mkfs/main.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include "kerncompat.h"
  20 #include "androidcompat.h"
  21
  22 #include <sys/ioctl.h>
  23 #include <sys/mount.h>
  24 #include "ioctl.h"
  25 #include <stdio.h>
  26 #include <stdlib.h>
  27 /* #include <sys/dir.h> included via androidcompat.h */
  28 #include <fcntl.h>
  29 #include <unistd.h>
  30 #include <getopt.h>
  31 #include <uuid/uuid.h>
  32 #include <ctype.h>
  33 #include <blkid/blkid.h>
  34 #include "ctree.h"
  35 #include "disk-io.h"
  36 #include "volumes.h"
  37 #include "transaction.h"
  38 #include "utils.h"
  39 #include "list_sort.h"
  40 #include "help.h"
  41 #include "mkfs/common.h"
  42 #include "mkfs/rootdir.h"
  43 #include "fsfeatures.h"
  44
  45 static int verbose = 1;
  46
  47 struct mkfs_allocation {
  48         u64 data;
  49         u64 metadata;
  50         u64 mixed;
  51         u64 system;
  52 };
  53
  54 static int create_metadata_block_groups(struct btrfs_root *root, int mixed,
  55                                 struct mkfs_allocation *allocation)
  56 {
  57         struct btrfs_fs_info *fs_info = root->fs_info;
  58         struct btrfs_trans_handle *trans;
  59         u64 bytes_used;
  60         u64 chunk_start = 0;
  61         u64 chunk_size = 0;
  62         int ret;
  63
  64         trans = btrfs_start_transaction(root, 1);
  65         BUG_ON(IS_ERR(trans));
  66         bytes_used = btrfs_super_bytes_used(fs_info->super_copy);
  67
  68         root->fs_info->system_allocs = 1;
  69         /*
  70          * First temporary system chunk must match the chunk layout
  71          * created in make_btrfs().
  72          */
  73         ret = btrfs_make_block_group(trans, fs_info, bytes_used,
  74                                      BTRFS_BLOCK_GROUP_SYSTEM,
  75                                      BTRFS_BLOCK_RESERVED_1M_FOR_SUPER,
  76                                      BTRFS_MKFS_SYSTEM_GROUP_SIZE);
  77         allocation->system += BTRFS_MKFS_SYSTEM_GROUP_SIZE;
  78         if (ret)
  79                 return ret;
  80
  81         if (mixed) {
  82                 ret = btrfs_alloc_chunk(trans, fs_info,
  83                                         &chunk_start, &chunk_size,
  84                                         BTRFS_BLOCK_GROUP_METADATA |
  85                                         BTRFS_BLOCK_GROUP_DATA);
  86                 if (ret == -ENOSPC) {
  87                         error("no space to allocate data/metadata chunk");
  88                         goto err;
  89                 }
  90                 if (ret)
  91                         return ret;
  92                 ret = btrfs_make_block_group(trans, fs_info, 0,
  93                                              BTRFS_BLOCK_GROUP_METADATA |
  94                                              BTRFS_BLOCK_GROUP_DATA,
  95                                              chunk_start, chunk_size);
  96                 if (ret)
  97                         return ret;
  98                 allocation->mixed += chunk_size;
  99         } else {
 100                 ret = btrfs_alloc_chunk(trans, fs_info,
 101                                         &chunk_start, &chunk_size,
 102                                         BTRFS_BLOCK_GROUP_METADATA);
 103                 if (ret == -ENOSPC) {
 104                         error("no space to allocate metadata chunk");
 105                         goto err;
 106                 }
 107                 if (ret)
 108                         return ret;
 109                 ret = btrfs_make_block_group(trans, fs_info, 0,
 110                                              BTRFS_BLOCK_GROUP_METADATA,
 111                                              chunk_start, chunk_size);
 112                 allocation->metadata += chunk_size;
 113                 if (ret)
 114                         return ret;
 115         }
 116
 117         root->fs_info->system_allocs = 0;
 118         ret = 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         struct btrfs_fs_info *fs_info = root->fs_info;
 129         u64 chunk_start = 0;
 130         u64 chunk_size = 0;
 131         int ret = 0;
 132
 133         if (!mixed) {
 134                 ret = btrfs_alloc_chunk(trans, fs_info,
 135                                         &chunk_start, &chunk_size,
 136                                         BTRFS_BLOCK_GROUP_DATA);
 137                 if (ret == -ENOSPC) {
 138                         error("no space to allocate data chunk");
 139                         goto err;
 140                 }
 141                 if (ret)
 142                         return ret;
 143                 ret = btrfs_make_block_group(trans, fs_info, 0,
 144                                              BTRFS_BLOCK_GROUP_DATA,
 145                                              chunk_start, chunk_size);
 146                 allocation->data += chunk_size;
 147                 if (ret)
 148                         return ret;
 149         }
 150
 151 err:
 152         return ret;
 153 }
 154
 155 static int make_root_dir(struct btrfs_trans_handle *trans,
 156                 struct btrfs_root *root)
 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 int __recow_root(struct btrfs_trans_handle *trans,
 188                          struct btrfs_root *root)
 189 {
 190         struct extent_buffer *tmp;
 191         int ret;
 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                 if (ret)
 198                         return ret;
 199                 free_extent_buffer(tmp);
 200         }
 201
 202         return 0;
 203 }
 204
 205 static int recow_roots(struct btrfs_trans_handle *trans,
 206                        struct btrfs_root *root)
 207 {
 208         struct btrfs_fs_info *info = root->fs_info;
 209         int ret;
 210
 211         ret = __recow_root(trans, info->fs_root);
 212         if (ret)
 213                 return ret;
 214         ret = __recow_root(trans, info->tree_root);
 215         if (ret)
 216                 return ret;
 217         ret = __recow_root(trans, info->extent_root);
 218         if (ret)
 219                 return ret;
 220         ret = __recow_root(trans, info->chunk_root);
 221         if (ret)
 222                 return ret;
 223         ret = __recow_root(trans, info->dev_root);
 224         if (ret)
 225                 return ret;
 226         ret = __recow_root(trans, info->csum_root);
 227         if (ret)
 228                 return ret;
 229
 230         return 0;
 231 }
 232
 233 static int create_one_raid_group(struct btrfs_trans_handle *trans,
 234                               struct btrfs_root *root, u64 type,
 235                               struct mkfs_allocation *allocation)
 236
 237 {
 238         struct btrfs_fs_info *fs_info = root->fs_info;
 239         u64 chunk_start;
 240         u64 chunk_size;
 241         int ret;
 242
 243         ret = btrfs_alloc_chunk(trans, fs_info,
 244                                 &chunk_start, &chunk_size, type);
 245         if (ret == -ENOSPC) {
 246                 error("not enough free space to allocate chunk");
 247                 exit(1);
 248         }
 249         if (ret)
 250                 return ret;
 251
 252         ret = btrfs_make_block_group(trans, fs_info, 0,
 253                                      type, chunk_start, chunk_size);
 254
 255         type &= BTRFS_BLOCK_GROUP_TYPE_MASK;
 256         if (type == BTRFS_BLOCK_GROUP_DATA) {
 257                 allocation->data += chunk_size;
 258         } else if (type == BTRFS_BLOCK_GROUP_METADATA) {
 259                 allocation->metadata += chunk_size;
 260         } else if (type == BTRFS_BLOCK_GROUP_SYSTEM) {
 261                 allocation->system += chunk_size;
 262         } else if (type ==
 263                         (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA)) {
 264                 allocation->mixed += chunk_size;
 265         } else {
 266                 error("unrecognized profile type: 0x%llx",
 267                                 (unsigned long long)type);
 268                 ret = -EINVAL;
 269         }
 270
 271         return ret;
 272 }
 273
 274 static int create_raid_groups(struct btrfs_trans_handle *trans,
 275                               struct btrfs_root *root, u64 data_profile,
 276                               u64 metadata_profile, int mixed,
 277                               struct mkfs_allocation *allocation)
 278 {
 279         int ret;
 280
 281         if (metadata_profile) {
 282                 u64 meta_flags = BTRFS_BLOCK_GROUP_METADATA;
 283
 284                 ret = create_one_raid_group(trans, root,
 285                                             BTRFS_BLOCK_GROUP_SYSTEM |
 286                                             metadata_profile, allocation);
 287                 if (ret)
 288                         return ret;
 289
 290                 if (mixed)
 291                         meta_flags |= BTRFS_BLOCK_GROUP_DATA;
 292
 293                 ret = create_one_raid_group(trans, root, meta_flags |
 294                                             metadata_profile, allocation);
 295                 if (ret)
 296                         return ret;
 297
 298         }
 299         if (!mixed && data_profile) {
 300                 ret = create_one_raid_group(trans, root,
 301                                             BTRFS_BLOCK_GROUP_DATA |
 302                                             data_profile, allocation);
 303                 if (ret)
 304                         return ret;
 305         }
 306         ret = recow_roots(trans, root);
 307
 308         return ret;
 309 }
 310
 311 static int create_tree(struct btrfs_trans_handle *trans,
 312                         struct btrfs_root *root, u64 objectid)
 313 {
 314         struct btrfs_key location;
 315         struct btrfs_root_item root_item;
 316         struct extent_buffer *tmp;
 317         int ret;
 318
 319         ret = btrfs_copy_root(trans, root, root->node, &tmp, objectid);
 320         if (ret)
 321                 return ret;
 322
 323         memcpy(&root_item, &root->root_item, sizeof(root_item));
 324         btrfs_set_root_bytenr(&root_item, tmp->start);
 325         btrfs_set_root_level(&root_item, btrfs_header_level(tmp));
 326         btrfs_set_root_generation(&root_item, trans->transid);
 327         free_extent_buffer(tmp);
 328
 329         location.objectid = objectid;
 330         location.type = BTRFS_ROOT_ITEM_KEY;
 331         location.offset = 0;
 332         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
 333                                 &location, &root_item);
 334
 335         return ret;
 336 }
 337
 338 static void print_usage(int ret)
 339 {
 340         printf("Usage: mkfs.btrfs [options] dev [ dev ... ]\n");
 341         printf("Options:\n");
 342         printf("  allocation profiles:\n");
 343         printf("\t-d|--data PROFILE       data profile, raid0, raid1, raid5, raid6, raid10, dup or single\n");
 344         printf("\t-m|--metadata PROFILE   metadata profile, values like for data profile\n");
 345         printf("\t-M|--mixed              mix metadata and data together\n");
 346         printf("  features:\n");
 347         printf("\t-n|--nodesize SIZE      size of btree nodes\n");
 348         printf("\t-s|--sectorsize SIZE    data block size (may not be mountable by current kernel)\n");
 349         printf("\t-O|--features LIST      comma separated list of filesystem features (use '-O list-all' to list features)\n");
 350         printf("\t-L|--label LABEL        set the filesystem label\n");
 351         printf("\t-U|--uuid UUID          specify the filesystem UUID (must be unique)\n");
 352         printf("  creation:\n");
 353         printf("\t-b|--byte-count SIZE    set filesystem size to SIZE (on the first device)\n");
 354         printf("\t-r|--rootdir DIR        copy files from DIR to the image root directory\n");
 355         printf("\t--shrink                (with --rootdir) shrink the filled filesystem to minimal size\n");
 356         printf("\t-K|--nodiscard          do not perform whole device TRIM\n");
 357         printf("\t-f|--force              force overwrite of existing filesystem\n");
 358         printf("  general:\n");
 359         printf("\t-q|--quiet              no messages except errors\n");
 360         printf("\t-V|--version            print the mkfs.btrfs version and exit\n");
 361         printf("\t--help                  print this help and exit\n");
 362         printf("  deprecated:\n");
 363         printf("\t-A|--alloc-start START  the offset to start the filesystem\n");
 364         printf("\t-l|--leafsize SIZE      deprecated, alias for nodesize\n");
 365         exit(ret);
 366 }
 367
 368 static u64 parse_profile(const char *s)
 369 {
 370         if (strcasecmp(s, "raid0") == 0) {
 371                 return BTRFS_BLOCK_GROUP_RAID0;
 372         } else if (strcasecmp(s, "raid1") == 0) {
 373                 return BTRFS_BLOCK_GROUP_RAID1;
 374         } else if (strcasecmp(s, "raid5") == 0) {
 375                 return BTRFS_BLOCK_GROUP_RAID5;
 376         } else if (strcasecmp(s, "raid6") == 0) {
 377                 return BTRFS_BLOCK_GROUP_RAID6;
 378         } else if (strcasecmp(s, "raid10") == 0) {
 379                 return BTRFS_BLOCK_GROUP_RAID10;
 380         } else if (strcasecmp(s, "dup") == 0) {
 381                 return BTRFS_BLOCK_GROUP_DUP;
 382         } else if (strcasecmp(s, "single") == 0) {
 383                 return 0;
 384         } else {
 385                 error("unknown profile %s", s);
 386                 exit(1);
 387         }
 388         /* not reached */
 389         return 0;
 390 }
 391
 392 static char *parse_label(const char *input)
 393 {
 394         int len = strlen(input);
 395
 396         if (len >= BTRFS_LABEL_SIZE) {
 397                 error("label %s is too long (max %d)", input,
 398                         BTRFS_LABEL_SIZE - 1);
 399                 exit(1);
 400         }
 401         return strdup(input);
 402 }
 403
 404 static int zero_output_file(int out_fd, u64 size)
 405 {
 406         int loop_num;
 407         u64 location = 0;
 408         char buf[SZ_4K];
 409         int ret = 0, i;
 410         ssize_t written;
 411
 412         memset(buf, 0, SZ_4K);
 413
 414         /* Only zero out the first 1M */
 415         loop_num = SZ_1M / SZ_4K;
 416         for (i = 0; i < loop_num; i++) {
 417                 written = pwrite64(out_fd, buf, SZ_4K, location);
 418                 if (written != SZ_4K)
 419                         ret = -EIO;
 420                 location += SZ_4K;
 421         }
 422
 423         /* Then enlarge the file to size */
 424         written = pwrite64(out_fd, buf, 1, size - 1);
 425         if (written < 1)
 426                 ret = -EIO;
 427         return ret;
 428 }
 429
 430 static int is_ssd(const char *file)
 431 {
 432         blkid_probe probe;
 433         char wholedisk[PATH_MAX];
 434         char sysfs_path[PATH_MAX];
 435         dev_t devno;
 436         int fd;
 437         char rotational;
 438         int ret;
 439
 440         probe = blkid_new_probe_from_filename(file);
 441         if (!probe)
 442                 return 0;
 443
 444         /* Device number of this disk (possibly a partition) */
 445         devno = blkid_probe_get_devno(probe);
 446         if (!devno) {
 447                 blkid_free_probe(probe);
 448                 return 0;
 449         }
 450
 451         /* Get whole disk name (not full path) for this devno */
 452         ret = blkid_devno_to_wholedisk(devno,
 453                         wholedisk, sizeof(wholedisk), NULL);
 454         if (ret) {
 455                 blkid_free_probe(probe);
 456                 return 0;
 457         }
 458
 459         snprintf(sysfs_path, PATH_MAX, "/sys/block/%s/queue/rotational",
 460                  wholedisk);
 461
 462         blkid_free_probe(probe);
 463
 464         fd = open(sysfs_path, O_RDONLY);
 465         if (fd < 0) {
 466                 return 0;
 467         }
 468
 469         if (read(fd, &rotational, 1) < 1) {
 470                 close(fd);
 471                 return 0;
 472         }
 473         close(fd);
 474
 475         return rotational == '0';
 476 }
 477
 478 static int _cmp_device_by_id(void *priv, struct list_head *a,
 479                              struct list_head *b)
 480 {
 481         return list_entry(a, struct btrfs_device, dev_list)->devid -
 482                list_entry(b, struct btrfs_device, dev_list)->devid;
 483 }
 484
 485 static void list_all_devices(struct btrfs_root *root)
 486 {
 487         struct btrfs_fs_devices *fs_devices;
 488         struct btrfs_device *device;
 489         int number_of_devices = 0;
 490         u64 total_block_count = 0;
 491
 492         fs_devices = root->fs_info->fs_devices;
 493
 494         list_for_each_entry(device, &fs_devices->devices, dev_list)
 495                 number_of_devices++;
 496
 497         list_sort(NULL, &fs_devices->devices, _cmp_device_by_id);
 498
 499         printf("Number of devices:  %d\n", number_of_devices);
 500         /* printf("Total devices size: %10s\n", */
 501                 /* pretty_size(total_block_count)); */
 502         printf("Devices:\n");
 503         printf("   ID        SIZE  PATH\n");
 504         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 505                 printf("  %3llu  %10s  %s\n",
 506                         device->devid,
 507                         pretty_size(device->total_bytes),
 508                         device->name);
 509                 total_block_count += device->total_bytes;
 510         }
 511
 512         printf("\n");
 513 }
 514
 515 static int is_temp_block_group(struct extent_buffer *node,
 516                                struct btrfs_block_group_item *bgi,
 517                                u64 data_profile, u64 meta_profile,
 518                                u64 sys_profile)
 519 {
 520         u64 flag = btrfs_disk_block_group_flags(node, bgi);
 521         u64 flag_type = flag & BTRFS_BLOCK_GROUP_TYPE_MASK;
 522         u64 flag_profile = flag & BTRFS_BLOCK_GROUP_PROFILE_MASK;
 523         u64 used = btrfs_disk_block_group_used(node, bgi);
 524
 525         /*
 526          * Chunks meets all the following conditions is a temp chunk
 527          * 1) Empty chunk
 528          * Temp chunk is always empty.
 529          *
 530          * 2) profile mismatch with mkfs profile.
 531          * Temp chunk is always in SINGLE
 532          *
 533          * 3) Size differs with mkfs_alloc
 534          * Special case for SINGLE/SINGLE btrfs.
 535          * In that case, temp data chunk and real data chunk are always empty.
 536          * So we need to use mkfs_alloc to be sure which chunk is the newly
 537          * allocated.
 538          *
 539          * Normally, new chunk size is equal to mkfs one (One chunk)
 540          * If it has multiple chunks, we just refuse to delete any one.
 541          * As they are all single, so no real problem will happen.
 542          * So only use condition 1) and 2) to judge them.
 543          */
 544         if (used != 0)
 545                 return 0;
 546         switch (flag_type) {
 547         case BTRFS_BLOCK_GROUP_DATA:
 548         case BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA:
 549                 data_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
 550                 if (flag_profile != data_profile)
 551                         return 1;
 552                 break;
 553         case BTRFS_BLOCK_GROUP_METADATA:
 554                 meta_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
 555                 if (flag_profile != meta_profile)
 556                         return 1;
 557                 break;
 558         case BTRFS_BLOCK_GROUP_SYSTEM:
 559                 sys_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
 560                 if (flag_profile != sys_profile)
 561                         return 1;
 562                 break;
 563         }
 564         return 0;
 565 }
 566
 567 /* Note: if current is a block group, it will skip it anyway */
 568 static int next_block_group(struct btrfs_root *root,
 569                             struct btrfs_path *path)
 570 {
 571         struct btrfs_key key;
 572         int ret = 0;
 573
 574         while (1) {
 575                 ret = btrfs_next_item(root, path);
 576                 if (ret)
 577                         goto out;
 578
 579                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 580                 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
 581                         goto out;
 582         }
 583 out:
 584         return ret;
 585 }
 586
 587 /* This function will cleanup  */
 588 static int cleanup_temp_chunks(struct btrfs_fs_info *fs_info,
 589                                struct mkfs_allocation *alloc,
 590                                u64 data_profile, u64 meta_profile,
 591                                u64 sys_profile)
 592 {
 593         struct btrfs_trans_handle *trans = NULL;
 594         struct btrfs_block_group_item *bgi;
 595         struct btrfs_root *root = fs_info->extent_root;
 596         struct btrfs_key key;
 597         struct btrfs_key found_key;
 598         struct btrfs_path path;
 599         int ret = 0;
 600
 601         btrfs_init_path(&path);
 602         trans = btrfs_start_transaction(root, 1);
 603         BUG_ON(IS_ERR(trans));
 604
 605         key.objectid = 0;
 606         key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
 607         key.offset = 0;
 608
 609         while (1) {
 610                 /*
 611                  * as the rest of the loop may modify the tree, we need to
 612                  * start a new search each time.
 613                  */
 614                 ret = btrfs_search_slot(trans, root, &key, &path, 0, 0);
 615                 if (ret < 0)
 616                         goto out;
 617                 /* Don't pollute ret for >0 case */
 618                 if (ret > 0)
 619                         ret = 0;
 620
 621                 btrfs_item_key_to_cpu(path.nodes[0], &found_key,
 622                                       path.slots[0]);
 623                 if (found_key.objectid < key.objectid)
 624                         goto out;
 625                 if (found_key.type != BTRFS_BLOCK_GROUP_ITEM_KEY) {
 626                         ret = next_block_group(root, &path);
 627                         if (ret < 0)
 628                                 goto out;
 629                         if (ret > 0) {
 630                                 ret = 0;
 631                                 goto out;
 632                         }
 633                         btrfs_item_key_to_cpu(path.nodes[0], &found_key,
 634                                               path.slots[0]);
 635                 }
 636
 637                 bgi = btrfs_item_ptr(path.nodes[0], path.slots[0],
 638                                      struct btrfs_block_group_item);
 639                 if (is_temp_block_group(path.nodes[0], bgi,
 640                                         data_profile, meta_profile,
 641                                         sys_profile)) {
 642                         u64 flags = btrfs_disk_block_group_flags(path.nodes[0],
 643                                                              bgi);
 644
 645                         ret = btrfs_free_block_group(trans, fs_info,
 646                                         found_key.objectid, found_key.offset);
 647                         if (ret < 0)
 648                                 goto out;
 649
 650                         if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
 651                             BTRFS_BLOCK_GROUP_DATA)
 652                                 alloc->data -= found_key.offset;
 653                         else if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
 654                                  BTRFS_BLOCK_GROUP_METADATA)
 655                                 alloc->metadata -= found_key.offset;
 656                         else if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
 657                                  BTRFS_BLOCK_GROUP_SYSTEM)
 658                                 alloc->system -= found_key.offset;
 659                         else if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
 660                                  (BTRFS_BLOCK_GROUP_METADATA |
 661                                   BTRFS_BLOCK_GROUP_DATA))
 662                                 alloc->mixed -= found_key.offset;
 663                 }
 664                 btrfs_release_path(&path);
 665                 key.objectid = found_key.objectid + found_key.offset;
 666         }
 667 out:
 668         if (trans)
 669                 btrfs_commit_transaction(trans, root);
 670         btrfs_release_path(&path);
 671         return ret;
 672 }
 673
 674 /*
 675  * Just update chunk allocation info, since --rootdir may allocate new
 676  * chunks which is not updated in @allocation structure.
 677  */
 678 static void update_chunk_allocation(struct btrfs_fs_info *fs_info,
 679                                     struct mkfs_allocation *allocation)
 680 {
 681         struct btrfs_block_group_cache *bg_cache;
 682         const u64 mixed_flag = BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA;
 683         u64 search_start = 0;
 684
 685         allocation->mixed = 0;
 686         allocation->data = 0;
 687         allocation->metadata = 0;
 688         allocation->system = 0;
 689         while (1) {
 690                 bg_cache = btrfs_lookup_first_block_group(fs_info,
 691                                                           search_start);
 692                 if (!bg_cache)
 693                         break;
 694                 if ((bg_cache->flags & mixed_flag) == mixed_flag)
 695                         allocation->mixed += bg_cache->key.offset;
 696                 else if (bg_cache->flags & BTRFS_BLOCK_GROUP_DATA)
 697                         allocation->data += bg_cache->key.offset;
 698                 else if (bg_cache->flags & BTRFS_BLOCK_GROUP_METADATA)
 699                         allocation->metadata += bg_cache->key.offset;
 700                 else
 701                         allocation->system += bg_cache->key.offset;
 702                 search_start = bg_cache->key.objectid + bg_cache->key.offset;
 703         }
 704 }
 705
 706 int main(int argc, char **argv)
 707 {
 708         char *file;
 709         struct btrfs_root *root;
 710         struct btrfs_fs_info *fs_info;
 711         struct btrfs_trans_handle *trans;
 712         char *label = NULL;
 713         u64 block_count = 0;
 714         u64 dev_block_count = 0;
 715         u64 alloc_start = 0;
 716         u64 metadata_profile = 0;
 717         u64 data_profile = 0;
 718         u32 nodesize = max_t(u32, sysconf(_SC_PAGESIZE),
 719                         BTRFS_MKFS_DEFAULT_NODE_SIZE);
 720         u32 sectorsize = 4096;
 721         u32 stripesize = 4096;
 722         int zero_end = 1;
 723         int fd = -1;
 724         int ret = 0;
 725         int close_ret;
 726         int i;
 727         int mixed = 0;
 728         int nodesize_forced = 0;
 729         int data_profile_opt = 0;
 730         int metadata_profile_opt = 0;
 731         int discard = 1;
 732         int ssd = 0;
 733         int force_overwrite = 0;
 734         char *source_dir = NULL;
 735         bool source_dir_set = false;
 736         bool shrink_rootdir = false;
 737         u64 source_dir_size = 0;
 738         u64 min_dev_size;
 739         u64 shrink_size;
 740         int dev_cnt = 0;
 741         int saved_optind;
 742         char fs_uuid[BTRFS_UUID_UNPARSED_SIZE] = { 0 };
 743         u64 features = BTRFS_MKFS_DEFAULT_FEATURES;
 744         struct mkfs_allocation allocation = { 0 };
 745         struct btrfs_mkfs_config mkfs_cfg;
 746
 747         while(1) {
 748                 int c;
 749                 enum { GETOPT_VAL_SHRINK = 257 };
 750                 static const struct option long_options[] = {
 751                         { "alloc-start", required_argument, NULL, 'A'},
 752                         { "byte-count", required_argument, NULL, 'b' },
 753                         { "force", no_argument, NULL, 'f' },
 754                         { "leafsize", required_argument, NULL, 'l' },
 755                         { "label", required_argument, NULL, 'L'},
 756                         { "metadata", required_argument, NULL, 'm' },
 757                         { "mixed", no_argument, NULL, 'M' },
 758                         { "nodesize", required_argument, NULL, 'n' },
 759                         { "sectorsize", required_argument, NULL, 's' },
 760                         { "data", required_argument, NULL, 'd' },
 761                         { "version", no_argument, NULL, 'V' },
 762                         { "rootdir", required_argument, NULL, 'r' },
 763                         { "nodiscard", no_argument, NULL, 'K' },
 764                         { "features", required_argument, NULL, 'O' },
 765                         { "uuid", required_argument, NULL, 'U' },
 766                         { "quiet", 0, NULL, 'q' },
 767                         { "shrink", no_argument, NULL, GETOPT_VAL_SHRINK },
 768                         { "help", no_argument, NULL, GETOPT_VAL_HELP },
 769                         { NULL, 0, NULL, 0}
 770                 };
 771
 772                 c = getopt_long(argc, argv, "A:b:fl:n:s:m:d:L:O:r:U:VMKq",
 773                                 long_options, NULL);
 774                 if (c < 0)
 775                         break;
 776                 switch(c) {
 777                         case 'A':
 778                                 alloc_start = parse_size(optarg);
 779                                 break;
 780                         case 'f':
 781                                 force_overwrite = 1;
 782                                 break;
 783                         case 'd':
 784                                 data_profile = parse_profile(optarg);
 785                                 data_profile_opt = 1;
 786                                 break;
 787                         case 'l':
 788                                 warning("--leafsize is deprecated, use --nodesize");
 789                                 /* fall through */
 790                         case 'n':
 791                                 nodesize = parse_size(optarg);
 792                                 nodesize_forced = 1;
 793                                 break;
 794                         case 'L':
 795                                 label = parse_label(optarg);
 796                                 break;
 797                         case 'm':
 798                                 metadata_profile = parse_profile(optarg);
 799                                 metadata_profile_opt = 1;
 800                                 break;
 801                         case 'M':
 802                                 mixed = 1;
 803                                 break;
 804                         case 'O': {
 805                                 char *orig = strdup(optarg);
 806                                 char *tmp = orig;
 807
 808                                 tmp = btrfs_parse_fs_features(tmp, &features);
 809                                 if (tmp) {
 810                                         error("unrecognized filesystem feature '%s'",
 811                                                         tmp);
 812                                         free(orig);
 813                                         goto error;
 814                                 }
 815                                 free(orig);
 816                                 if (features & BTRFS_FEATURE_LIST_ALL) {
 817                                         btrfs_list_all_fs_features(0);
 818                                         goto success;
 819                                 }
 820                                 break;
 821                                 }
 822                         case 's':
 823                                 sectorsize = parse_size(optarg);
 824                                 break;
 825                         case 'b':
 826                                 block_count = parse_size(optarg);
 827                                 zero_end = 0;
 828                                 break;
 829                         case 'V':
 830                                 printf("mkfs.btrfs, part of %s\n",
 831                                                 PACKAGE_STRING);
 832                                 goto success;
 833                         case 'r':
 834                                 source_dir = optarg;
 835                                 source_dir_set = true;
 836                                 break;
 837                         case 'U':
 838                                 strncpy(fs_uuid, optarg,
 839                                         BTRFS_UUID_UNPARSED_SIZE - 1);
 840                                 break;
 841                         case 'K':
 842                                 discard = 0;
 843                                 break;
 844                         case 'q':
 845                                 verbose = 0;
 846                                 break;
 847                         case GETOPT_VAL_SHRINK:
 848                                 shrink_rootdir = true;
 849                                 break;
 850                         case GETOPT_VAL_HELP:
 851                         default:
 852                                 print_usage(c != GETOPT_VAL_HELP);
 853                 }
 854         }
 855
 856         if (verbose) {
 857                 printf("%s\n", PACKAGE_STRING);
 858                 printf("See %s for more information.\n\n", PACKAGE_URL);
 859         }
 860
 861         sectorsize = max(sectorsize, (u32)sysconf(_SC_PAGESIZE));
 862         stripesize = sectorsize;
 863         saved_optind = optind;
 864         dev_cnt = argc - optind;
 865         if (dev_cnt == 0)
 866                 print_usage(1);
 867
 868         if (source_dir_set && dev_cnt > 1) {
 869                 error("the option -r is limited to a single device");
 870                 goto error;
 871         }
 872         if (shrink_rootdir && !source_dir_set) {
 873                 error("the option --shrink must be used with --rootdir");
 874                 goto error;
 875         }
 876
 877         if (*fs_uuid) {
 878                 uuid_t dummy_uuid;
 879
 880                 if (uuid_parse(fs_uuid, dummy_uuid) != 0) {
 881                         error("could not parse UUID: %s", fs_uuid);
 882                         goto error;
 883                 }
 884                 if (!test_uuid_unique(fs_uuid)) {
 885                         error("non-unique UUID: %s", fs_uuid);
 886                         goto error;
 887                 }
 888         }
 889
 890         while (dev_cnt-- > 0) {
 891                 file = argv[optind++];
 892                 if (source_dir_set && is_path_exist(file) == 0)
 893                         ret = 0;
 894                 else if (is_block_device(file) == 1)
 895                         ret = test_dev_for_mkfs(file, force_overwrite);
 896                 else
 897                         ret = test_status_for_mkfs(file, force_overwrite);
 898
 899                 if (ret)
 900                         goto error;
 901         }
 902
 903         optind = saved_optind;
 904         dev_cnt = argc - optind;
 905
 906         file = argv[optind++];
 907         ssd = is_ssd(file);
 908
 909         /*
 910         * Set default profiles according to number of added devices.
 911         * For mixed groups defaults are single/single.
 912         */
 913         if (!mixed) {
 914                 if (!metadata_profile_opt) {
 915                         if (dev_cnt == 1 && ssd && verbose)
 916                                 printf("Detected a SSD, turning off metadata "
 917                                 "duplication.  Mkfs with -m dup if you want to "
 918                                 "force metadata duplication.\n");
 919
 920                         metadata_profile = (dev_cnt > 1) ?
 921                                         BTRFS_BLOCK_GROUP_RAID1 : (ssd) ?
 922                                         0: BTRFS_BLOCK_GROUP_DUP;
 923                 }
 924                 if (!data_profile_opt) {
 925                         data_profile = (dev_cnt > 1) ?
 926                                 BTRFS_BLOCK_GROUP_RAID0 : 0; /* raid0 or single */
 927                 }
 928         } else {
 929                 u32 best_nodesize = max_t(u32, sysconf(_SC_PAGESIZE), sectorsize);
 930
 931                 if (metadata_profile_opt || data_profile_opt) {
 932                         if (metadata_profile != data_profile) {
 933                                 error(
 934         "with mixed block groups data and metadata profiles must be the same");
 935                                 goto error;
 936                         }
 937                 }
 938
 939                 if (!nodesize_forced)
 940                         nodesize = best_nodesize;
 941         }
 942
 943         /*
 944          * FS features that can be set by other means than -O
 945          * just set the bit here
 946          */
 947         if (mixed)
 948                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS;
 949
 950         if ((data_profile | metadata_profile) &
 951             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
 952                 features |= BTRFS_FEATURE_INCOMPAT_RAID56;
 953         }
 954
 955         if (btrfs_check_nodesize(nodesize, sectorsize,
 956                                  features))
 957                 goto error;
 958
 959         if (sectorsize < sizeof(struct btrfs_super_block)) {
 960                 error("sectorsize smaller than superblock: %u < %zu",
 961                                 sectorsize, sizeof(struct btrfs_super_block));
 962                 goto error;
 963         }
 964
 965         min_dev_size = btrfs_min_dev_size(nodesize, mixed, metadata_profile,
 966                                           data_profile);
 967         /*
 968          * Enlarge the destination file or create a new one, using the size
 969          * calculated from source dir.
 970          *
 971          * This must be done before minimal device size checks.
 972          */
 973         if (source_dir_set) {
 974                 int oflags = O_RDWR;
 975                 struct stat statbuf;
 976
 977                 if (is_path_exist(file) == 0)
 978                         oflags |= O_CREAT;
 979
 980                 fd = open(file, oflags, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP |
 981                                          S_IROTH);
 982                 if (fd < 0) {
 983                         error("unable to open %s: %m", file);
 984                         goto error;
 985                 }
 986
 987                 ret = fstat(fd, &statbuf);
 988                 if (ret < 0) {
 989                         error("unable to stat %s: %m", file);
 990                         ret = -errno;
 991                         goto error;
 992                 }
 993
 994                 /*
 995                  * Block_count not specified, use file/device size first.
 996                  * Or we will always use source_dir_size calculated for mkfs.
 997                  */
 998                 if (!block_count)
 999                         block_count = btrfs_device_size(fd, &statbuf);
1000                 source_dir_size = btrfs_mkfs_size_dir(source_dir, sectorsize,
1001                                 min_dev_size, metadata_profile, data_profile);
1002                 if (block_count < source_dir_size)
1003                         block_count = source_dir_size;
1004                 ret = zero_output_file(fd, block_count);
1005                 if (ret) {
1006                         error("unable to zero the output file");
1007                         close(fd);
1008                         goto error;
1009                 }
1010                 /* our "device" is the new image file */
1011                 dev_block_count = block_count;
1012                 close(fd);
1013         }
1014         /* Check device/block_count after the nodesize is determined */
1015         if (block_count && block_count < min_dev_size) {
1016                 error("size %llu is too small to make a usable filesystem",
1017                         block_count);
1018                 error("minimum size for btrfs filesystem is %llu",
1019                         min_dev_size);
1020                 goto error;
1021         }
1022         for (i = saved_optind; i < saved_optind + dev_cnt; i++) {
1023                 char *path;
1024
1025                 path = argv[i];
1026                 ret = test_minimum_size(path, min_dev_size);
1027                 if (ret < 0) {
1028                         error("failed to check size for %s: %m", path);
1029                         goto error;
1030                 }
1031                 if (ret > 0) {
1032                         error("'%s' is too small to make a usable filesystem",
1033                                 path);
1034                         error("minimum size for each btrfs device is %llu",
1035                                 min_dev_size);
1036                         goto error;
1037                 }
1038         }
1039         ret = test_num_disk_vs_raid(metadata_profile, data_profile,
1040                         dev_cnt, mixed, ssd);
1041         if (ret)
1042                 goto error;
1043
1044         dev_cnt--;
1045
1046         /*
1047          * Open without O_EXCL so that the problem should not occur by the
1048          * following operation in kernel:
1049          * (btrfs_register_one_device() fails if O_EXCL is on)
1050          */
1051         fd = open(file, O_RDWR);
1052         if (fd < 0) {
1053                 error("unable to open %s: %m", file);
1054                 goto error;
1055         }
1056         ret = btrfs_prepare_device(fd, file, &dev_block_count, block_count,
1057                         (zero_end ? PREP_DEVICE_ZERO_END : 0) |
1058                         (discard ? PREP_DEVICE_DISCARD : 0) |
1059                         (verbose ? PREP_DEVICE_VERBOSE : 0));
1060         if (ret)
1061                 goto error;
1062         if (block_count && block_count > dev_block_count) {
1063                 error("%s is smaller than requested size, expected %llu, found %llu",
1064                       file, (unsigned long long)block_count,
1065                       (unsigned long long)dev_block_count);
1066                 goto error;
1067         }
1068
1069         /* To create the first block group and chunk 0 in make_btrfs */
1070         if (dev_block_count < BTRFS_MKFS_SYSTEM_GROUP_SIZE) {
1071                 error("device is too small to make filesystem, must be at least %llu",
1072                                 (unsigned long long)BTRFS_MKFS_SYSTEM_GROUP_SIZE);
1073                 goto error;
1074         }
1075
1076         if (group_profile_max_safe_loss(metadata_profile) <
1077                 group_profile_max_safe_loss(data_profile)){
1078                 warning("metadata has lower redundancy than data!\n");
1079         }
1080
1081         mkfs_cfg.label = label;
1082         memcpy(mkfs_cfg.fs_uuid, fs_uuid, sizeof(mkfs_cfg.fs_uuid));
1083         mkfs_cfg.num_bytes = dev_block_count;
1084         mkfs_cfg.nodesize = nodesize;
1085         mkfs_cfg.sectorsize = sectorsize;
1086         mkfs_cfg.stripesize = stripesize;
1087         mkfs_cfg.features = features;
1088
1089         ret = make_btrfs(fd, &mkfs_cfg);
1090         if (ret) {
1091                 error("error during mkfs: %s", strerror(-ret));
1092                 goto error;
1093         }
1094
1095         fs_info = open_ctree_fs_info(file, 0, 0, 0,
1096                         OPEN_CTREE_WRITES | OPEN_CTREE_FS_PARTIAL);
1097         if (!fs_info) {
1098                 error("open ctree failed");
1099                 goto error;
1100         }
1101         close(fd);
1102         fd = -1;
1103         root = fs_info->fs_root;
1104         fs_info->alloc_start = alloc_start;
1105
1106         ret = create_metadata_block_groups(root, mixed, &allocation);
1107         if (ret) {
1108                 error("failed to create default block groups: %d", ret);
1109                 goto error;
1110         }
1111
1112         trans = btrfs_start_transaction(root, 1);
1113         if (IS_ERR(trans)) {
1114                 error("failed to start transaction");
1115                 goto error;
1116         }
1117
1118         ret = create_data_block_groups(trans, root, mixed, &allocation);
1119         if (ret) {
1120                 error("failed to create default data block groups: %d", ret);
1121                 goto error;
1122         }
1123
1124         ret = make_root_dir(trans, root);
1125         if (ret) {
1126                 error("failed to setup the root directory: %d", ret);
1127                 goto error;
1128         }
1129
1130         ret = btrfs_commit_transaction(trans, root);
1131         if (ret) {
1132                 error("unable to commit transaction: %d", ret);
1133                 goto out;
1134         }
1135
1136         trans = btrfs_start_transaction(root, 1);
1137         if (IS_ERR(trans)) {
1138                 error("failed to start transaction");
1139                 goto error;
1140         }
1141
1142         if (dev_cnt == 0)
1143                 goto raid_groups;
1144
1145         while (dev_cnt-- > 0) {
1146                 file = argv[optind++];
1147
1148                 /*
1149                  * open without O_EXCL so that the problem should not
1150                  * occur by the following processing.
1151                  * (btrfs_register_one_device() fails if O_EXCL is on)
1152                  */
1153                 fd = open(file, O_RDWR);
1154                 if (fd < 0) {
1155                         error("unable to open %s: %m", file);
1156                         goto error;
1157                 }
1158                 ret = btrfs_device_already_in_root(root, fd,
1159                                                    BTRFS_SUPER_INFO_OFFSET);
1160                 if (ret) {
1161                         error("skipping duplicate device %s in the filesystem",
1162                                 file);
1163                         close(fd);
1164                         continue;
1165                 }
1166                 ret = btrfs_prepare_device(fd, file, &dev_block_count,
1167                                 block_count,
1168                                 (verbose ? PREP_DEVICE_VERBOSE : 0) |
1169                                 (zero_end ? PREP_DEVICE_ZERO_END : 0) |
1170                                 (discard ? PREP_DEVICE_DISCARD : 0));
1171                 if (ret) {
1172                         goto error;
1173                 }
1174
1175                 ret = btrfs_add_to_fsid(trans, root, fd, file, dev_block_count,
1176                                         sectorsize, sectorsize, sectorsize);
1177                 if (ret) {
1178                         error("unable to add %s to filesystem: %d", file, ret);
1179                         goto out;
1180                 }
1181                 if (verbose >= 2) {
1182                         struct btrfs_device *device;
1183
1184                         device = container_of(fs_info->fs_devices->devices.next,
1185                                         struct btrfs_device, dev_list);
1186                         printf("adding device %s id %llu\n", file,
1187                                 (unsigned long long)device->devid);
1188                 }
1189         }
1190
1191 raid_groups:
1192         ret = create_raid_groups(trans, root, data_profile,
1193                          metadata_profile, mixed, &allocation);
1194         if (ret) {
1195                 error("unable to create raid groups: %d", ret);
1196                 goto out;
1197         }
1198
1199         ret = create_tree(trans, root, BTRFS_DATA_RELOC_TREE_OBJECTID);
1200         if (ret) {
1201                 error("unable to create data reloc tree: %d", ret);
1202                 goto out;
1203         }
1204
1205         ret = btrfs_commit_transaction(trans, root);
1206         if (ret) {
1207                 error("unable to commit transaction: %d", ret);
1208                 goto out;
1209         }
1210
1211         ret = cleanup_temp_chunks(fs_info, &allocation, data_profile,
1212                                   metadata_profile, metadata_profile);
1213         if (ret < 0) {
1214                 error("failed to cleanup temporary chunks: %d", ret);
1215                 goto out;
1216         }
1217
1218         if (source_dir_set) {
1219                 ret = btrfs_mkfs_fill_dir(source_dir, root, verbose);
1220                 if (ret) {
1221                         error("error wihle filling filesystem: %d", ret);
1222                         goto out;
1223                 }
1224                 if (shrink_rootdir) {
1225                         ret = btrfs_mkfs_shrink_fs(fs_info, &shrink_size,
1226                                                    shrink_rootdir);
1227                         if (ret < 0) {
1228                                 error("error while shrinking filesystem: %d",
1229                                         ret);
1230                                 goto out;
1231                         }
1232                 }
1233         }
1234
1235         if (verbose) {
1236                 char features_buf[64];
1237
1238                 update_chunk_allocation(fs_info, &allocation);
1239                 printf("Label:              %s\n", label);
1240                 printf("UUID:               %s\n", mkfs_cfg.fs_uuid);
1241                 printf("Node size:          %u\n", nodesize);
1242                 printf("Sector size:        %u\n", sectorsize);
1243                 printf("Filesystem size:    %s\n",
1244                         pretty_size(btrfs_super_total_bytes(fs_info->super_copy)));
1245                 printf("Block group profiles:\n");
1246                 if (allocation.data)
1247                         printf("  Data:             %-8s %16s\n",
1248                                 btrfs_group_profile_str(data_profile),
1249                                 pretty_size(allocation.data));
1250                 if (allocation.metadata)
1251                         printf("  Metadata:         %-8s %16s\n",
1252                                 btrfs_group_profile_str(metadata_profile),
1253                                 pretty_size(allocation.metadata));
1254                 if (allocation.mixed)
1255                         printf("  Data+Metadata:    %-8s %16s\n",
1256                                 btrfs_group_profile_str(data_profile),
1257                                 pretty_size(allocation.mixed));
1258                 printf("  System:           %-8s %16s\n",
1259                         btrfs_group_profile_str(metadata_profile),
1260                         pretty_size(allocation.system));
1261                 printf("SSD detected:       %s\n", ssd ? "yes" : "no");
1262                 btrfs_parse_features_to_string(features_buf, features);
1263                 printf("Incompat features:  %s", features_buf);
1264                 printf("\n");
1265
1266                 list_all_devices(root);
1267         }
1268
1269         /*
1270          * The filesystem is now fully set up, commit the remaining changes and
1271          * fix the signature as the last step before closing the devices.
1272          */
1273         fs_info->finalize_on_close = 1;
1274 out:
1275         close_ret = close_ctree(root);
1276
1277         if (!close_ret) {
1278                 optind = saved_optind;
1279                 dev_cnt = argc - optind;
1280                 while (dev_cnt-- > 0) {
1281                         file = argv[optind++];
1282                         if (is_block_device(file) == 1)
1283                                 btrfs_register_one_device(file);
1284                 }
1285         }
1286
1287         btrfs_close_all_devices();
1288         free(label);
1289
1290         return !!ret;
1291 error:
1292         if (fd > 0)
1293                 close(fd);
1294
1295         free(label);
1296         exit(1);
1297 success:
1298         exit(0);
1299 }