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
  21 #include <sys/ioctl.h>
  22 #include <sys/mount.h>
  23 #include "ioctl.h"
  24 #include <stdio.h>
  25 #include <stdlib.h>
  26 #include <sys/types.h>
  27 #include <sys/stat.h>
  28 #include <sys/dir.h>
  29 #include <fcntl.h>
  30 #include <unistd.h>
  31 #include <getopt.h>
  32 #include <uuid/uuid.h>
  33 #include <ctype.h>
  34 #include <sys/xattr.h>
  35 #include <limits.h>
  36 #include <linux/limits.h>
  37 #include <blkid/blkid.h>
  38 #include <ftw.h>
  39 #include "ctree.h"
  40 #include "disk-io.h"
  41 #include "volumes.h"
  42 #include "transaction.h"
  43 #include "utils.h"
  44
  45 static u64 index_cnt = 2;
  46 static int verbose = 1;
  47
  48 struct directory_name_entry {
  49         char *dir_name;
  50         char *path;
  51         ino_t inum;
  52         struct list_head list;
  53 };
  54
  55 struct mkfs_allocation {
  56         u64 data;
  57         u64 metadata;
  58         u64 mixed;
  59         u64 system;
  60 };
  61
  62 static int create_metadata_block_groups(struct btrfs_root *root, int mixed,
  63                                 struct mkfs_allocation *allocation)
  64 {
  65         struct btrfs_trans_handle *trans;
  66         u64 bytes_used;
  67         u64 chunk_start = 0;
  68         u64 chunk_size = 0;
  69         int ret;
  70
  71         trans = btrfs_start_transaction(root, 1);
  72         bytes_used = btrfs_super_bytes_used(root->fs_info->super_copy);
  73
  74         root->fs_info->system_allocs = 1;
  75         ret = btrfs_make_block_group(trans, root, bytes_used,
  76                                      BTRFS_BLOCK_GROUP_SYSTEM,
  77                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
  78                                      0, BTRFS_MKFS_SYSTEM_GROUP_SIZE);
  79         allocation->system += BTRFS_MKFS_SYSTEM_GROUP_SIZE;
  80         BUG_ON(ret);
  81
  82         if (mixed) {
  83                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
  84                                         &chunk_start, &chunk_size,
  85                                         BTRFS_BLOCK_GROUP_METADATA |
  86                                         BTRFS_BLOCK_GROUP_DATA);
  87                 if (ret == -ENOSPC) {
  88                         fprintf(stderr,
  89                                 "no space to alloc data/metadata chunk\n");
  90                         goto err;
  91                 }
  92                 BUG_ON(ret);
  93                 ret = btrfs_make_block_group(trans, root, 0,
  94                                              BTRFS_BLOCK_GROUP_METADATA |
  95                                              BTRFS_BLOCK_GROUP_DATA,
  96                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
  97                                              chunk_start, chunk_size);
  98                 BUG_ON(ret);
  99                 allocation->mixed += chunk_size;
 100         } else {
 101                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 102                                         &chunk_start, &chunk_size,
 103                                         BTRFS_BLOCK_GROUP_METADATA);
 104                 if (ret == -ENOSPC) {
 105                         fprintf(stderr, "no space to alloc metadata chunk\n");
 106                         goto err;
 107                 }
 108                 BUG_ON(ret);
 109                 ret = btrfs_make_block_group(trans, root, 0,
 110                                              BTRFS_BLOCK_GROUP_METADATA,
 111                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 112                                              chunk_start, chunk_size);
 113                 allocation->metadata += chunk_size;
 114                 BUG_ON(ret);
 115         }
 116
 117         root->fs_info->system_allocs = 0;
 118         btrfs_commit_transaction(trans, root);
 119
 120 err:
 121         return ret;
 122 }
 123
 124 static int create_data_block_groups(struct btrfs_trans_handle *trans,
 125                 struct btrfs_root *root, int mixed,
 126                 struct mkfs_allocation *allocation)
 127 {
 128         u64 chunk_start = 0;
 129         u64 chunk_size = 0;
 130         int ret = 0;
 131
 132         if (!mixed) {
 133                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 134                                         &chunk_start, &chunk_size,
 135                                         BTRFS_BLOCK_GROUP_DATA);
 136                 if (ret == -ENOSPC) {
 137                         fprintf(stderr, "no space to alloc data chunk\n");
 138                         goto err;
 139                 }
 140                 BUG_ON(ret);
 141                 ret = btrfs_make_block_group(trans, root, 0,
 142                                              BTRFS_BLOCK_GROUP_DATA,
 143                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 144                                              chunk_start, chunk_size);
 145                 allocation->data += chunk_size;
 146                 BUG_ON(ret);
 147         }
 148
 149 err:
 150         return ret;
 151 }
 152
 153 static int make_root_dir(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 154                 int mixed, struct mkfs_allocation *allocation)
 155 {
 156         struct btrfs_key location;
 157         int ret;
 158
 159         ret = btrfs_make_root_dir(trans, root->fs_info->tree_root,
 160                               BTRFS_ROOT_TREE_DIR_OBJECTID);
 161         if (ret)
 162                 goto err;
 163         ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
 164         if (ret)
 165                 goto err;
 166         memcpy(&location, &root->fs_info->fs_root->root_key, sizeof(location));
 167         location.offset = (u64)-1;
 168         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
 169                         "default", 7,
 170                         btrfs_super_root_dir(root->fs_info->super_copy),
 171                         &location, BTRFS_FT_DIR, 0);
 172         if (ret)
 173                 goto err;
 174
 175         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 176                              "default", 7, location.objectid,
 177                              BTRFS_ROOT_TREE_DIR_OBJECTID, 0);
 178         if (ret)
 179                 goto err;
 180
 181 err:
 182         return ret;
 183 }
 184
 185 static void __recow_root(struct btrfs_trans_handle *trans,
 186                          struct btrfs_root *root)
 187 {
 188         int ret;
 189         struct extent_buffer *tmp;
 190
 191         if (trans->transid != btrfs_root_generation(&root->root_item)) {
 192                 extent_buffer_get(root->node);
 193                 ret = __btrfs_cow_block(trans, root, root->node,
 194                                         NULL, 0, &tmp, 0, 0);
 195                 BUG_ON(ret);
 196                 free_extent_buffer(tmp);
 197         }
 198 }
 199
 200 static void recow_roots(struct btrfs_trans_handle *trans,
 201                        struct btrfs_root *root)
 202 {
 203         struct btrfs_fs_info *info = root->fs_info;
 204
 205         __recow_root(trans, info->fs_root);
 206         __recow_root(trans, info->tree_root);
 207         __recow_root(trans, info->extent_root);
 208         __recow_root(trans, info->chunk_root);
 209         __recow_root(trans, info->dev_root);
 210         __recow_root(trans, info->csum_root);
 211 }
 212
 213 static int create_one_raid_group(struct btrfs_trans_handle *trans,
 214                               struct btrfs_root *root, u64 type,
 215                               struct mkfs_allocation *allocation)
 216
 217 {
 218         u64 chunk_start;
 219         u64 chunk_size;
 220         int ret;
 221
 222         ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 223                                 &chunk_start, &chunk_size, type);
 224         if (ret == -ENOSPC) {
 225                 fprintf(stderr, "not enough free space\n");
 226                 exit(1);
 227         }
 228         BUG_ON(ret);
 229         ret = btrfs_make_block_group(trans, root->fs_info->extent_root, 0,
 230                                      type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 231                                      chunk_start, chunk_size);
 232         if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == BTRFS_BLOCK_GROUP_DATA)
 233                 allocation->data += chunk_size;
 234         else if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == BTRFS_BLOCK_GROUP_METADATA)
 235                 allocation->metadata += chunk_size;
 236         else if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == BTRFS_BLOCK_GROUP_SYSTEM)
 237                 allocation->system += chunk_size;
 238         else if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
 239                         (BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA))
 240                 allocation->mixed += chunk_size;
 241         else
 242                 BUG_ON(1);
 243
 244         BUG_ON(ret);
 245         return ret;
 246 }
 247
 248 static int create_raid_groups(struct btrfs_trans_handle *trans,
 249                               struct btrfs_root *root, u64 data_profile,
 250                               u64 metadata_profile, int mixed,
 251                               struct mkfs_allocation *allocation)
 252 {
 253         u64 num_devices = btrfs_super_num_devices(root->fs_info->super_copy);
 254         int ret;
 255
 256         if (metadata_profile) {
 257                 u64 meta_flags = BTRFS_BLOCK_GROUP_METADATA;
 258
 259                 ret = create_one_raid_group(trans, root,
 260                                             BTRFS_BLOCK_GROUP_SYSTEM |
 261                                             metadata_profile, allocation);
 262                 BUG_ON(ret);
 263
 264                 if (mixed)
 265                         meta_flags |= BTRFS_BLOCK_GROUP_DATA;
 266
 267                 ret = create_one_raid_group(trans, root, meta_flags |
 268                                             metadata_profile, allocation);
 269                 BUG_ON(ret);
 270
 271         }
 272         if (!mixed && num_devices > 1 && data_profile) {
 273                 ret = create_one_raid_group(trans, root,
 274                                             BTRFS_BLOCK_GROUP_DATA |
 275                                             data_profile, allocation);
 276                 BUG_ON(ret);
 277         }
 278         recow_roots(trans, root);
 279
 280         return 0;
 281 }
 282
 283 static int create_data_reloc_tree(struct btrfs_trans_handle *trans,
 284                                   struct btrfs_root *root)
 285 {
 286         struct btrfs_key location;
 287         struct btrfs_root_item root_item;
 288         struct extent_buffer *tmp;
 289         u64 objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
 290         int ret;
 291
 292         ret = btrfs_copy_root(trans, root, root->node, &tmp, objectid);
 293         BUG_ON(ret);
 294
 295         memcpy(&root_item, &root->root_item, sizeof(root_item));
 296         btrfs_set_root_bytenr(&root_item, tmp->start);
 297         btrfs_set_root_level(&root_item, btrfs_header_level(tmp));
 298         btrfs_set_root_generation(&root_item, trans->transid);
 299         free_extent_buffer(tmp);
 300
 301         location.objectid = objectid;
 302         location.type = BTRFS_ROOT_ITEM_KEY;
 303         location.offset = 0;
 304         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
 305                                 &location, &root_item);
 306         BUG_ON(ret);
 307         return 0;
 308 }
 309
 310 static void print_usage(int ret)
 311 {
 312         fprintf(stderr, "usage: mkfs.btrfs [options] dev [ dev ... ]\n");
 313         fprintf(stderr, "options:\n");
 314         fprintf(stderr, "\t-A|--alloc-start START  the offset to start the FS\n");
 315         fprintf(stderr, "\t-b|--byte-count SIZE    total number of bytes in the FS\n");
 316         fprintf(stderr, "\t-d|--data PROFILE       data profile, raid0, raid1, raid5, raid6, raid10, dup or single\n");
 317         fprintf(stderr, "\t-f|--force              force overwrite of existing filesystem\n");
 318         fprintf(stderr, "\t-l|--leafsize SIZE      deprecated, alias for nodesize\n");
 319         fprintf(stderr, "\t-L|--label LABEL        set a label\n");
 320         fprintf(stderr, "\t-m|--metadata PROFILE   metadata profile, values like data profile\n");
 321         fprintf(stderr, "\t-M|--mixed              mix metadata and data together\n");
 322         fprintf(stderr, "\t-n|--nodesize SIZE      size of btree nodes\n");
 323         fprintf(stderr, "\t-s|--sectorsize SIZE    min block allocation (may not mountable by current kernel)\n");
 324         fprintf(stderr, "\t-r|--rootdir DIR        the source directory\n");
 325         fprintf(stderr, "\t-K|--nodiscard          do not perform whole device TRIM\n");
 326         fprintf(stderr, "\t-O|--features LIST      comma separated list of filesystem features, use '-O list-all' to list features\n");
 327         fprintf(stderr, "\t-U|--uuid UUID          specify the filesystem UUID\n");
 328         fprintf(stderr, "\t-q|--quiet              no messages except errors\n");
 329         fprintf(stderr, "\t-V|--version            print the mkfs.btrfs version and exit\n");
 330         fprintf(stderr, "%s\n", PACKAGE_STRING);
 331         exit(ret);
 332 }
 333
 334 static void print_version(void) __attribute__((noreturn));
 335 static void print_version(void)
 336 {
 337         fprintf(stderr, "mkfs.btrfs, part of %s\n", PACKAGE_STRING);
 338         exit(0);
 339 }
 340
 341 static u64 parse_profile(char *s)
 342 {
 343         if (strcmp(s, "raid0") == 0) {
 344                 return BTRFS_BLOCK_GROUP_RAID0;
 345         } else if (strcasecmp(s, "raid1") == 0) {
 346                 return BTRFS_BLOCK_GROUP_RAID1;
 347         } else if (strcasecmp(s, "raid5") == 0) {
 348                 return BTRFS_BLOCK_GROUP_RAID5;
 349         } else if (strcasecmp(s, "raid6") == 0) {
 350                 return BTRFS_BLOCK_GROUP_RAID6;
 351         } else if (strcasecmp(s, "raid10") == 0) {
 352                 return BTRFS_BLOCK_GROUP_RAID10;
 353         } else if (strcasecmp(s, "dup") == 0) {
 354                 return BTRFS_BLOCK_GROUP_DUP;
 355         } else if (strcasecmp(s, "single") == 0) {
 356                 return 0;
 357         } else {
 358                 fprintf(stderr, "Unknown profile %s\n", s);
 359                 exit(1);
 360         }
 361         /* not reached */
 362         return 0;
 363 }
 364
 365 static char *parse_label(char *input)
 366 {
 367         int len = strlen(input);
 368
 369         if (len >= BTRFS_LABEL_SIZE) {
 370                 fprintf(stderr, "Label %s is too long (max %d)\n", input,
 371                         BTRFS_LABEL_SIZE - 1);
 372                 exit(1);
 373         }
 374         return strdup(input);
 375 }
 376
 377 static int add_directory_items(struct btrfs_trans_handle *trans,
 378                                struct btrfs_root *root, u64 objectid,
 379                                ino_t parent_inum, const char *name,
 380                                struct stat *st, int *dir_index_cnt)
 381 {
 382         int ret;
 383         int name_len;
 384         struct btrfs_key location;
 385         u8 filetype = 0;
 386
 387         name_len = strlen(name);
 388
 389         location.objectid = objectid;
 390         location.offset = 0;
 391         btrfs_set_key_type(&location, BTRFS_INODE_ITEM_KEY);
 392
 393         if (S_ISDIR(st->st_mode))
 394                 filetype = BTRFS_FT_DIR;
 395         if (S_ISREG(st->st_mode))
 396                 filetype = BTRFS_FT_REG_FILE;
 397         if (S_ISLNK(st->st_mode))
 398                 filetype = BTRFS_FT_SYMLINK;
 399
 400         ret = btrfs_insert_dir_item(trans, root, name, name_len,
 401                                     parent_inum, &location,
 402                                     filetype, index_cnt);
 403         if (ret)
 404                 return ret;
 405         ret = btrfs_insert_inode_ref(trans, root, name, name_len,
 406                                      objectid, parent_inum, index_cnt);
 407         *dir_index_cnt = index_cnt;
 408         index_cnt++;
 409
 410         return ret;
 411 }
 412
 413 static int fill_inode_item(struct btrfs_trans_handle *trans,
 414                            struct btrfs_root *root,
 415                            struct btrfs_inode_item *dst, struct stat *src)
 416 {
 417         u64 blocks = 0;
 418         u64 sectorsize = root->sectorsize;
 419
 420         /*
 421          * btrfs_inode_item has some reserved fields
 422          * and represents on-disk inode entry, so
 423          * zero everything to prevent information leak
 424          */
 425         memset(dst, 0, sizeof (*dst));
 426
 427         btrfs_set_stack_inode_generation(dst, trans->transid);
 428         btrfs_set_stack_inode_size(dst, src->st_size);
 429         btrfs_set_stack_inode_nbytes(dst, 0);
 430         btrfs_set_stack_inode_block_group(dst, 0);
 431         btrfs_set_stack_inode_nlink(dst, src->st_nlink);
 432         btrfs_set_stack_inode_uid(dst, src->st_uid);
 433         btrfs_set_stack_inode_gid(dst, src->st_gid);
 434         btrfs_set_stack_inode_mode(dst, src->st_mode);
 435         btrfs_set_stack_inode_rdev(dst, 0);
 436         btrfs_set_stack_inode_flags(dst, 0);
 437         btrfs_set_stack_timespec_sec(&dst->atime, src->st_atime);
 438         btrfs_set_stack_timespec_nsec(&dst->atime, 0);
 439         btrfs_set_stack_timespec_sec(&dst->ctime, src->st_ctime);
 440         btrfs_set_stack_timespec_nsec(&dst->ctime, 0);
 441         btrfs_set_stack_timespec_sec(&dst->mtime, src->st_mtime);
 442         btrfs_set_stack_timespec_nsec(&dst->mtime, 0);
 443         btrfs_set_stack_timespec_sec(&dst->otime, 0);
 444         btrfs_set_stack_timespec_nsec(&dst->otime, 0);
 445
 446         if (S_ISDIR(src->st_mode)) {
 447                 btrfs_set_stack_inode_size(dst, 0);
 448                 btrfs_set_stack_inode_nlink(dst, 1);
 449         }
 450         if (S_ISREG(src->st_mode)) {
 451                 btrfs_set_stack_inode_size(dst, (u64)src->st_size);
 452                 if (src->st_size <= BTRFS_MAX_INLINE_DATA_SIZE(root))
 453                         btrfs_set_stack_inode_nbytes(dst, src->st_size);
 454                 else {
 455                         blocks = src->st_size / sectorsize;
 456                         if (src->st_size % sectorsize)
 457                                 blocks += 1;
 458                         blocks *= sectorsize;
 459                         btrfs_set_stack_inode_nbytes(dst, blocks);
 460                 }
 461         }
 462         if (S_ISLNK(src->st_mode))
 463                 btrfs_set_stack_inode_nbytes(dst, src->st_size + 1);
 464
 465         return 0;
 466 }
 467
 468 static int directory_select(const struct direct *entry)
 469 {
 470         if ((strncmp(entry->d_name, ".", entry->d_reclen) == 0) ||
 471                 (strncmp(entry->d_name, "..", entry->d_reclen) == 0))
 472                 return 0;
 473         else
 474                 return 1;
 475 }
 476
 477 static void free_namelist(struct direct **files, int count)
 478 {
 479         int i;
 480
 481         if (count < 0)
 482                 return;
 483
 484         for (i = 0; i < count; ++i)
 485                 free(files[i]);
 486         free(files);
 487 }
 488
 489 static u64 calculate_dir_inode_size(char *dirname)
 490 {
 491         int count, i;
 492         struct direct **files, *cur_file;
 493         u64 dir_inode_size = 0;
 494
 495         count = scandir(dirname, &files, directory_select, NULL);
 496
 497         for (i = 0; i < count; i++) {
 498                 cur_file = files[i];
 499                 dir_inode_size += strlen(cur_file->d_name);
 500         }
 501
 502         free_namelist(files, count);
 503
 504         dir_inode_size *= 2;
 505         return dir_inode_size;
 506 }
 507
 508 static int add_inode_items(struct btrfs_trans_handle *trans,
 509                            struct btrfs_root *root,
 510                            struct stat *st, char *name,
 511                            u64 self_objectid, ino_t parent_inum,
 512                            int dir_index_cnt, struct btrfs_inode_item *inode_ret)
 513 {
 514         int ret;
 515         struct btrfs_key inode_key;
 516         struct btrfs_inode_item btrfs_inode;
 517         u64 objectid;
 518         u64 inode_size = 0;
 519
 520         fill_inode_item(trans, root, &btrfs_inode, st);
 521         objectid = self_objectid;
 522
 523         if (S_ISDIR(st->st_mode)) {
 524                 inode_size = calculate_dir_inode_size(name);
 525                 btrfs_set_stack_inode_size(&btrfs_inode, inode_size);
 526         }
 527
 528         inode_key.objectid = objectid;
 529         inode_key.offset = 0;
 530         btrfs_set_key_type(&inode_key, BTRFS_INODE_ITEM_KEY);
 531
 532         ret = btrfs_insert_inode(trans, root, objectid, &btrfs_inode);
 533
 534         *inode_ret = btrfs_inode;
 535         return ret;
 536 }
 537
 538 static int add_xattr_item(struct btrfs_trans_handle *trans,
 539                           struct btrfs_root *root, u64 objectid,
 540                           const char *file_name)
 541 {
 542         int ret;
 543         int cur_name_len;
 544         char xattr_list[XATTR_LIST_MAX];
 545         char *cur_name;
 546         char cur_value[XATTR_SIZE_MAX];
 547         char delimiter = '\0';
 548         char *next_location = xattr_list;
 549
 550         ret = llistxattr(file_name, xattr_list, XATTR_LIST_MAX);
 551         if (ret < 0) {
 552                 if(errno == ENOTSUP)
 553                         return 0;
 554                 fprintf(stderr, "get a list of xattr failed for %s\n",
 555                         file_name);
 556                 return ret;
 557         }
 558         if (ret == 0)
 559                 return ret;
 560
 561         cur_name = strtok(xattr_list, &delimiter);
 562         while (cur_name != NULL) {
 563                 cur_name_len = strlen(cur_name);
 564                 next_location += cur_name_len + 1;
 565
 566                 ret = getxattr(file_name, cur_name, cur_value, XATTR_SIZE_MAX);
 567                 if (ret < 0) {
 568                         if(errno == ENOTSUP)
 569                                 return 0;
 570                         fprintf(stderr, "get a xattr value failed for %s attr %s\n",
 571                                 file_name, cur_name);
 572                         return ret;
 573                 }
 574
 575                 ret = btrfs_insert_xattr_item(trans, root, cur_name,
 576                                               cur_name_len, cur_value,
 577                                               ret, objectid);
 578                 if (ret) {
 579                         fprintf(stderr, "insert a xattr item failed for %s\n",
 580                                 file_name);
 581                 }
 582
 583                 cur_name = strtok(next_location, &delimiter);
 584         }
 585
 586         return ret;
 587 }
 588
 589 static int add_symbolic_link(struct btrfs_trans_handle *trans,
 590                              struct btrfs_root *root,
 591                              u64 objectid, const char *path_name)
 592 {
 593         int ret;
 594         u64 sectorsize = root->sectorsize;
 595         char *buf = malloc(sectorsize);
 596
 597         ret = readlink(path_name, buf, sectorsize);
 598         if (ret <= 0) {
 599                 fprintf(stderr, "readlink failed for %s\n", path_name);
 600                 goto fail;
 601         }
 602         if (ret >= sectorsize) {
 603                 fprintf(stderr, "symlink too long for %s", path_name);
 604                 ret = -1;
 605                 goto fail;
 606         }
 607
 608         buf[ret] = '\0'; /* readlink does not do it for us */
 609         ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
 610                                          buf, ret + 1);
 611 fail:
 612         free(buf);
 613         return ret;
 614 }
 615
 616 static int add_file_items(struct btrfs_trans_handle *trans,
 617                           struct btrfs_root *root,
 618                           struct btrfs_inode_item *btrfs_inode, u64 objectid,
 619                           ino_t parent_inum, struct stat *st,
 620                           const char *path_name, int out_fd)
 621 {
 622         int ret = -1;
 623         ssize_t ret_read;
 624         u64 bytes_read = 0;
 625         struct btrfs_key key;
 626         int blocks;
 627         u32 sectorsize = root->sectorsize;
 628         u64 first_block = 0;
 629         u64 file_pos = 0;
 630         u64 cur_bytes;
 631         u64 total_bytes;
 632         struct extent_buffer *eb = NULL;
 633         int fd;
 634
 635         if (st->st_size == 0)
 636                 return 0;
 637
 638         fd = open(path_name, O_RDONLY);
 639         if (fd == -1) {
 640                 fprintf(stderr, "%s open failed\n", path_name);
 641                 return ret;
 642         }
 643
 644         blocks = st->st_size / sectorsize;
 645         if (st->st_size % sectorsize)
 646                 blocks += 1;
 647
 648         if (st->st_size <= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 649                 char *buffer = malloc(st->st_size);
 650                 ret_read = pread64(fd, buffer, st->st_size, bytes_read);
 651                 if (ret_read == -1) {
 652                         fprintf(stderr, "%s read failed\n", path_name);
 653                         free(buffer);
 654                         goto end;
 655                 }
 656
 657                 ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
 658                                                  buffer, st->st_size);
 659                 free(buffer);
 660                 goto end;
 661         }
 662
 663         /* round up our st_size to the FS blocksize */
 664         total_bytes = (u64)blocks * sectorsize;
 665
 666         /*
 667          * do our IO in extent buffers so it can work
 668          * against any raid type
 669          */
 670         eb = malloc(sizeof(*eb) + sectorsize);
 671         if (!eb) {
 672                 ret = -ENOMEM;
 673                 goto end;
 674         }
 675         memset(eb, 0, sizeof(*eb) + sectorsize);
 676
 677 again:
 678
 679         /*
 680          * keep our extent size at 1MB max, this makes it easier to work inside
 681          * the tiny block groups created during mkfs
 682          */
 683         cur_bytes = min(total_bytes, 1024ULL * 1024);
 684         ret = btrfs_reserve_extent(trans, root, cur_bytes, 0, 0, (u64)-1,
 685                                    &key, 1);
 686         if (ret)
 687                 goto end;
 688
 689         first_block = key.objectid;
 690         bytes_read = 0;
 691
 692         while (bytes_read < cur_bytes) {
 693
 694                 memset(eb->data, 0, sectorsize);
 695
 696                 ret_read = pread64(fd, eb->data, sectorsize, file_pos + bytes_read);
 697                 if (ret_read == -1) {
 698                         fprintf(stderr, "%s read failed\n", path_name);
 699                         goto end;
 700                 }
 701
 702                 eb->start = first_block + bytes_read;
 703                 eb->len = sectorsize;
 704
 705                 /*
 706                  * we're doing the csum before we record the extent, but
 707                  * that's ok
 708                  */
 709                 ret = btrfs_csum_file_block(trans, root->fs_info->csum_root,
 710                                             first_block + bytes_read + sectorsize,
 711                                             first_block + bytes_read,
 712                                             eb->data, sectorsize);
 713                 if (ret)
 714                         goto end;
 715
 716                 ret = write_and_map_eb(trans, root, eb);
 717                 if (ret) {
 718                         fprintf(stderr, "output file write failed\n");
 719                         goto end;
 720                 }
 721
 722                 bytes_read += sectorsize;
 723         }
 724
 725         if (bytes_read) {
 726                 ret = btrfs_record_file_extent(trans, root, objectid, btrfs_inode,
 727                                                file_pos, first_block, cur_bytes);
 728                 if (ret)
 729                         goto end;
 730
 731         }
 732
 733         file_pos += cur_bytes;
 734         total_bytes -= cur_bytes;
 735
 736         if (total_bytes)
 737                 goto again;
 738
 739 end:
 740         free(eb);
 741         close(fd);
 742         return ret;
 743 }
 744
 745 static char *make_path(char *dir, char *name)
 746 {
 747         char *path;
 748
 749         path = malloc(strlen(dir) + strlen(name) + 2);
 750         if (!path)
 751                 return NULL;
 752         strcpy(path, dir);
 753         if (dir[strlen(dir) - 1] != '/')
 754                 strcat(path, "/");
 755         strcat(path, name);
 756         return path;
 757 }
 758
 759 static int traverse_directory(struct btrfs_trans_handle *trans,
 760                               struct btrfs_root *root, char *dir_name,
 761                               struct directory_name_entry *dir_head, int out_fd)
 762 {
 763         int ret = 0;
 764
 765         struct btrfs_inode_item cur_inode;
 766         struct btrfs_inode_item *inode_item;
 767         int count, i, dir_index_cnt;
 768         struct direct **files;
 769         struct stat st;
 770         struct directory_name_entry *dir_entry, *parent_dir_entry;
 771         struct direct *cur_file;
 772         ino_t parent_inum, cur_inum;
 773         ino_t highest_inum = 0;
 774         char *parent_dir_name;
 775         char real_path[PATH_MAX];
 776         struct btrfs_path path;
 777         struct extent_buffer *leaf;
 778         struct btrfs_key root_dir_key;
 779         u64 root_dir_inode_size = 0;
 780
 781         /* Add list for source directory */
 782         dir_entry = malloc(sizeof(struct directory_name_entry));
 783         dir_entry->dir_name = dir_name;
 784         dir_entry->path = realpath(dir_name, real_path);
 785         if (!dir_entry->path) {
 786                 fprintf(stderr, "get directory real path error\n");
 787                 ret = -1;
 788                 goto fail_no_dir;
 789         }
 790
 791         parent_inum = highest_inum + BTRFS_FIRST_FREE_OBJECTID;
 792         dir_entry->inum = parent_inum;
 793         list_add_tail(&dir_entry->list, &dir_head->list);
 794
 795         btrfs_init_path(&path);
 796
 797         root_dir_key.objectid = btrfs_root_dirid(&root->root_item);
 798         root_dir_key.offset = 0;
 799         btrfs_set_key_type(&root_dir_key, BTRFS_INODE_ITEM_KEY);
 800         ret = btrfs_lookup_inode(trans, root, &path, &root_dir_key, 1);
 801         if (ret) {
 802                 fprintf(stderr, "root dir lookup error\n");
 803                 goto fail_no_dir;
 804         }
 805
 806         leaf = path.nodes[0];
 807         inode_item = btrfs_item_ptr(leaf, path.slots[0],
 808                                     struct btrfs_inode_item);
 809
 810         root_dir_inode_size = calculate_dir_inode_size(dir_name);
 811         btrfs_set_inode_size(leaf, inode_item, root_dir_inode_size);
 812         btrfs_mark_buffer_dirty(leaf);
 813
 814         btrfs_release_path(&path);
 815
 816         do {
 817                 parent_dir_entry = list_entry(dir_head->list.next,
 818                                               struct directory_name_entry,
 819                                               list);
 820                 list_del(&parent_dir_entry->list);
 821
 822                 parent_inum = parent_dir_entry->inum;
 823                 parent_dir_name = parent_dir_entry->dir_name;
 824                 if (chdir(parent_dir_entry->path)) {
 825                         fprintf(stderr, "chdir error for %s\n",
 826                                 parent_dir_name);
 827                         ret = -1;
 828                         goto fail_no_files;
 829                 }
 830
 831                 count = scandir(parent_dir_entry->path, &files,
 832                                 directory_select, NULL);
 833                 if (count == -1)
 834                 {
 835                         fprintf(stderr, "scandir for %s failed: %s\n",
 836                                 parent_dir_name, strerror (errno));
 837                         ret = -1;
 838                         goto fail;
 839                 }
 840
 841                 for (i = 0; i < count; i++) {
 842                         cur_file = files[i];
 843
 844                         if (lstat(cur_file->d_name, &st) == -1) {
 845                                 fprintf(stderr, "lstat failed for file %s\n",
 846                                         cur_file->d_name);
 847                                 ret = -1;
 848                                 goto fail;
 849                         }
 850
 851                         cur_inum = st.st_ino;
 852                         ret = add_directory_items(trans, root,
 853                                                   cur_inum, parent_inum,
 854                                                   cur_file->d_name,
 855                                                   &st, &dir_index_cnt);
 856                         if (ret) {
 857                                 fprintf(stderr, "add_directory_items failed\n");
 858                                 goto fail;
 859                         }
 860
 861                         ret = add_inode_items(trans, root, &st,
 862                                               cur_file->d_name, cur_inum,
 863                                               parent_inum, dir_index_cnt,
 864                                               &cur_inode);
 865                         if (ret == -EEXIST) {
 866                                 BUG_ON(st.st_nlink <= 1);
 867                                 continue;
 868                         }
 869                         if (ret) {
 870                                 fprintf(stderr, "add_inode_items failed\n");
 871                                 goto fail;
 872                         }
 873
 874                         ret = add_xattr_item(trans, root,
 875                                              cur_inum, cur_file->d_name);
 876                         if (ret) {
 877                                 fprintf(stderr, "add_xattr_item failed\n");
 878                                 if(ret != -ENOTSUP)
 879                                         goto fail;
 880                         }
 881
 882                         if (S_ISDIR(st.st_mode)) {
 883                                 dir_entry = malloc(sizeof(struct directory_name_entry));
 884                                 dir_entry->dir_name = cur_file->d_name;
 885                                 dir_entry->path = make_path(parent_dir_entry->path,
 886                                                             cur_file->d_name);
 887                                 dir_entry->inum = cur_inum;
 888                                 list_add_tail(&dir_entry->list, &dir_head->list);
 889                         } else if (S_ISREG(st.st_mode)) {
 890                                 ret = add_file_items(trans, root, &cur_inode,
 891                                                      cur_inum, parent_inum, &st,
 892                                                      cur_file->d_name, out_fd);
 893                                 if (ret) {
 894                                         fprintf(stderr, "add_file_items failed\n");
 895                                         goto fail;
 896                                 }
 897                         } else if (S_ISLNK(st.st_mode)) {
 898                                 ret = add_symbolic_link(trans, root,
 899                                                         cur_inum, cur_file->d_name);
 900                                 if (ret) {
 901                                         fprintf(stderr, "add_symbolic_link failed\n");
 902                                         goto fail;
 903                                 }
 904                         }
 905                 }
 906
 907                 free_namelist(files, count);
 908                 free(parent_dir_entry);
 909
 910                 index_cnt = 2;
 911
 912         } while (!list_empty(&dir_head->list));
 913
 914 out:
 915         return !!ret;
 916 fail:
 917         free_namelist(files, count);
 918 fail_no_files:
 919         free(parent_dir_entry);
 920         goto out;
 921 fail_no_dir:
 922         free(dir_entry);
 923         goto out;
 924 }
 925
 926 static int open_target(char *output_name)
 927 {
 928         int output_fd;
 929         output_fd = open(output_name, O_CREAT | O_RDWR | O_TRUNC,
 930                          S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH);
 931
 932         return output_fd;
 933 }
 934
 935 static int create_chunks(struct btrfs_trans_handle *trans,
 936                          struct btrfs_root *root, u64 num_of_meta_chunks,
 937                          u64 size_of_data,
 938                          struct mkfs_allocation *allocation)
 939 {
 940         u64 chunk_start;
 941         u64 chunk_size;
 942         u64 meta_type = BTRFS_BLOCK_GROUP_METADATA;
 943         u64 data_type = BTRFS_BLOCK_GROUP_DATA;
 944         u64 minimum_data_chunk_size = 8 * 1024 * 1024;
 945         u64 i;
 946         int ret;
 947
 948         for (i = 0; i < num_of_meta_chunks; i++) {
 949                 ret = btrfs_alloc_chunk(trans, root->fs_info->extent_root,
 950                                         &chunk_start, &chunk_size, meta_type);
 951                 BUG_ON(ret);
 952                 ret = btrfs_make_block_group(trans, root->fs_info->extent_root, 0,
 953                                              meta_type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 954                                              chunk_start, chunk_size);
 955                 allocation->metadata += chunk_size;
 956                 BUG_ON(ret);
 957                 set_extent_dirty(&root->fs_info->free_space_cache,
 958                                  chunk_start, chunk_start + chunk_size - 1, 0);
 959         }
 960
 961         if (size_of_data < minimum_data_chunk_size)
 962                 size_of_data = minimum_data_chunk_size;
 963
 964         ret = btrfs_alloc_data_chunk(trans, root->fs_info->extent_root,
 965                                      &chunk_start, size_of_data, data_type);
 966         BUG_ON(ret);
 967         ret = btrfs_make_block_group(trans, root->fs_info->extent_root, 0,
 968                                      data_type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 969                                      chunk_start, size_of_data);
 970         allocation->data += size_of_data;
 971         BUG_ON(ret);
 972         set_extent_dirty(&root->fs_info->free_space_cache,
 973                          chunk_start, chunk_start + size_of_data - 1, 0);
 974         return ret;
 975 }
 976
 977 static int make_image(char *source_dir, struct btrfs_root *root, int out_fd)
 978 {
 979         int ret;
 980         struct btrfs_trans_handle *trans;
 981
 982         struct stat root_st;
 983
 984         struct directory_name_entry dir_head;
 985
 986         struct directory_name_entry *dir_entry = NULL;
 987
 988         ret = lstat(source_dir, &root_st);
 989         if (ret) {
 990                 fprintf(stderr, "unable to lstat the %s\n", source_dir);
 991                 goto out;
 992         }
 993
 994         INIT_LIST_HEAD(&dir_head.list);
 995
 996         trans = btrfs_start_transaction(root, 1);
 997         ret = traverse_directory(trans, root, source_dir, &dir_head, out_fd);
 998         if (ret) {
 999                 fprintf(stderr, "unable to traverse_directory\n");
1000                 goto fail;
1001         }
1002         btrfs_commit_transaction(trans, root);
1003
1004         if (verbose)
1005                 printf("Making image is completed.\n");
1006         return 0;
1007 fail:
1008         while (!list_empty(&dir_head.list)) {
1009                 dir_entry = list_entry(dir_head.list.next,
1010                                        struct directory_name_entry, list);
1011                 list_del(&dir_entry->list);
1012                 free(dir_entry);
1013         }
1014 out:
1015         fprintf(stderr, "Making image is aborted.\n");
1016         return -1;
1017 }
1018
1019 /*
1020  * This ignores symlinks with unreadable targets and subdirs that can't
1021  * be read.  It's a best-effort to give a rough estimate of the size of
1022  * a subdir.  It doesn't guarantee that prepopulating btrfs from this
1023  * tree won't still run out of space.
1024  *
1025  * The rounding up to 4096 is questionable.  Previous code used du -B 4096.
1026  */
1027 static u64 global_total_size;
1028 static int ftw_add_entry_size(const char *fpath, const struct stat *st,
1029                               int type)
1030 {
1031         if (type == FTW_F || type == FTW_D)
1032                 global_total_size += round_up(st->st_size, 4096);
1033
1034         return 0;
1035 }
1036
1037 static u64 size_sourcedir(char *dir_name, u64 sectorsize,
1038                           u64 *num_of_meta_chunks_ret, u64 *size_of_data_ret)
1039 {
1040         u64 dir_size = 0;
1041         u64 total_size = 0;
1042         int ret;
1043         u64 default_chunk_size = 8 * 1024 * 1024;       /* 8MB */
1044         u64 allocated_meta_size = 8 * 1024 * 1024;      /* 8MB */
1045         u64 allocated_total_size = 20 * 1024 * 1024;    /* 20MB */
1046         u64 num_of_meta_chunks = 0;
1047         u64 num_of_data_chunks = 0;
1048         u64 num_of_allocated_meta_chunks =
1049                         allocated_meta_size / default_chunk_size;
1050
1051         global_total_size = 0;
1052         ret = ftw(dir_name, ftw_add_entry_size, 10);
1053         dir_size = global_total_size;
1054         if (ret < 0) {
1055                 fprintf(stderr, "ftw subdir walk of '%s' failed: %s\n",
1056                         dir_name, strerror(errno));
1057                 exit(1);
1058         }
1059
1060         num_of_data_chunks = (dir_size + default_chunk_size - 1) /
1061                 default_chunk_size;
1062
1063         num_of_meta_chunks = (dir_size / 2) / default_chunk_size;
1064         if (((dir_size / 2) % default_chunk_size) != 0)
1065                 num_of_meta_chunks++;
1066         if (num_of_meta_chunks <= num_of_allocated_meta_chunks)
1067                 num_of_meta_chunks = 0;
1068         else
1069                 num_of_meta_chunks -= num_of_allocated_meta_chunks;
1070
1071         total_size = allocated_total_size +
1072                      (num_of_data_chunks * default_chunk_size) +
1073                      (num_of_meta_chunks * default_chunk_size);
1074
1075         *num_of_meta_chunks_ret = num_of_meta_chunks;
1076         *size_of_data_ret = num_of_data_chunks * default_chunk_size;
1077         return total_size;
1078 }
1079
1080 static int zero_output_file(int out_fd, u64 size, u32 sectorsize)
1081 {
1082         int len = sectorsize;
1083         int loop_num = size / sectorsize;
1084         u64 location = 0;
1085         char *buf = malloc(len);
1086         int ret = 0, i;
1087         ssize_t written;
1088
1089         if (!buf)
1090                 return -ENOMEM;
1091         memset(buf, 0, len);
1092         for (i = 0; i < loop_num; i++) {
1093                 written = pwrite64(out_fd, buf, len, location);
1094                 if (written != len)
1095                         ret = -EIO;
1096                 location += sectorsize;
1097         }
1098         free(buf);
1099         return ret;
1100 }
1101
1102 static int is_ssd(const char *file)
1103 {
1104         blkid_probe probe;
1105         char wholedisk[32];
1106         char sysfs_path[PATH_MAX];
1107         dev_t devno;
1108         int fd;
1109         char rotational;
1110         int ret;
1111
1112         probe = blkid_new_probe_from_filename(file);
1113         if (!probe)
1114                 return 0;
1115
1116         /* Device number of this disk (possibly a partition) */
1117         devno = blkid_probe_get_devno(probe);
1118         if (!devno) {
1119                 blkid_free_probe(probe);
1120                 return 0;
1121         }
1122
1123         /* Get whole disk name (not full path) for this devno */
1124         ret = blkid_devno_to_wholedisk(devno,
1125                         wholedisk, sizeof(wholedisk), NULL);
1126         if (ret) {
1127                 blkid_free_probe(probe);
1128                 return 0;
1129         }
1130
1131         snprintf(sysfs_path, PATH_MAX, "/sys/block/%s/queue/rotational",
1132                  wholedisk);
1133
1134         blkid_free_probe(probe);
1135
1136         fd = open(sysfs_path, O_RDONLY);
1137         if (fd < 0) {
1138                 return 0;
1139         }
1140
1141         if (read(fd, &rotational, sizeof(char)) < sizeof(char)) {
1142                 close(fd);
1143                 return 0;
1144         }
1145         close(fd);
1146
1147         return !atoi((const char *)&rotational);
1148 }
1149
1150 static void list_all_devices(struct btrfs_root *root)
1151 {
1152         struct btrfs_fs_devices *fs_devices;
1153         struct btrfs_device *device;
1154         int number_of_devices = 0;
1155         u64 total_block_count = 0;
1156
1157         fs_devices = root->fs_info->fs_devices;
1158
1159         list_for_each_entry(device, &fs_devices->devices, dev_list)
1160                 number_of_devices++;
1161
1162         printf("Number of devices:  %d\n", number_of_devices);
1163         /* printf("Total devices size: %10s\n", */
1164                 /* pretty_size(total_block_count)); */
1165         printf("Devices:\n");
1166         printf("   ID        SIZE  PATH\n");
1167         list_for_each_entry_reverse(device, &fs_devices->devices, dev_list) {
1168                 char dev_uuid[BTRFS_UUID_UNPARSED_SIZE];
1169
1170                 uuid_unparse(device->uuid, dev_uuid);
1171                 printf("  %3llu  %10s  %s\n",
1172                         device->devid,
1173                         pretty_size(device->total_bytes),
1174                         device->name);
1175                 total_block_count += device->total_bytes;
1176         }
1177
1178         printf("\n");
1179 }
1180
1181 static int is_temp_block_group(struct extent_buffer *node,
1182                                struct btrfs_block_group_item *bgi,
1183                                u64 data_profile, u64 meta_profile,
1184                                u64 sys_profile)
1185 {
1186         u64 flag = btrfs_disk_block_group_flags(node, bgi);
1187         u64 flag_type = flag & BTRFS_BLOCK_GROUP_TYPE_MASK;
1188         u64 flag_profile = flag & BTRFS_BLOCK_GROUP_PROFILE_MASK;
1189         u64 used = btrfs_disk_block_group_used(node, bgi);
1190
1191         /*
1192          * Chunks meets all the following conditions is a temp chunk
1193          * 1) Empty chunk
1194          * Temp chunk is always empty.
1195          *
1196          * 2) profile dismatch with mkfs profile.
1197          * Temp chunk is always in SINGLE
1198          *
1199          * 3) Size differs with mkfs_alloc
1200          * Special case for SINGLE/SINGLE btrfs.
1201          * In that case, temp data chunk and real data chunk are always empty.
1202          * So we need to use mkfs_alloc to be sure which chunk is the newly
1203          * allocated.
1204          *
1205          * Normally, new chunk size is equal to mkfs one (One chunk)
1206          * If it has multiple chunks, we just refuse to delete any one.
1207          * As they are all single, so no real problem will happen.
1208          * So only use condition 1) and 2) to judge them.
1209          */
1210         if (used != 0)
1211                 return 0;
1212         switch (flag_type) {
1213         case BTRFS_BLOCK_GROUP_DATA:
1214         case BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA:
1215                 data_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1216                 if (flag_profile != data_profile)
1217                         return 1;
1218                 break;
1219         case BTRFS_BLOCK_GROUP_METADATA:
1220                 meta_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1221                 if (flag_profile != meta_profile)
1222                         return 1;
1223                 break;
1224         case BTRFS_BLOCK_GROUP_SYSTEM:
1225                 sys_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
1226                 if (flag_profile != sys_profile)
1227                         return 1;
1228                 break;
1229         }
1230         return 0;
1231 }
1232
1233 /* Note: if current is a block group, it will skip it anyway */
1234 static int next_block_group(struct btrfs_root *root,
1235                             struct btrfs_path *path)
1236 {
1237         struct btrfs_key key;
1238         int ret = 0;
1239
1240         while (1) {
1241                 ret = btrfs_next_item(root, path);
1242                 if (ret)
1243                         goto out;
1244
1245                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1246                 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
1247                         goto out;
1248         }
1249 out:
1250         return ret;
1251 }
1252
1253 /* This function will cleanup  */
1254 static int cleanup_temp_chunks(struct btrfs_fs_info *fs_info,
1255                                struct mkfs_allocation *alloc,
1256                                u64 data_profile, u64 meta_profile,
1257                                u64 sys_profile)
1258 {
1259         struct btrfs_trans_handle *trans = NULL;
1260         struct btrfs_block_group_item *bgi;
1261         struct btrfs_root *root = fs_info->extent_root;
1262         struct btrfs_key key;
1263         struct btrfs_key found_key;
1264         struct btrfs_path *path;
1265         int ret = 0;
1266
1267         path = btrfs_alloc_path();
1268         if (!path) {
1269                 ret = -ENOMEM;
1270                 goto out;
1271         }
1272
1273         trans = btrfs_start_transaction(root, 1);
1274
1275         key.objectid = 0;
1276         key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
1277         key.offset = 0;
1278
1279         while (1) {
1280                 /*
1281                  * as the rest of the loop may modify the tree, we need to
1282                  * start a new search each time.
1283                  */
1284                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
1285                 if (ret < 0)
1286                         goto out;
1287
1288                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1289                                       path->slots[0]);
1290                 if (found_key.objectid < key.objectid)
1291                         goto out;
1292                 if (found_key.type != BTRFS_BLOCK_GROUP_ITEM_KEY) {
1293                         ret = next_block_group(root, path);
1294                         if (ret < 0)
1295                                 goto out;
1296                         if (ret > 0) {
1297                                 ret = 0;
1298                                 goto out;
1299                         }
1300                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1301                                               path->slots[0]);
1302                 }
1303
1304                 bgi = btrfs_item_ptr(path->nodes[0], path->slots[0],
1305                                      struct btrfs_block_group_item);
1306                 if (is_temp_block_group(path->nodes[0], bgi,
1307                                         data_profile, meta_profile,
1308                                         sys_profile)) {
1309                         ret = btrfs_free_block_group(trans, fs_info,
1310                                         found_key.objectid, found_key.offset);
1311                         if (ret < 0)
1312                                 goto out;
1313                 }
1314                 btrfs_release_path(path);
1315                 key.objectid = found_key.objectid + found_key.offset;
1316         }
1317 out:
1318         if (trans)
1319                 btrfs_commit_transaction(trans, root);
1320         btrfs_free_path(path);
1321         return ret;
1322 }
1323
1324 int main(int ac, char **av)
1325 {
1326         char *file;
1327         struct btrfs_root *root;
1328         struct btrfs_trans_handle *trans;
1329         char *label = NULL;
1330         u64 block_count = 0;
1331         u64 dev_block_count = 0;
1332         u64 blocks[7];
1333         u64 alloc_start = 0;
1334         u64 metadata_profile = 0;
1335         u64 data_profile = 0;
1336         u32 nodesize = max_t(u32, sysconf(_SC_PAGESIZE),
1337                         BTRFS_MKFS_DEFAULT_NODE_SIZE);
1338         u32 sectorsize = 4096;
1339         u32 stripesize = 4096;
1340         int zero_end = 1;
1341         int fd;
1342         int ret;
1343         int i;
1344         int mixed = 0;
1345         int nodesize_forced = 0;
1346         int data_profile_opt = 0;
1347         int metadata_profile_opt = 0;
1348         int discard = 1;
1349         int ssd = 0;
1350         int force_overwrite = 0;
1351         char *source_dir = NULL;
1352         int source_dir_set = 0;
1353         u64 num_of_meta_chunks = 0;
1354         u64 size_of_data = 0;
1355         u64 source_dir_size = 0;
1356         int dev_cnt = 0;
1357         int saved_optind;
1358         char fs_uuid[BTRFS_UUID_UNPARSED_SIZE] = { 0 };
1359         u64 features = BTRFS_MKFS_DEFAULT_FEATURES;
1360         struct mkfs_allocation allocation = { 0 };
1361         struct btrfs_mkfs_config mkfs_cfg;
1362
1363         while(1) {
1364                 int c;
1365                 static const struct option long_options[] = {
1366                         { "alloc-start", required_argument, NULL, 'A'},
1367                         { "byte-count", required_argument, NULL, 'b' },
1368                         { "force", no_argument, NULL, 'f' },
1369                         { "leafsize", required_argument, NULL, 'l' },
1370                         { "label", required_argument, NULL, 'L'},
1371                         { "metadata", required_argument, NULL, 'm' },
1372                         { "mixed", no_argument, NULL, 'M' },
1373                         { "nodesize", required_argument, NULL, 'n' },
1374                         { "sectorsize", required_argument, NULL, 's' },
1375                         { "data", required_argument, NULL, 'd' },
1376                         { "version", no_argument, NULL, 'V' },
1377                         { "rootdir", required_argument, NULL, 'r' },
1378                         { "nodiscard", no_argument, NULL, 'K' },
1379                         { "features", required_argument, NULL, 'O' },
1380                         { "uuid", required_argument, NULL, 'U' },
1381                         { "quiet", 0, NULL, 'q' },
1382                         { "help", no_argument, NULL, GETOPT_VAL_HELP },
1383                         { NULL, 0, NULL, 0}
1384                 };
1385
1386                 c = getopt_long(ac, av, "A:b:fl:n:s:m:d:L:O:r:U:VMKq",
1387                                 long_options, NULL);
1388                 if (c < 0)
1389                         break;
1390                 switch(c) {
1391                         case 'A':
1392                                 alloc_start = parse_size(optarg);
1393                                 break;
1394                         case 'f':
1395                                 force_overwrite = 1;
1396                                 break;
1397                         case 'd':
1398                                 data_profile = parse_profile(optarg);
1399                                 data_profile_opt = 1;
1400                                 break;
1401                         case 'l':
1402                                 fprintf(stderr,
1403                         "WARNING: --leafsize is deprecated, use --nodesize\n");
1404                         case 'n':
1405                                 nodesize = parse_size(optarg);
1406                                 nodesize_forced = 1;
1407                                 break;
1408                         case 'L':
1409                                 label = parse_label(optarg);
1410                                 break;
1411                         case 'm':
1412                                 metadata_profile = parse_profile(optarg);
1413                                 metadata_profile_opt = 1;
1414                                 break;
1415                         case 'M':
1416                                 mixed = 1;
1417                                 break;
1418                         case 'O': {
1419                                 char *orig = strdup(optarg);
1420                                 char *tmp = orig;
1421
1422                                 tmp = btrfs_parse_fs_features(tmp, &features);
1423                                 if (tmp) {
1424                                         fprintf(stderr,
1425                                                 "Unrecognized filesystem feature '%s'\n",
1426                                                         tmp);
1427                                         free(orig);
1428                                         exit(1);
1429                                 }
1430                                 free(orig);
1431                                 if (features & BTRFS_FEATURE_LIST_ALL) {
1432                                         btrfs_list_all_fs_features(0);
1433                                         exit(0);
1434                                 }
1435                                 break;
1436                                 }
1437                         case 's':
1438                                 sectorsize = parse_size(optarg);
1439                                 break;
1440                         case 'b':
1441                                 block_count = parse_size(optarg);
1442                                 if (block_count <= BTRFS_MKFS_SMALL_VOLUME_SIZE)
1443                                         mixed = 1;
1444                                 zero_end = 0;
1445                                 break;
1446                         case 'V':
1447                                 print_version();
1448                                 break;
1449                         case 'r':
1450                                 source_dir = optarg;
1451                                 source_dir_set = 1;
1452                                 break;
1453                         case 'U':
1454                                 strncpy(fs_uuid, optarg,
1455                                         BTRFS_UUID_UNPARSED_SIZE - 1);
1456                                 break;
1457                         case 'K':
1458                                 discard = 0;
1459                                 break;
1460                         case 'q':
1461                                 verbose = 0;
1462                                 break;
1463                         case GETOPT_VAL_HELP:
1464                         default:
1465                                 print_usage(c != GETOPT_VAL_HELP);
1466                 }
1467         }
1468         sectorsize = max(sectorsize, (u32)sysconf(_SC_PAGESIZE));
1469         if (btrfs_check_nodesize(nodesize, sectorsize))
1470                 exit(1);
1471         saved_optind = optind;
1472         dev_cnt = ac - optind;
1473         if (dev_cnt == 0)
1474                 print_usage(1);
1475
1476         if (source_dir_set && dev_cnt > 1) {
1477                 fprintf(stderr,
1478                         "The -r option is limited to a single device\n");
1479                 exit(1);
1480         }
1481
1482         if (*fs_uuid) {
1483                 uuid_t dummy_uuid;
1484
1485                 if (uuid_parse(fs_uuid, dummy_uuid) != 0) {
1486                         fprintf(stderr, "could not parse UUID: %s\n", fs_uuid);
1487                         exit(1);
1488                 }
1489                 if (!test_uuid_unique(fs_uuid)) {
1490                         fprintf(stderr, "non-unique UUID: %s\n", fs_uuid);
1491                         exit(1);
1492                 }
1493         }
1494
1495         while (dev_cnt-- > 0) {
1496                 file = av[optind++];
1497                 if (is_block_device(file))
1498                         if (test_dev_for_mkfs(file, force_overwrite))
1499                                 exit(1);
1500         }
1501
1502         optind = saved_optind;
1503         dev_cnt = ac - optind;
1504
1505         file = av[optind++];
1506         ssd = is_ssd(file);
1507
1508         if (is_vol_small(file) || mixed) {
1509                 if (verbose)
1510                         printf("SMALL VOLUME: forcing mixed metadata/data groups\n");
1511                 mixed = 1;
1512         }
1513
1514         /*
1515         * Set default profiles according to number of added devices.
1516         * For mixed groups defaults are single/single.
1517         */
1518         if (!mixed) {
1519                 if (!metadata_profile_opt) {
1520                         if (dev_cnt == 1 && ssd && verbose)
1521                                 printf("Detected a SSD, turning off metadata "
1522                                 "duplication.  Mkfs with -m dup if you want to "
1523                                 "force metadata duplication.\n");
1524
1525                         metadata_profile = (dev_cnt > 1) ?
1526                                         BTRFS_BLOCK_GROUP_RAID1 : (ssd) ?
1527                                         0: BTRFS_BLOCK_GROUP_DUP;
1528                 }
1529                 if (!data_profile_opt) {
1530                         data_profile = (dev_cnt > 1) ?
1531                                 BTRFS_BLOCK_GROUP_RAID0 : 0; /* raid0 or single */
1532                 }
1533         } else {
1534                 u32 best_nodesize = max_t(u32, sysconf(_SC_PAGESIZE), sectorsize);
1535
1536                 if (metadata_profile_opt || data_profile_opt) {
1537                         if (metadata_profile != data_profile) {
1538                                 fprintf(stderr,
1539         "ERROR: With mixed block groups data and metadata profiles must be the same\n");
1540                                 exit(1);
1541                         }
1542                 }
1543
1544                 if (!nodesize_forced) {
1545                         nodesize = best_nodesize;
1546                         if (btrfs_check_nodesize(nodesize, sectorsize))
1547                                 exit(1);
1548                 }
1549                 if (nodesize != sectorsize) {
1550                         fprintf(stderr, "Error: mixed metadata/data block groups "
1551                                 "require metadata blocksizes equal to the sectorsize\n");
1552                         exit(1);
1553                 }
1554         }
1555
1556         /* Check device/block_count after the nodesize is determined */
1557         if (block_count && block_count < btrfs_min_dev_size(nodesize)) {
1558                 fprintf(stderr,
1559                         "Size '%llu' is too small to make a usable filesystem\n",
1560                         block_count);
1561                 fprintf(stderr,
1562                         "Minimum size for btrfs filesystem is %llu\n",
1563                         btrfs_min_dev_size(nodesize));
1564                 exit(1);
1565         }
1566         for (i = saved_optind; i < saved_optind + dev_cnt; i++) {
1567                 char *path;
1568
1569                 path = av[i];
1570                 ret = test_minimum_size(path, nodesize);
1571                 if (ret < 0) {
1572                         fprintf(stderr, "Failed to check size for '%s': %s\n",
1573                                 path, strerror(-ret));
1574                         exit (1);
1575                 }
1576                 if (ret > 0) {
1577                         fprintf(stderr,
1578                                 "'%s' is too small to make a usable filesystem\n",
1579                                 path);
1580                         fprintf(stderr,
1581                                 "Minimum size for each btrfs device is %llu.\n",
1582                                 btrfs_min_dev_size(nodesize));
1583                         exit(1);
1584                 }
1585         }
1586         ret = test_num_disk_vs_raid(metadata_profile, data_profile,
1587                         dev_cnt, mixed);
1588         if (ret)
1589                 exit(1);
1590
1591         /* if we are here that means all devs are good to btrfsify */
1592         if (verbose) {
1593                 printf("%s\n", PACKAGE_STRING);
1594                 printf("See %s for more information.\n\n", PACKAGE_URL);
1595         }
1596
1597         dev_cnt--;
1598
1599         if (!source_dir_set) {
1600                 /*
1601                  * open without O_EXCL so that the problem should not
1602                  * occur by the following processing.
1603                  * (btrfs_register_one_device() fails if O_EXCL is on)
1604                  */
1605                 fd = open(file, O_RDWR);
1606                 if (fd < 0) {
1607                         fprintf(stderr, "unable to open %s: %s\n", file,
1608                                 strerror(errno));
1609                         exit(1);
1610                 }
1611                 ret = btrfs_prepare_device(fd, file, zero_end, &dev_block_count,
1612                                            block_count, &mixed, discard);
1613                 if (ret) {
1614                         close(fd);
1615                         exit(1);
1616                 }
1617                 if (block_count && block_count > dev_block_count) {
1618                         fprintf(stderr, "%s is smaller than requested size\n", file);
1619                         exit(1);
1620                 }
1621         } else {
1622                 fd = open_target(file);
1623                 if (fd < 0) {
1624                         fprintf(stderr, "unable to open the %s\n", file);
1625                         exit(1);
1626                 }
1627
1628                 source_dir_size = size_sourcedir(source_dir, sectorsize,
1629                                              &num_of_meta_chunks, &size_of_data);
1630                 if(block_count < source_dir_size)
1631                         block_count = source_dir_size;
1632                 ret = zero_output_file(fd, block_count, sectorsize);
1633                 if (ret) {
1634                         fprintf(stderr, "unable to zero the output file\n");
1635                         exit(1);
1636                 }
1637                 /* our "device" is the new image file */
1638                 dev_block_count = block_count;
1639         }
1640
1641         /* To create the first block group and chunk 0 in make_btrfs */
1642         if (dev_block_count < BTRFS_MKFS_SYSTEM_GROUP_SIZE) {
1643                 fprintf(stderr, "device is too small to make filesystem\n");
1644                 exit(1);
1645         }
1646
1647         blocks[0] = BTRFS_SUPER_INFO_OFFSET;
1648         for (i = 1; i < 7; i++) {
1649                 blocks[i] = BTRFS_SUPER_INFO_OFFSET + 1024 * 1024 +
1650                         nodesize * i;
1651         }
1652
1653         if (group_profile_max_safe_loss(metadata_profile) <
1654                 group_profile_max_safe_loss(data_profile)){
1655                 fprintf(stderr,
1656                         "WARNING: metatdata has lower redundancy than data!\n\n");
1657         }
1658
1659         /*
1660          * FS features that can be set by other means than -O
1661          * just set the bit here
1662          */
1663         if (mixed)
1664                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS;
1665
1666         if ((data_profile | metadata_profile) &
1667             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
1668                 features |= BTRFS_FEATURE_INCOMPAT_RAID56;
1669         }
1670
1671         mkfs_cfg.label = label;
1672         mkfs_cfg.fs_uuid = fs_uuid;
1673         memcpy(mkfs_cfg.blocks, blocks, sizeof(blocks));
1674         mkfs_cfg.num_bytes = dev_block_count;
1675         mkfs_cfg.nodesize = nodesize;
1676         mkfs_cfg.sectorsize = sectorsize;
1677         mkfs_cfg.stripesize = stripesize;
1678         mkfs_cfg.features = features;
1679
1680         ret = make_btrfs(fd, &mkfs_cfg);
1681         if (ret) {
1682                 fprintf(stderr, "error during mkfs: %s\n", strerror(-ret));
1683                 exit(1);
1684         }
1685
1686         root = open_ctree(file, 0, OPEN_CTREE_WRITES);
1687         if (!root) {
1688                 fprintf(stderr, "Open ctree failed\n");
1689                 close(fd);
1690                 exit(1);
1691         }
1692         root->fs_info->alloc_start = alloc_start;
1693
1694         ret = create_metadata_block_groups(root, mixed, &allocation);
1695         if (ret) {
1696                 fprintf(stderr, "failed to create default block groups\n");
1697                 exit(1);
1698         }
1699
1700         trans = btrfs_start_transaction(root, 1);
1701         if (!trans) {
1702                 fprintf(stderr, "failed to start transaction\n");
1703                 exit(1);
1704         }
1705
1706         ret = create_data_block_groups(trans, root, mixed, &allocation);
1707         if (ret) {
1708                 fprintf(stderr, "failed to create default data block groups\n");
1709                 exit(1);
1710         }
1711
1712         ret = make_root_dir(trans, root, mixed, &allocation);
1713         if (ret) {
1714                 fprintf(stderr, "failed to setup the root directory\n");
1715                 exit(1);
1716         }
1717
1718         btrfs_commit_transaction(trans, root);
1719
1720         trans = btrfs_start_transaction(root, 1);
1721         if (!trans) {
1722                 fprintf(stderr, "failed to start transaction\n");
1723                 exit(1);
1724         }
1725
1726         if (is_block_device(file))
1727                 btrfs_register_one_device(file);
1728
1729         if (dev_cnt == 0)
1730                 goto raid_groups;
1731
1732         while (dev_cnt-- > 0) {
1733                 int old_mixed = mixed;
1734
1735                 file = av[optind++];
1736
1737                 /*
1738                  * open without O_EXCL so that the problem should not
1739                  * occur by the following processing.
1740                  * (btrfs_register_one_device() fails if O_EXCL is on)
1741                  */
1742                 fd = open(file, O_RDWR);
1743                 if (fd < 0) {
1744                         fprintf(stderr, "unable to open %s: %s\n", file,
1745                                 strerror(errno));
1746                         exit(1);
1747                 }
1748                 ret = btrfs_device_already_in_root(root, fd,
1749                                                    BTRFS_SUPER_INFO_OFFSET);
1750                 if (ret) {
1751                         fprintf(stderr, "skipping duplicate device %s in FS\n",
1752                                 file);
1753                         close(fd);
1754                         continue;
1755                 }
1756                 ret = btrfs_prepare_device(fd, file, zero_end, &dev_block_count,
1757                                            block_count, &mixed, discard);
1758                 if (ret) {
1759                         close(fd);
1760                         exit(1);
1761                 }
1762                 mixed = old_mixed;
1763
1764                 ret = btrfs_add_to_fsid(trans, root, fd, file, dev_block_count,
1765                                         sectorsize, sectorsize, sectorsize);
1766                 BUG_ON(ret);
1767                 if (verbose >= 2) {
1768                         struct btrfs_device *device;
1769
1770                         device = container_of(root->fs_info->fs_devices->devices.next,
1771                                         struct btrfs_device, dev_list);
1772                         printf("adding device %s id %llu\n", file,
1773                                 (unsigned long long)device->devid);
1774                 }
1775
1776                 if (is_block_device(file))
1777                         btrfs_register_one_device(file);
1778         }
1779
1780 raid_groups:
1781         if (!source_dir_set) {
1782                 ret = create_raid_groups(trans, root, data_profile,
1783                                  metadata_profile, mixed, &allocation);
1784                 BUG_ON(ret);
1785         }
1786
1787         ret = create_data_reloc_tree(trans, root);
1788         BUG_ON(ret);
1789
1790         btrfs_commit_transaction(trans, root);
1791
1792         if (source_dir_set) {
1793                 trans = btrfs_start_transaction(root, 1);
1794                 ret = create_chunks(trans, root,
1795                                     num_of_meta_chunks, size_of_data,
1796                                     &allocation);
1797                 BUG_ON(ret);
1798                 btrfs_commit_transaction(trans, root);
1799
1800                 ret = make_image(source_dir, root, fd);
1801                 BUG_ON(ret);
1802         }
1803         ret = cleanup_temp_chunks(root->fs_info, &allocation, data_profile,
1804                                   metadata_profile, metadata_profile);
1805         if (ret < 0) {
1806                 fprintf(stderr, "Failed to cleanup temporary chunks\n");
1807                 goto out;
1808         }
1809
1810         if (verbose) {
1811                 char features_buf[64];
1812
1813                 printf("Label:              %s\n", label);
1814                 printf("UUID:               %s\n", fs_uuid);
1815                 printf("Node size:          %u\n", nodesize);
1816                 printf("Sector size:        %u\n", sectorsize);
1817                 printf("Filesystem size:    %s\n",
1818                         pretty_size(btrfs_super_total_bytes(root->fs_info->super_copy)));
1819                 printf("Block group profiles:\n");
1820                 if (allocation.data)
1821                         printf("  Data:             %-8s %16s\n",
1822                                 btrfs_group_profile_str(data_profile),
1823                                 pretty_size(allocation.data));
1824                 if (allocation.metadata)
1825                         printf("  Metadata:         %-8s %16s\n",
1826                                 btrfs_group_profile_str(metadata_profile),
1827                                 pretty_size(allocation.metadata));
1828                 if (allocation.mixed)
1829                         printf("  Data+Metadata:    %-8s %16s\n",
1830                                 btrfs_group_profile_str(data_profile),
1831                                 pretty_size(allocation.mixed));
1832                 printf("  System:           %-8s %16s\n",
1833                         btrfs_group_profile_str(metadata_profile),
1834                         pretty_size(allocation.system));
1835                 printf("SSD detected:       %s\n", ssd ? "yes" : "no");
1836                 btrfs_parse_features_to_string(features_buf, features);
1837                 printf("Incompat features:  %s", features_buf);
1838                 printf("\n");
1839
1840                 list_all_devices(root);
1841         }
1842
1843 out:
1844         ret = close_ctree(root);
1845         BUG_ON(ret);
1846         btrfs_close_all_devices();
1847         free(label);
1848         return 0;
1849 }