index_cnt, struct btrfs_inode_item *inode_ret)
 514 {
 515         int ret;
 516         struct btrfs_key inode_key;
 517         struct btrfs_inode_item btrfs_inode;
 518         u64 objectid;
 519         u64 inode_size = 0;
 520
 521         fill_inode_item(trans, root, &btrfs_inode, st);
 522         objectid = self_objectid;
 523
 524         if (S_ISDIR(st->st_mode)) {
 525                 inode_size = calculate_dir_inode_size(name);
 526                 btrfs_set_stack_inode_size(&btrfs_inode, inode_size);
 527         }
 528
 529         inode_key.objectid = objectid;
 530         inode_key.offset = 0;
 531         btrfs_set_key_type(&inode_key, BTRFS_INODE_ITEM_KEY);
 532
 533         ret = btrfs_insert_inode(trans, root, objectid, &btrfs_inode);
 534
 535         *inode_ret = btrfs_inode;
 536         return ret;
 537 }
 538
 539 static int add_xattr_item(struct btrfs_trans_handle *trans,
 540                           struct btrfs_root *root, u64 objectid,
 541                           const char *file_name)
 542 {
 543         int ret;
 544         int cur_name_len;
 545         char xattr_list[XATTR_LIST_MAX];
 546         char *cur_name;
 547         char cur_value[XATTR_SIZE_MAX];
 548         char delimiter = '\0';
 549         char *next_location = xattr_list;
 550
 551         ret = llistxattr(file_name, xattr_list, XATTR_LIST_MAX);
 552         if (ret < 0) {
 553                 if(errno == ENOTSUP)
 554                         return 0;
 555                 fprintf(stderr, "get a list of xattr failed for %s\n",
 556                         file_name);
 557                 return ret;
 558         }
 559         if (ret == 0)
 560                 return ret;
 561
 562         cur_name = strtok(xattr_list, &delimiter);
 563         while (cur_name != NULL) {
 564                 cur_name_len = strlen(cur_name);
 565                 next_location += cur_name_len + 1;
 566
 567                 ret = getxattr(file_name, cur_name, cur_value, XATTR_SIZE_MAX);
 568                 if (ret < 0) {
 569                         if(errno == ENOTSUP)
 570                                 return 0;
 571                         fprintf(stderr, "get a xattr value failed for %s attr %s\n",
 572                                 file_name, cur_name);
 573                         return ret;
 574                 }
 575
 576                 ret = btrfs_insert_xattr_item(trans, root, cur_name,
 577                                               cur_name_len, cur_value,
 578                                               ret, objectid);
 579                 if (ret) {
 580                         fprintf(stderr, "insert a xattr item failed for %s\n",
 581                                 file_name);
 582                 }
 583
 584                 cur_name = strtok(next_location, &delimiter);
 585         }
 586
 587         return ret;
 588 }
 589
 590 static int add_symbolic_link(struct btrfs_trans_handle *trans,
 591                              struct btrfs_root *root,
 592                              u64 objectid, const char *path_name)
 593 {
 594         int ret;
 595         u64 sectorsize = root->sectorsize;
 596         char *buf = malloc(sectorsize);
 597
 598         ret = readlink(path_name, buf, sectorsize);
 599         if (ret <= 0) {
 600                 fprintf(stderr, "readlink failed for %s\n", path_name);
 601                 goto fail;
 602         }
 603         if (ret >= sectorsize) {
 604                 fprintf(stderr, "symlink too long for %s\n", path_name);
 605                 ret = -1;
 606                 goto fail;
 607         }
 608
 609         buf[ret] = '\0'; /* readlink does not do it for us */
 610         ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
 611                                          buf, ret + 1);
 612 fail:
 613         free(buf);
 614         return ret;
 615 }
 616
 617 static int add_file_items(struct btrfs_trans_handle *trans,
 618                           struct btrfs_root *root,
 619                           struct btrfs_inode_item *btrfs_inode, u64 objectid,
 620                           ino_t parent_inum, struct stat *st,
 621                           const char *path_name, int out_fd)
 622 {
 623         int ret = -1;
 624         ssize_t ret_read;
 625         u64 bytes_read = 0;
 626         struct btrfs_key key;
 627         int blocks;
 628         u32 sectorsize = root->sectorsize;
 629         u64 first_block = 0;
 630         u64 file_pos = 0;
 631         u64 cur_bytes;
 632         u64 total_bytes;
 633         struct extent_buffer *eb = NULL;
 634         int fd;
 635
 636         if (st->st_size == 0)
 637                 return 0;
 638
 639         fd = open(path_name, O_RDONLY);
 640         if (fd == -1) {
 641                 fprintf(stderr, "%s open failed\n", path_name);
 642                 return ret;
 643         }
 644
 645         blocks = st->st_size / sectorsize;
 646         if (st->st_size % sectorsize)
 647                 blocks += 1;
 648
 649         if (st->st_size <= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 650                 char *buffer = malloc(st->st_size);
 651
 652                 if (!buffer) {
 653                         ret = -ENOMEM;
 654                         goto end;
 655                 }
 656
 657                 ret_read = pread64(fd, buffer, st->st_size, bytes_read);
 658                 if (ret_read == -1) {
 659                         fprintf(stderr, "%s read failed\n", path_name);
 660                         free(buffer);
 661                         goto end;
 662                 }
 663
 664                 ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
 665                                                  buffer, st->st_size);
 666                 free(buffer);
 667                 goto end;
 668         }
 669
 670         /* round up our st_size to the FS blocksize */
 671         total_bytes = (u64)blocks * sectorsize;
 672
 673         /*
 674          * do our IO in extent buffers so it can work
 675          * against any raid type
 676          */
 677         eb = calloc(1, sizeof(*eb) + sectorsize);
 678         if (!eb) {
 679                 ret = -ENOMEM;
 680                 goto end;
 681         }
 682
 683 again:
 684
 685         /*
 686          * keep our extent size at 1MB max, this makes it easier to work inside
 687          * the tiny block groups created during mkfs
 688          */
 689         cur_bytes = min(total_bytes, 1024ULL * 1024);
 690         ret = btrfs_reserve_extent(trans, root, cur_bytes, 0, 0, (u64)-1,
 691                                    &key, 1);
 692         if (ret)
 693                 goto end;
 694
 695         first_block = key.objectid;
 696         bytes_read = 0;
 697
 698         while (bytes_read < cur_bytes) {
 699
 700                 memset(eb->data, 0, sectorsize);
 701
 702                 ret_read = pread64(fd, eb->data, sectorsize, file_pos + bytes_read);
 703                 if (ret_read == -1) {
 704                         fprintf(stderr, "%s read failed\n", path_name);
 705                         goto end;
 706                 }
 707
 708                 eb->start = first_block + bytes_read;
 709                 eb->len = sectorsize;
 710
 711                 /*
 712                  * we're doing the csum before we record the extent, but
 713                  * that's ok
 714                  */
 715                 ret = btrfs_csum_file_block(trans, root->fs_info->csum_root,
 716                                             first_block + bytes_read + sectorsize,
 717                                             first_block + bytes_read,
 718                                             eb->data, sectorsize);
 719                 if (ret)
 720                         goto end;
 721
 722                 ret = write_and_map_eb(trans, root, eb);
 723                 if (ret) {
 724                         fprintf(stderr, "output file write failed\n");
 725                         goto end;
 726                 }
 727
 728                 bytes_read += sectorsize;
 729         }
 730
 731         if (bytes_read) {
 732                 ret = btrfs_record_file_extent(trans, root, objectid, btrfs_inode,
 733                                                file_pos, first_block, cur_bytes);
 734                 if (ret)
 735                         goto end;
 736
 737         }
 738
 739         file_pos += cur_bytes;
 740         total_bytes -= cur_bytes;
 741
 742         if (total_bytes)
 743                 goto again;
 744
 745 end:
 746         free(eb);
 747         close(fd);
 748         return ret;
 749 }
 750
 751 static char *make_path(char *dir, char *name)
 752 {
 753         char *path;
 754
 755         path = malloc(strlen(dir) + strlen(name) + 2);
 756         if (!path)
 757                 return NULL;
 758         strcpy(path, dir);
 759         if (dir[strlen(dir) - 1] != '/')
 760                 strcat(path, "/");
 761         strcat(path, name);
 762         return path;
 763 }
 764
 765 static int traverse_directory(struct btrfs_trans_handle *trans,
 766                               struct btrfs_root *root, char *dir_name,
 767                               struct directory_name_entry *dir_head, int out_fd)
 768 {
 769         int ret = 0;
 770
 771         struct btrfs_inode_item cur_inode;
 772         struct btrfs_inode_item *inode_item;
 773         int count, i, dir_index_cnt;
 774         struct direct **files;
 775         struct stat st;
 776         struct directory_name_entry *dir_entry, *parent_dir_entry;
 777         struct direct *cur_file;
 778         ino_t parent_inum, cur_inum;
 779         ino_t highest_inum = 0;
 780         char *parent_dir_name;
 781         char real_path[PATH_MAX];
 782         struct btrfs_path path;
 783         struct extent_buffer *leaf;
 784         struct btrfs_key root_dir_key;
 785         u64 root_dir_inode_size = 0;
 786
 787         /* Add list for source directory */
 788         dir_entry = malloc(sizeof(struct directory_name_entry));
 789         if (!dir_entry)
 790                 return -ENOMEM;
 791         dir_entry->dir_name = dir_name;
 792         dir_entry->path = realpath(dir_name, real_path);
 793         if (!dir_entry->path) {
 794                 fprintf(stderr, "get directory real path error\n");
 795                 ret = -1;
 796                 goto fail_no_dir;
 797         }
 798
 799         parent_inum = highest_inum + BTRFS_FIRST_FREE_OBJECTID;
 800         dir_entry->inum = parent_inum;
 801         list_add_tail(&dir_entry->list, &dir_head->list);
 802
 803         btrfs_init_path(&path);
 804
 805         root_dir_key.objectid = btrfs_root_dirid(&root->root_item);
 806         root_dir_key.offset = 0;
 807         btrfs_set_key_type(&root_dir_key, BTRFS_INODE_ITEM_KEY);
 808         ret = btrfs_lookup_inode(trans, root, &path, &root_dir_key, 1);
 809         if (ret) {
 810                 fprintf(stderr, "root dir lookup error\n");
 811                 goto fail_no_dir;
 812         }
 813
 814         leaf = path.nodes[0];
 815         inode_item = btrfs_item_ptr(leaf, path.slots[0],
 816                                     struct btrfs_inode_item);
 817
 818         root_dir_inode_size = calculate_dir_inode_size(dir_name);
 819         btrfs_set_inode_size(leaf, inode_item, root_dir_inode_size);
 820         btrfs_mark_buffer_dirty(leaf);
 821
 822         btrfs_release_path(&path);
 823
 824         do {
 825                 parent_dir_entry = list_entry(dir_head->list.next,
 826                                               struct directory_name_entry,
 827                                               list);
 828                 list_del(&parent_dir_entry->list);
 829
 830                 parent_inum = parent_dir_entry->inum;
 831                 parent_dir_name = parent_dir_entry->dir_name;
 832                 if (chdir(parent_dir_entry->path)) {
 833                         fprintf(stderr, "chdir error for %s\n",
 834                                 parent_dir_name);
 835                         ret = -1;
 836                         goto fail_no_files;
 837                 }
 838
 839                 count = scandir(parent_dir_entry->path, &files,
 840                                 directory_select, NULL);
 841                 if (count == -1)
 842                 {
 843                         fprintf(stderr, "scandir for %s failed: %s\n",
 844                                 parent_dir_name, strerror (errno));
 845                         ret = -1;
 846                         goto fail;
 847                 }
 848
 849                 for (i = 0; i < count; i++) {
 850                         cur_file = files[i];
 851
 852                         if (lstat(cur_file->d_name, &st) == -1) {
 853                                 fprintf(stderr, "lstat failed for file %s\n",
 854                                         cur_file->d_name);
 855                                 ret = -1;
 856                                 goto fail;
 857                         }
 858
 859                         cur_inum = st.st_ino;
 860                         ret = add_directory_items(trans, root,
 861                                                   cur_inum, parent_inum,
 862                                                   cur_file->d_name,
 863                                                   &st, &dir_index_cnt);
 864                         if (ret) {
 865                                 fprintf(stderr, "add_directory_items failed\n");
 866                                 goto fail;
 867                         }
 868
 869                         ret = add_inode_items(trans, root, &st,
 870                                               cur_file->d_name, cur_inum,
 871                                               parent_inum, dir_index_cnt,
 872                                               &cur_inode);
 873                         if (ret == -EEXIST) {
 874                                 BUG_ON(st.st_nlink <= 1);
 875                                 continue;
 876                         }
 877                         if (ret) {
 878                                 fprintf(stderr, "add_inode_items failed\n");
 879                                 goto fail;
 880                         }
 881
 882                         ret = add_xattr_item(trans, root,
 883                                              cur_inum, cur_file->d_name);
 884                         if (ret) {
 885                                 fprintf(stderr, "add_xattr_item failed\n");
 886                                 if(ret != -ENOTSUP)
 887                                         goto fail;
 888                         }
 889
 890                         if (S_ISDIR(st.st_mode)) {
 891                                 dir_entry = malloc(sizeof(struct directory_name_entry));
 892                                 if (!dir_entry) {
 893                                         ret = -ENOMEM;
 894                                         goto fail;
 895                                 }
 896                                 dir_entry->dir_name = cur_file->d_name;
 897                                 dir_entry->path = make_path(parent_dir_entry->path,
 898                                                             cur_file->d_name);
 899                                 dir_entry->inum = cur_inum;
 900                                 list_add_tail(&dir_entry->list, &dir_head->list);
 901                         } else if (S_ISREG(st.st_mode)) {
 902                                 ret = add_file_items(trans, root, &cur_inode,
 903                                                      cur_inum, parent_inum, &st,
 904                                                      cur_file->d_name, out_fd);
 905                                 if (ret) {
 906                                         fprintf(stderr, "add_file_items failed\n");
 907                                         goto fail;
 908                                 }
 909                         } else if (S_ISLNK(st.st_mode)) {
 910                                 ret = add_symbolic_link(trans, root,
 911                                                         cur_inum, cur_file->d_name);
 912                                 if (ret) {
 913                                         fprintf(stderr, "add_symbolic_link failed\n");
 914                                         goto fail;
 915                                 }
 916                         }
 917                 }
 918
 919                 free_namelist(files, count);
 920                 free(parent_dir_entry);
 921
 922                 index_cnt = 2;
 923
 924         } while (!list_empty(&dir_head->list));
 925
 926 out:
 927         return !!ret;
 928 fail:
 929         free_namelist(files, count);
 930 fail_no_files:
 931         free(parent_dir_entry);
 932         goto out;
 933 fail_no_dir:
 934         free(dir_entry);
 935         goto out;
 936 }
 937
 938 static int open_target(char *output_name)
 939 {
 940         int output_fd;
 941         output_fd = open(output_name, O_CREAT | O_RDWR,
 942                          S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH);
 943
 944         return output_fd;
 945 }
 946
 947 static int create_chunks(struct btrfs_trans_handle *trans,
 948                          struct btrfs_root *root, u64 num_of_meta_chunks,
 949                          u64 size_of_data,
 950                          struct mkfs_allocation *allocation)
 951 {
 952         u64 chunk_start;
 953         u64 chunk_size;
 954         u64 meta_type = BTRFS_BLOCK_GROUP_METADATA;
 955         u64 data_type = BTRFS_BLOCK_GROUP_DATA;
 956         u64 minimum_data_chunk_size = 8 * 1024 * 1024;
 957         u64 i;
 958         int ret;
 959
 960         for (i = 0; i < num_of_meta_chunks; i++) {
 961                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 962                                         &chunk_start, &chunk_size, meta_type);
 963                 BUG_ON(ret);
 964                 ret = btrfs_make_block_group(trans, root->fs_info->extent_root, 0,
 965                                              meta_type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 966                                              chunk_start, chunk_size);
 967                 allocation->metadata += chunk_size;
 968                 BUG_ON(ret);
 969                 set_extent_dirty(&root->fs_info->free_space_cache,
 970                                  chunk_start, chunk_start + chunk_size - 1, 0);
 971         }
 972
 973         if (size_of_data < minimum_data_chunk_size)
 974                 size_of_data = minimum_data_chunk_size;
 975
 976         ret = btrfs_alloc_data_chunk(trans, root->fs_info->extent_root,
 977                                      &chunk_start, size_of_data, data_type);
 978         BUG_ON(ret);
 979         ret = btrfs_make_block_group(trans, root->fs_info->extent_root, 0,
 980                                      data_type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 981                                      chunk_start, size_of_data);
 982         allocation->data += size_of_data;
 983         BUG_ON(ret);
 984         set_extent_dirty(&root->fs_info->free_space_cache,
 985                          chunk_start, chunk_start + size_of_data - 1, 0);
 986         return ret;
 987 }
 988
 989 static int make_image(char *source_dir, struct btrfs_root *root, int out_fd)
 990 {
 991         int ret;
 992         struct btrfs_trans_handle *trans;
 993
 994         struct stat root_st;
 995
 996         struct directory_name_entry dir_head;
 997
 998         struct directory_name_entry *dir_entry = NULL;
 999
1000         ret = lstat(source_dir, &root_st);
1001         if (ret) {
1002                 fprintf(stderr, "unable to lstat the %s\n", source_dir);
1003                 goto out;
1004         }
1005
1006         INIT_LIST_HEAD(&dir_head.list);
1007
1008         trans = btrfs_start_transaction(root, 1);
1009         ret = traverse_directory(trans, root, source_dir, &dir_head, out_fd);
1010         if (ret) {
1011                 fprintf(stderr, "unable to traverse_directory\n");
1012                 goto fail;
1013         }
1014         btrfs_commit_transaction(trans, root);
1015
1016         if (verbose)
1017                 printf("Making image is completed.\n");
1018         return 0;
1019 fail:
1020         while (!list_empty(&dir_head.list)) {
1021                 dir_entry = list_entry(dir_head.list.next,
1022                                        struct directory_name_entry, list);
1023                 list_del(&dir_entry->list);
1024                 free(dir_entry);
1025         }
1026 out:
1027         fprintf(stderr, "Making image is aborted.\n");
1028         return -1;
1029 }
1030
1031 /*
1032  * This ignores symlinks with unreadable targets and subdirs that can't
1033  * be read.  It's a best-effort to give a rough estimate of the size of
1034  * a subdir.  It doesn't guarantee that prepopulating btrfs from this
1035  * tree won't still run out of space.
1036  *
1037  * The rounding up to 4096 is questionable.  Previous code used du -B 4096.
1038  */
1039 static u64 global_total_size;
1040 static int ftw_add_entry_size(const char *fpath, const struct stat *st,
1041                               int type)
1042 {
1043         if (type == FTW_F || type == FTW_D)
1044                 global_total_size += round_up(st->st_size, 4096);
1045
1046         return 0;
1047 }
1048
1049 static u64 size_sourcedir(char *dir_name, u64 sectorsize,
1050                           u64 *num_of_meta_chunks_ret, u64 *size_of_data_ret)
1051 {
1052         u64 dir_size = 0;
1053         u64 total_size = 0;
1054         int ret;
1055         u64 default_chunk_size = 8 * 1024 * 1024;       /* 8MB */
1056         u64 allocated_meta_size = 8 * 1024 * 1024;      /* 8MB */
1057         u64 allocated_total_size = 20 * 1024 * 1024;    /* 20MB */
1058         u64 num_of_meta_chunks = 0;
1059         u64 num_of_data_chunks = 0;
1060         u64 num_of_allocated_meta_chunks =
1061                         allocated_meta_size / default_chunk_size;
1062
1063         global_total_size = 0;
1064         ret = ftw(dir_name, ftw_add_entry_size, 10);
1065         dir_size = global_total_size;
1066         if (ret < 0) {
1067                 fprintf(stderr, "ftw subdir walk of '%s' failed: %s\n",
1068                         dir_name, strerror(errno));
1069                 exit(1);
1070         }
1071
1072         num_of_data_chunks = (dir_size + default_chunk_size - 1) /
1073                 default_chunk_size;
1074
1075         num_of_meta_chunks = (dir_size / 2) / default_chunk_size;
1076         if (((dir_size / 2) % default_chunk_size) != 0)
1077                 num_of_meta_chunks++;
1078         if (num_of_meta_chunks <= num_of_allocated_meta_chunks)
1079                 num_of_meta_chunks = 0;
1080         else
1081                 num_of_meta_chunks -= num_of_allocated_meta_chunks;
1082
1083         total_size = allocated_total_size +
1084                      (num_of_data_chunks * default_chunk_size) +
1085                      (num_of_meta_chunks * default_chunk_size);
1086
1087         *num_of_meta_chunks_ret = num_of_meta_chunks;
1088         *size_of_data_ret = num_of_data_chunks * default_chunk_size;
1089         return total_size;
1090 }
1091
1092 static int zero_output_file(int out_fd, u64 size)
1093 {
1094         int loop_num;
1095         u64 location = 0;
1096         char buf[4096];
1097         int ret = 0, i;
1098         ssize_t written;
1099
1100         memset(buf, 0, 4096);
1101         loop_num = size / 4096;
1102         for (i = 0; i < loop_num; i++) {
1103                 written = pwrite64(out_fd, buf, 4096, location);
1104                 if (written != 4096)
1105                         ret = -EIO;
1106                 location += 4096;
1107         }
1108         return ret;
1109 }
1110
1111 static int is_ssd(const char *file)
1112 {
1113         blkid_probe probe;
1114         char wholedisk[32];
1115         char sysfs_path[PATH_MAX];
1116         dev_t devno;
1117         int fd;
1118         char rotational;
1119         int ret;
1120
1121         probe = blkid_new_probe_from_filename(file);
1122         if (!probe)
1123                 return 0;
1124
1125         /* Device number of this disk (possibly a partition) */
1126         devno = blkid_probe_get_devno(probe);
1127         if (!devno) {
1128                 blkid_free_probe(probe);
1129                 return 0;
1130         }
1131
1132         /* Get whole disk name (not full path) for this devno */
1133         ret = blkid_devno_to_wholedisk(devno,
1134                         wholedisk, sizeof(wholedisk), NULL);
1135         if (ret) {
1136                 blkid_free_probe(probe);
1137                 return 0;
1138         }
1139
1140         snprintf(sysfs_path, PATH_MAX, "/sys/block/%s/queue/rotational",
1141                  wholedisk);
1142
1143         blkid_free_probe(probe);
1144
1145         fd = open(sysfs_path, O_RDONLY);
1146         if (fd < 0) {
1147                 return 0;
1148         }
1149
1150         if (read(fd, &rotational, sizeof(char)) < sizeof(char)) {
1151                 close(fd);
1152                 return 0;
1153         }
1154         close(fd);
1155
1156         return !atoi((const char *)&rotational);
1157 }
1158
1159 static int _cmp_device_by_id(void *priv, struct list_head *a,
1160                              struct list_head *b)
1161 {
1162         return list_entry(a, struct btrfs_device, dev_list)->devid -
1163                list_entry(b, struct btrfs_device, dev_list)->devid;
1164 }
1165
1166 static void list_all_devices(struct btrfs_root *root)
1167 {
1168         struct btrfs_fs_devices *fs_devices;
1169         struct btrfs_device *device;
1170         int number_of_devices = 0;
1171         u64 total_block_count = 0;
1172
1173         fs_devices = root->fs_info->fs_devices;
1174
1175         list_for_each_entry(device, &fs_devices->devices, dev_list)
1176                 number_of_devices++;
1177
1178         list_sort(NULL, &fs_devices->devices, _cmp_device_by_id);
1179
1180         printf("Number of devices:  %d\n", number_of_devices);
1181         /* printf("Total devices size: %10s\n", */
1182                 /* pretty_size(total_block_count)); */
1183         printf("Devices:\n");
1184         printf("   ID        SIZE  PATH\n");
1185         list_for_each_entry(device, &fs_devices->devices, dev_list) {
1186                 printf("  %3llu  %10s  %s\n",
1187                         device->devid,
1188                         pretty_size(device->total_bytes),
1189                         device->name);
1190                 total_block_count += device->total_bytes;
1191         }
1192
1193         printf("\n");
1194 }
1195
1196 static int is_temp_block_group(struct extent_buffer *node,
1197                                struct btrfs_block_group_item *bgi,
1198                                u64 data_profile, u64 meta_profile,
1199                                u64 sys_profile)
1200 {
1201         u64 flag = btrfs_disk_block_group_flags(node, bgi);
1202         u64 flag_type = flag & BTRFS_BLOCK_GROUP_TYPE_MASK;
1203         u64 flag_profile = flag & BTRFS_BLOCK_GROUP_PROFILE_MASK;
1204         u64 used = btrfs_disk_block_group_used(node, bgi);
1205
1206         /*
1207          * Chunks meets all the following conditions is a temp chunk
1208          * 1) Empty chunk
1209          * Temp chunk is always empty.
1210          *
1211          * 2) profile dismatch with mkfs profile.
1212          * Temp chunk is always in SINGLE
1213          *
1214          * 3) Size differs with mkfs_alloc
1215          * Special case for SINGLE/SINGLE btrfs.
1216          * In that case, temp data chunk and real data chunk are always empty.
1217          * So we need to use mkfs_alloc to be sure which chunk is the newly
1218          * allocated.
1219          *
1220          * Normally, new chunk size is equal to mkfs one (One chunk)
1221          * If it has multiple chunks, we just refuse to delete any one.
1222          * As they are all single, so no real problem will happen.
1223          * So only use condition 1) and 2) to judge them.
1224          */
1225         if (used != 0)
1226                 return 0;
1227         switch (flag_type) {
1228         case BTRFS_BLOCK_GROUP_DATA:
1229         case BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA:
1230                 data_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1231                 if (flag_profile != data_profile)
1232                         return 1;
1233                 break;
1234         case BTRFS_BLOCK_GROUP_METADATA:
1235                 meta_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1236                 if (flag_profile != meta_profile)
1237                         return 1;
1238                 break;
1239         case BTRFS_BLOCK_GROUP_SYSTEM:
1240                 sys_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1241                 if (flag_profile != sys_profile)
1242                         return 1;
1243                 break;
1244         }
1245         return 0;
1246 }
1247
1248 /* Note: if current is a block group, it will skip it anyway */
1249 static int next_block_group(struct btrfs_root *root,
1250                             struct btrfs_path *path)
1251 {
1252         struct btrfs_key key;
1253         int ret = 0;
1254
1255         while (1) {
1256                 ret = btrfs_next_item(root, path);
1257                 if (ret)
1258                         goto out;
1259
1260                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1261                 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
1262                         goto out;
1263         }
1264 out:
1265         return ret;
1266 }
1267
1268 /* This function will cleanup  */
1269 static int cleanup_temp_chunks(struct btrfs_fs_info *fs_info,
1270                                struct mkfs_allocation *alloc,
1271                                u64 data_profile, u64 meta_profile,
1272                                u64 sys_profile)
1273 {
1274         struct btrfs_trans_handle *trans = NULL;
1275         struct btrfs_block_group_item *bgi;
1276         struct btrfs_root *root = fs_info->extent_root;
1277         struct btrfs_key key;
1278         struct btrfs_key found_key;
1279         struct btrfs_path *path;
1280         int ret = 0;
1281
1282         path = btrfs_alloc_path();
1283         if (!path) {
1284                 ret = -ENOMEM;
1285                 goto out;
1286         }
1287
1288         trans = btrfs_start_transaction(root, 1);
1289
1290         key.objectid = 0;
1291         key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
1292         key.offset = 0;
1293
1294         while (1) {
1295                 /*
1296                  * as the rest of the loop may modify the tree, we need to
1297                  * start a new search each time.
1298                  */
1299                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
1300                 if (ret < 0)
1301                         goto out;
1302
1303                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1304                                       path->slots[0]);
1305                 if (found_key.objectid < key.objectid)
1306                         goto out;
1307                 if (found_key.type != BTRFS_BLOCK_GROUP_ITEM_KEY) {
1308                         ret = next_block_group(root, path);
1309                         if (ret < 0)
1310                                 goto out;
1311                         if (ret > 0) {
1312                                 ret = 0;
1313                                 goto out;
1314                         }
1315                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1316                                               path->slots[0]);
1317                 }
1318
1319                 bgi = btrfs_item_ptr(path->nodes[0], path->slots[0],
1320                                      struct btrfs_block_group_item);
1321                 if (is_temp_block_group(path->nodes[0], bgi,
1322                                         data_profile, meta_profile,
1323                                         sys_profile)) {
1324                         ret = btrfs_free_block_group(trans, fs_info,
1325                                         found_key.objectid, found_key.offset);
1326                         if (ret < 0)
1327                                 goto out;
1328                 }
1329                 btrfs_release_path(path);
1330                 key.objectid = found_key.objectid + found_key.offset;
1331         }
1332 out:
1333         if (trans)
1334                 btrfs_commit_transaction(trans, root);
1335         btrfs_free_path(path);
1336         return ret;
1337 }
1338
1339 int main(int ac, char **av)
1340 {
1341         char *file;
1342         struct btrfs_root *root;
1343         struct btrfs_trans_handle *trans;
1344         char *label = NULL;
1345         u64 block_count = 0;
1346         u64 dev_block_count = 0;
1347         u64 blocks[7];
1348         u64 alloc_start = 0;
1349         u64 metadata_profile = 0;
1350         u64 data_profile = 0;
1351         u32 nodesize = max_t(u32, sysconf(_SC_PAGESIZE),
1352                         BTRFS_MKFS_DEFAULT_NODE_SIZE);
1353         u32 sectorsize = 4096;
1354         u32 stripesize = 4096;
1355         int zero_end = 1;
1356         int fd;
1357         int ret;
1358         int i;
1359         int mixed = 0;
1360         int nodesize_forced = 0;
1361         int data_profile_opt = 0;
1362         int metadata_profile_opt = 0;
1363         int discard = 1;
1364         int ssd = 0;
1365         int force_overwrite = 0;
1366         char *source_dir = NULL;
1367         int source_dir_set = 0;
1368         u64 num_of_meta_chunks = 0;
1369         u64 size_of_data = 0;
1370         u64 source_dir_size = 0;
1371         int dev_cnt = 0;
1372         int saved_optind;
1373         char fs_uuid[BTRFS_UUID_UNPARSED_SIZE] = { 0 };
1374         u64 features = BTRFS_MKFS_DEFAULT_FEATURES;
1375         struct mkfs_allocation allocation = { 0 };
1376         struct btrfs_mkfs_config mkfs_cfg;
1377
1378         while(1) {
1379                 int c;
1380                 static const struct option long_options[] = {
1381                         { "alloc-start", required_argument, NULL, 'A'},
1382                         { "byte-count", required_argument, NULL, 'b' },
1383                         { "force", no_argument, NULL, 'f' },
1384                         { "leafsize", required_argument, NULL, 'l' },
1385                         { "label", required_argument, NULL, 'L'},
1386                         { "metadata", required_argument, NULL, 'm' },
1387                         { "mixed", no_argument, NULL, 'M' },
1388                         { "nodesize", required_argument, NULL, 'n' },
1389                         { "sectorsize", required_argument, NULL, 's' },
1390                         { "data", required_argument, NULL, 'd' },
1391                         { "version", no_argument, NULL, 'V' },
1392                         { "rootdir", required_argument, NULL, 'r' },
1393                         { "nodiscard", no_argument, NULL, 'K' },
1394                         { "features", required_argument, NULL, 'O' },
1395                         { "uuid", required_argument, NULL, 'U' },
1396                         { "quiet", 0, NULL, 'q' },
1397                         { "help", no_argument, NULL, GETOPT_VAL_HELP },
1398                         { NULL, 0, NULL, 0}
1399                 };
1400
1401                 c = getopt_long(ac, av, "A:b:fl:n:s:m:d:L:O:r:U:VMKq",
1402                                 long_options, NULL);
1403                 if (c < 0)
1404                         break;
1405                 switch(c) {
1406                         case 'A':
1407                                 alloc_start = parse_size(optarg);
1408                                 break;
1409                         case 'f':
1410                                 force_overwrite = 1;
1411                                 break;
1412                         case 'd':
1413                                 data_profile = parse_profile(optarg);
1414                                 data_profile_opt = 1;
1415                                 break;
1416                         case 'l':
1417                                 fprintf(stderr,
1418                         "WARNING: --leafsize is deprecated, use --nodesize\n");
1419                         case 'n':
1420                                 nodesize = parse_size(optarg);
1421                                 nodesize_forced = 1;
1422                                 break;
1423                         case 'L':
1424                                 label = parse_label(optarg);
1425                                 break;
1426                         case 'm':
1427                                 metadata_profile = parse_profile(optarg);
1428                                 metadata_profile_opt = 1;
1429                                 break;
1430                         case 'M':
1431                                 mixed = 1;
1432                                 break;
1433                         case 'O': {
1434                                 char *orig = strdup(optarg);
1435                                 char *tmp = orig;
1436
1437                                 tmp = btrfs_parse_fs_features(tmp, &features);
1438                                 if (tmp) {
1439                                         fprintf(stderr,
1440                                                 "Unrecognized filesystem feature '%s'\n",
1441                                                         tmp);
1442                                         free(orig);
1443                                         exit(1);
1444                                 }
1445                                 free(orig);
1446                                 if (features & BTRFS_FEATURE_LIST_ALL) {
1447                                         btrfs_list_all_fs_features(0);
1448                                         exit(0);
1449                                 }
1450                                 break;
1451                                 }
1452                         case 's':
1453                                 sectorsize = parse_size(optarg);
1454                                 break;
1455                         case 'b':
1456                                 block_count = parse_size(optarg);
1457                                 zero_end = 0;
1458                                 break;
1459                         case 'V':
1460                                 print_version();
1461                                 break;
1462                         case 'r':
1463                                 source_dir = optarg;
1464                                 source_dir_set = 1;
1465                                 break;
1466                         case 'U':
1467                                 strncpy(fs_uuid, optarg,
1468                                         BTRFS_UUID_UNPARSED_SIZE - 1);
1469                                 break;
1470                         case 'K':
1471                                 discard = 0;
1472                                 break;
1473                         case 'q':
1474                                 verbose = 0;
1475                                 break;
1476                         case GETOPT_VAL_HELP:
1477                         default:
1478                                 print_usage(c != GETOPT_VAL_HELP);
1479                 }
1480         }
1481
1482         if (verbose) {
1483                 printf("%s\n", PACKAGE_STRING);
1484                 printf("See %s for more information.\n\n", PACKAGE_URL);
1485         }
1486
1487         sectorsize = max(sectorsize, (u32)sysconf(_SC_PAGESIZE));
1488         saved_optind = optind;
1489         dev_cnt = ac - optind;
1490         if (dev_cnt == 0)
1491                 print_usage(1);
1492
1493         if (source_dir_set && dev_cnt > 1) {
1494                 fprintf(stderr,
1495                         "The -r option is limited to a single device\n");
1496                 exit(1);
1497         }
1498
1499         if (*fs_uuid) {
1500                 uuid_t dummy_uuid;
1501
1502                 if (uuid_parse(fs_uuid, dummy_uuid) != 0) {
1503                         fprintf(stderr, "could not parse UUID: %s\n", fs_uuid);
1504                         exit(1);
1505                 }
1506                 if (!test_uuid_unique(fs_uuid)) {
1507                         fprintf(stderr, "non-unique UUID: %s\n", fs_uuid);
1508                         exit(1);
1509                 }
1510         }
1511
1512         while (dev_cnt-- > 0) {
1513                 file = av[optind++];
1514                 if (is_block_device(file) == 1)
1515                         if (test_dev_for_mkfs(file, force_overwrite))
1516                                 exit(1);
1517         }
1518
1519         optind = saved_optind;
1520         dev_cnt = ac - optind;
1521
1522         file = av[optind++];
1523         ssd = is_ssd(file);
1524
1525         /*
1526         * Set default profiles according to number of added devices.
1527         * For mixed groups defaults are single/single.
1528         */
1529         if (!mixed) {
1530                 if (!metadata_profile_opt) {
1531                         if (dev_cnt == 1 && ssd && verbose)
1532                                 printf("Detected a SSD, turning off metadata "
1533                                 "duplication.  Mkfs with -m dup if you want to "
1534                                 "force metadata duplication.\n");
1535
1536                         metadata_profile = (dev_cnt > 1) ?
1537                                         BTRFS_BLOCK_GROUP_RAID1 : (ssd) ?
1538                                         0: BTRFS_BLOCK_GROUP_DUP;
1539                 }
1540                 if (!data_profile_opt) {
1541                         data_profile = (dev_cnt > 1) ?
1542                                 BTRFS_BLOCK_GROUP_RAID0 : 0; /* raid0 or single */
1543                 }
1544         } else {
1545                 u32 best_nodesize = max_t(u32, sysconf(_SC_PAGESIZE), sectorsize);
1546
1547                 if (metadata_profile_opt || data_profile_opt) {
1548                         if (metadata_profile != data_profile) {
1549                                 fprintf(stderr,
1550         "ERROR: With mixed block groups data and metadata profiles must be the same\n");
1551                                 exit(1);
1552                         }
1553                 }
1554
1555                 if (!nodesize_forced)
1556                         nodesize = best_nodesize;
1557         }
1558
1559         /*
1560          * FS features that can be set by other means than -O
1561          * just set the bit here
1562          */
1563         if (mixed)
1564                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS;
1565
1566         if ((data_profile | metadata_profile) &
1567             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
1568                 features |= BTRFS_FEATURE_INCOMPAT_RAID56;
1569         }
1570
1571         if (btrfs_check_nodesize(nodesize, sectorsize,
1572                                  features))
1573                 exit(1);
1574
1575         /* Check device/block_count after the nodesize is determined */
1576         if (block_count && block_count < btrfs_min_dev_size(nodesize)) {
1577                 fprintf(stderr,
1578                         "Size '%llu' is too small to make a usable filesystem\n",
1579                         block_count);
1580                 fprintf(stderr,
1581                         "Minimum size for btrfs filesystem is %llu\n",
1582                         btrfs_min_dev_size(nodesize));
1583                 exit(1);
1584         }
1585         for (i = saved_optind; i < saved_optind + dev_cnt; i++) {
1586                 char *path;
1587
1588                 path = av[i];
1589                 ret = test_minimum_size(path, nodesize);
1590                 if (ret < 0) {
1591                         fprintf(stderr, "Failed to check size for '%s': %s\n",
1592                                 path, strerror(-ret));
1593                         exit (1);
1594                 }
1595                 if (ret > 0) {
1596                         fprintf(stderr,
1597                                 "'%s' is too small to make a usable filesystem\n",
1598                                 path);
1599                         fprintf(stderr,
1600                                 "Minimum size for each btrfs device is %llu.\n",
1601                                 btrfs_min_dev_size(nodesize));
1602                         exit(1);
1603                 }
1604         }
1605         ret = test_num_disk_vs_raid(metadata_profile, data_profile,
1606                         dev_cnt, mixed);
1607         if (ret)
1608                 exit(1);
1609
1610         dev_cnt--;
1611
1612         if (!source_dir_set) {
1613                 /*
1614                  * open without O_EXCL so that the problem should not
1615                  * occur by the following processing.
1616                  * (btrfs_register_one_device() fails if O_EXCL is on)
1617                  */
1618                 fd = open(file, O_RDWR);
1619                 if (fd < 0) {
1620                         fprintf(stderr, "unable to open %s: %s\n", file,
1621                                 strerror(errno));
1622                         exit(1);
1623                 }
1624                 ret = btrfs_prepare_device(fd, file, zero_end, &dev_block_count,
1625                                            block_count, discard);
1626                 if (ret) {
1627                         close(fd);
1628                         exit(1);
1629                 }
1630                 if (block_count && block_count > dev_block_count) {
1631                         fprintf(stderr, "%s is smaller than requested size\n", file);
1632                         exit(1);
1633                 }
1634         } else {
1635                 fd = open_target(file);
1636                 if (fd < 0) {
1637                         fprintf(stderr, "unable to open the %s\n", file);
1638                         exit(1);
1639                 }
1640
1641                 source_dir_size = size_sourcedir(source_dir, sectorsize,
1642                                              &num_of_meta_chunks, &size_of_data);
1643                 if(block_count < source_dir_size)
1644                         block_count = source_dir_size;
1645                 ret = zero_output_file(fd, block_count);
1646                 if (ret) {
1647                         fprintf(stderr, "unable to zero the output file\n");
1648                         exit(1);
1649                 }
1650                 /* our "device" is the new image file */
1651                 dev_block_count = block_count;
1652         }
1653
1654         /* To create the first block group and chunk 0 in make_btrfs */
1655         if (dev_block_count < BTRFS_MKFS_SYSTEM_GROUP_SIZE) {
1656                 fprintf(stderr, "device is too small to make filesystem\n");
1657                 exit(1);
1658         }
1659
1660         blocks[0] = BTRFS_SUPER_INFO_OFFSET;
1661         for (i = 1; i < 7; i++) {
1662                 blocks[i] = BTRFS_SUPER_INFO_OFFSET + 1024 * 1024 +
1663                         nodesize * i;
1664         }
1665
1666         if (group_profile_max_safe_loss(metadata_profile) <
1667                 group_profile_max_safe_loss(data_profile)){
1668                 fprintf(stderr,
1669                         "WARNING: metatdata has lower redundancy than data!\n\n");
1670         }
1671
1672         mkfs_cfg.label = label;
1673         mkfs_cfg.fs_uuid = fs_uuid;
1674         memcpy(mkfs_cfg.blocks, blocks, sizeof(blocks));
1675         mkfs_cfg.num_bytes = dev_block_count;
1676         mkfs_cfg.nodesize = nodesize;
1677         mkfs_cfg.sectorsize = sectorsize;
1678         mkfs_cfg.stripesize = stripesize;
1679         mkfs_cfg.features = features;
1680
1681         ret = make_btrfs(fd, &mkfs_cfg);
1682         if (ret) {
1683                 fprintf(stderr, "error during mkfs: %s\n", strerror(-ret));
1684                 exit(1);
1685         }
1686
1687         root = open_ctree(file, 0, OPEN_CTREE_WRITES);
1688         if (!root) {
1689                 fprintf(stderr, "Open ctree failed\n");
1690                 close(fd);
1691                 exit(1);
1692         }
1693         root->fs_info->alloc_start = alloc_start;
1694
1695         ret = create_metadata_block_groups(root, mixed, &allocation);
1696         if (ret) {
1697                 fprintf(stderr, "failed to create default block groups\n");
1698                 exit(1);
1699         }
1700
1701         trans = btrfs_start_transaction(root, 1);
1702         if (!trans) {
1703                 fprintf(stderr, "failed to start transaction\n");
1704                 exit(1);
1705         }
1706
1707         ret = create_data_block_groups(trans, root, mixed, &allocation);
1708         if (ret) {
1709                 fprintf(stderr, "failed to create default data block groups\n");
1710                 exit(1);
1711         }
1712
1713         ret = make_root_dir(trans, root, &allocation);
1714         if (ret) {
1715                 fprintf(stderr, "failed to setup the root directory\n");
1716                 exit(1);
1717         }
1718
1719         btrfs_commit_transaction(trans, root);
1720
1721         trans = btrfs_start_transaction(root, 1);
1722         if (!trans) {
1723                 fprintf(stderr, "failed to start transaction\n");
1724                 exit(1);
1725         }
1726
1727         if (is_block_device(file) == 1)
1728                 btrfs_register_one_device(file);
1729
1730         if (dev_cnt == 0)
1731                 goto raid_groups;
1732
1733         while (dev_cnt-- > 0) {
1734                 file = av[optind++];
1735
1736                 /*
1737                  * open without O_EXCL so that the problem should not
1738                  * occur by the following processing.
1739                  * (btrfs_register_one_device() fails if O_EXCL is on)
1740                  */
1741                 fd = open(file, O_RDWR);
1742                 if (fd < 0) {
1743                         fprintf(stderr, "unable to open %s: %s\n", file,
1744                                 strerror(errno));
1745                         exit(1);
1746                 }
1747                 ret = btrfs_device_already_in_root(root, fd,
1748                                                    BTRFS_SUPER_INFO_OFFSET);
1749                 if (ret) {
1750                         fprintf(stderr, "skipping duplicate device %s in FS\n",
1751                                 file);
1752                         close(fd);
1753                         continue;
1754                 }
1755                 ret = btrfs_prepare_device(fd, file, zero_end, &dev_block_count,
1756                                            block_count, discard);
1757                 if (ret) {
1758                         close(fd);
1759                         exit(1);
1760                 }
1761
1762                 ret = btrfs_add_to_fsid(trans, root, fd, file, dev_block_count,
1763                                         sectorsize, sectorsize, sectorsize);
1764                 BUG_ON(ret);
1765                 if (verbose >= 2) {
1766                         struct btrfs_device *device;
1767
1768                         device = container_of(root->fs_info->fs_devices->devices.next,
1769                                         struct btrfs_device, dev_list);
1770                         printf("adding device %s id %llu\n", file,
1771                                 (unsigned long long)device->devid);
1772                 }
1773
1774                 if (is_block_device(file) == 1)
1775                         btrfs_register_one_device(file);
1776         }
1777
1778 raid_groups:
1779         if (!source_dir_set) {
1780                 ret = create_raid_groups(trans, root, data_profile,
1781                                  metadata_profile, mixed, &allocation);
1782                 BUG_ON(ret);
1783         }
1784
1785         ret = create_data_reloc_tree(trans, root);
1786         BUG_ON(ret);
1787
1788         btrfs_commit_transaction(trans, root);
1789
1790         if (source_dir_set) {
1791                 trans = btrfs_start_transaction(root, 1);
1792                 ret = create_chunks(trans, root,
1793                                     num_of_meta_chunks, size_of_data,
1794                                     &allocation);
1795                 BUG_ON(ret);
1796                 btrfs_commit_transaction(trans, root);
1797
1798                 ret = make_image(source_dir, root, fd);
1799                 BUG_ON(ret);
1800         }
1801         ret = cleanup_temp_chunks(root->fs_info, &allocation, data_profile,
1802                                   metadata_profile, metadata_profile);
1803         if (ret < 0) {
1804                 fprintf(stderr, "Failed to cleanup temporary chunks\n");
1805                 goto out;
1806         }
1807
1808         if (verbose) {
1809                 char features_buf[64];
1810
1811                 printf("Label:              %s\n", label);
1812                 printf("UUID:               %s\n", fs_uuid);
1813                 printf("Node size:          %u\n", nodesize);
1814                 printf("Sector size:        %u\n", sectorsize);
1815                 printf("Filesystem size:    %s\n",
1816                         pretty_size(btrfs_super_total_bytes(root->fs_info->super_copy)));
1817                 printf("Block group profiles:\n");
1818                 if (allocation.data)
1819                         printf("  Data:             %-8s %16s\n",
1820                                 btrfs_group_profile_str(data_profile),
1821                                 pretty_size(allocation.data));
1822                 if (allocation.metadata)
1823                         printf("  Metadata:         %-8s %16s\n",
1824                                 btrfs_group_profile_str(metadata_profile),
1825                                 pretty_size(allocation.metadata));
1826                 if (allocation.mixed)
1827                         printf("  Data+Metadata:    %-8s %16s\n",
1828                                 btrfs_group_profile_str(data_profile),
1829                                 pretty_size(allocation.mixed));
1830                 printf("  System:           %-8s %16s\n",
1831                         btrfs_group_profile_str(metadata_profile),
1832                         pretty_size(allocation.system));
1833                 printf("SSD detected:       %s\n", ssd ? "yes" : "no");
1834                 btrfs_parse_features_to_string(features_buf, features);
1835                 printf("Incompat features:  %s", features_buf);
1836                 printf("\n");
1837
1838                 list_all_devices(root);
1839         }
1840
1841 out:
1842         ret = close_ctree(root);
1843         BUG_ON(ret);
1844         btrfs_close_all_devices();
1845         free(label);
1846         return 0;
1847 }