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 <limits.h>
  30 #include <unistd.h>
  31 #include <getopt.h>
  32 #include <uuid/uuid.h>
  33 #include <ctype.h>
  34 #include <blkid/blkid.h>
  35 #include "ctree.h"
  36 #include "disk-io.h"
  37 #include "volumes.h"
  38 #include "transaction.h"
  39 #include "utils.h"
  40 #include "list_sort.h"
  41 #include "help.h"
  42 #include "mkfs/common.h"
  43 #include "mkfs/rootdir.h"
  44 #include "fsfeatures.h"
  45
  46 static int verbose = 1;
  47
  48 struct mkfs_allocation {
  49         u64 data;
  50         u64 metadata;
  51         u64 mixed;
  52         u64 system;
  53 };
  54
  55 static int create_metadata_block_groups(struct btrfs_root *root, int mixed,
  56                                 struct mkfs_allocation *allocation)
  57 {
  58         struct btrfs_fs_info *fs_info = root->fs_info;
  59         struct btrfs_trans_handle *trans;
  60         u64 bytes_used;
  61         u64 chunk_start = 0;
  62         u64 chunk_size = 0;
  63         int ret;
  64
  65         trans = btrfs_start_transaction(root, 1);
  66         BUG_ON(IS_ERR(trans));
  67         bytes_used = btrfs_super_bytes_used(fs_info->super_copy);
  68
  69         root->fs_info->system_allocs = 1;
  70         /*
  71          * First temporary system chunk must match the chunk layout
  72          * created in make_btrfs().
  73          */
  74         ret = btrfs_make_block_group(trans, fs_info, bytes_used,
  75                                      BTRFS_BLOCK_GROUP_SYSTEM,
  76                                      BTRFS_BLOCK_RESERVED_1M_FOR_SUPER,
  77                                      BTRFS_MKFS_SYSTEM_GROUP_SIZE);
  78         allocation->system += BTRFS_MKFS_SYSTEM_GROUP_SIZE;
  79         if (ret)
  80                 return ret;
  81
  82         if (mixed) {
  83                 ret = btrfs_alloc_chunk(trans, fs_info,
  84                                         &chunk_start, &chunk_size,
  85                                         BTRFS_BLOCK_GROUP_METADATA |
  86                                         BTRFS_BLOCK_GROUP_DATA);
  87                 if (ret == -ENOSPC) {
  88                         error("no space to allocate data/metadata chunk");
  89                         goto err;
  90                 }
  91                 if (ret)
  92                         return ret;
  93                 ret = btrfs_make_block_group(trans, fs_info, 0,
  94                                              BTRFS_BLOCK_GROUP_METADATA |
  95                                              BTRFS_BLOCK_GROUP_DATA,
  96                                              chunk_start, chunk_size);
  97                 if (ret)
  98                         return ret;
  99                 allocation->mixed += chunk_size;
 100         } else {
 101                 ret = btrfs_alloc_chunk(trans, fs_info,
 102                                         &chunk_start, &chunk_size,
 103                                         BTRFS_BLOCK_GROUP_METADATA);
 104                 if (ret == -ENOSPC) {
 105                         error("no space to allocate metadata chunk");
 106                         goto err;
 107                 }
 108                 if (ret)
 109                         return ret;
 110                 ret = btrfs_make_block_group(trans, fs_info, 0,
 111                                              BTRFS_BLOCK_GROUP_METADATA,
 112                                              chunk_start, chunk_size);
 113                 allocation->metadata += chunk_size;
 114                 if (ret)
 115                         return ret;
 116         }
 117
 118         root->fs_info->system_allocs = 0;
 119         ret = btrfs_commit_transaction(trans, root);
 120
 121 err:
 122         return ret;
 123 }
 124
 125 static int create_data_block_groups(struct btrfs_trans_handle *trans,
 126                 struct btrfs_root *root, int mixed,
 127                 struct mkfs_allocation *allocation)
 128 {
 129         struct btrfs_fs_info *fs_info = root->fs_info;
 130         u64 chunk_start = 0;
 131         u64 chunk_size = 0;
 132         int ret = 0;
 133
 134         if (!mixed) {
 135                 ret = btrfs_alloc_chunk(trans, fs_info,
 136                                         &chunk_start, &chunk_size,
 137                                         BTRFS_BLOCK_GROUP_DATA);
 138                 if (ret == -ENOSPC) {
 139                         error("no space to allocate data chunk");
 140                         goto err;
 141                 }
 142                 if (ret)
 143                         return ret;
 144                 ret = btrfs_make_block_group(trans, fs_info, 0,
 145                                              BTRFS_BLOCK_GROUP_DATA,
 146                                              chunk_start, chunk_size);
 147                 allocation->data += chunk_size;
 148                 if (ret)
 149                         return ret;
 150         }
 151
 152 err:
 153         return ret;
 154 }
 155
 156 static int make_root_dir(struct btrfs_trans_handle *trans,
 157                 struct btrfs_root *root)
 158 {
 159         struct btrfs_key location;
 160         int ret;
 161
 162         ret = btrfs_make_root_dir(trans, root->fs_info->tree_root,
 163                               BTRFS_ROOT_TREE_DIR_OBJECTID);
 164         if (ret)
 165                 goto err;
 166         ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
 167         if (ret)
 168                 goto err;
 169         memcpy(&location, &root->fs_info->fs_root->root_key, sizeof(location));
 170         location.offset = (u64)-1;
 171         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
 172                         "default", 7,
 173                         btrfs_super_root_dir(root->fs_info->super_copy),
 174                         &location, BTRFS_FT_DIR, 0);
 175         if (ret)
 176                 goto err;
 177
 178         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 179                              "default", 7, location.objectid,
 180                              BTRFS_ROOT_TREE_DIR_OBJECTID, 0);
 181         if (ret)
 182                 goto err;
 183
 184 err:
 185         return ret;
 186 }
 187
 188 static int __recow_root(struct btrfs_trans_handle *trans,
 189                          struct btrfs_root *root)
 190 {
 191         struct extent_buffer *tmp;
 192         int ret;
 193
 194         if (trans->transid != btrfs_root_generation(&root->root_item)) {
 195                 extent_buffer_get(root->node);
 196                 ret = __btrfs_cow_block(trans, root, root->node,
 197                                         NULL, 0, &tmp, 0, 0);
 198                 if (ret)
 199                         return ret;
 200                 free_extent_buffer(tmp);
 201         }
 202
 203         return 0;
 204 }
 205
 206 static int recow_roots(struct btrfs_trans_handle *trans,
 207                        struct btrfs_root *root)
 208 {
 209         struct btrfs_fs_info *info = root->fs_info;
 210         int ret;
 211
 212         ret = __recow_root(trans, info->fs_root);
 213         if (ret)
 214                 return ret;
 215         ret = __recow_root(trans, info->tree_root);
 216         if (ret)
 217                 return ret;
 218         ret = __recow_root(trans, info->extent_root);
 219         if (ret)
 220                 return ret;
 221         ret = __recow_root(trans, info->chunk_root);
 222         if (ret)
 223                 return ret;
 224         ret = __recow_root(trans, info->dev_root);
 225         if (ret)
 226                 return ret;
 227         ret = __recow_root(trans, info->csum_root);
 228         if (ret)
 229                 return ret;
 230
 231         return 0;
 232 }
 233
 234 static int create_one_raid_group(struct btrfs_trans_handle *trans,
 235                               struct btrfs_root *root, u64 type,
 236                               struct mkfs_allocation *allocation)
 237
 238 {
 239         struct btrfs_fs_info *fs_info = root->fs_info;
 240         u64 chunk_start;
 241         u64 chunk_size;
 242         int ret;
 243
 244         ret = btrfs_alloc_chunk(trans, fs_info,
 245                                 &chunk_start, &chunk_size, type);
 246         if (ret == -ENOSPC) {
 247                 error("not enough free space to allocate chunk");
 248                 exit(1);
 249         }
 250         if (ret)
 251                 return ret;
 252
 253         ret = btrfs_make_block_group(trans, fs_info, 0,
 254                                      type, chunk_start, chunk_size);
 255
 256         type &= BTRFS_BLOCK_GROUP_TYPE_MASK;
 257         if (type == BTRFS_BLOCK_GROUP_DATA) {
 258                 allocation->data += chunk_size;
 259         } else if (type == BTRFS_BLOCK_GROUP_METADATA) {
 260                 allocation->metadata += chunk_size;
 261         } else if (type == BTRFS_BLOCK_GROUP_SYSTEM) {
 262                 allocation->system += chunk_size;
 263         } else if (type ==
 264                         (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA)) {
 265                 allocation->mixed += chunk_size;
 266         } else {
 267                 error("unrecognized profile type: 0x%llx",
 268                                 (unsigned long long)type);
 269                 ret = -EINVAL;
 270         }
 271
 272         return ret;
 273 }
 274
 275 static int create_raid_groups(struct btrfs_trans_handle *trans,
 276                               struct btrfs_root *root, u64 data_profile,
 277                               u64 metadata_profile, int mixed,
 278                               struct mkfs_allocation *allocation)
 279 {
 280         int ret;
 281
 282         if (metadata_profile) {
 283                 u64 meta_flags = BTRFS_BLOCK_GROUP_METADATA;
 284
 285                 ret = create_one_raid_group(trans, root,
 286                                             BTRFS_BLOCK_GROUP_SYSTEM |
 287                                             metadata_profile, allocation);
 288                 if (ret)
 289                         return ret;
 290
 291                 if (mixed)
 292                         meta_flags |= BTRFS_BLOCK_GROUP_DATA;
 293
 294                 ret = create_one_raid_group(trans, root, meta_flags |
 295                                             metadata_profile, allocation);
 296                 if (ret)
 297                         return ret;
 298
 299         }
 300         if (!mixed && data_profile) {
 301                 ret = create_one_raid_group(trans, root,
 302                                             BTRFS_BLOCK_GROUP_DATA |
 303                                             data_profile, allocation);
 304                 if (ret)
 305                         return ret;
 306         }
 307         ret = recow_roots(trans, root);
 308
 309         return ret;
 310 }
 311
 312 static int create_tree(struct btrfs_trans_handle *trans,
 313                         struct btrfs_root *root, u64 objectid)
 314 {
 315         struct btrfs_key location;
 316         struct btrfs_root_item root_item;
 317         struct extent_buffer *tmp;
 318         int ret;
 319
 320         ret = btrfs_copy_root(trans, root, root->node, &tmp, objectid);
 321         if (ret)
 322                 return ret;
 323
 324         memcpy(&root_item, &root->root_item, sizeof(root_item));
 325         btrfs_set_root_bytenr(&root_item, tmp->start);
 326         btrfs_set_root_level(&root_item, btrfs_header_level(tmp));
 327         btrfs_set_root_generation(&root_item, trans->transid);
 328         free_extent_buffer(tmp);
 329
 330         location.objectid = objectid;
 331         location.type = BTRFS_ROOT_ITEM_KEY;
 332         location.offset = 0;
 333         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
 334                                 &location, &root_item);
 335
 336         return ret;
 337 }
 338
 339 static void print_usage(int ret)
 340 {
 341         printf("Usage: mkfs.btrfs [options] dev [ dev ... ]\n");
 342         printf("Options:\n");
 343         printf("  allocation profiles:\n");
 344         printf("\t-d|--data PROFILE       data profile, raid0, raid1, raid5, raid6, raid10, dup or single\n");
 345         printf("\t-m|--metadata PROFILE   metadata profile, values like for data profile\n");
 346         printf("\t-M|--mixed              mix metadata and data together\n");
 347         printf("  features:\n");
 348         printf("\t-n|--nodesize SIZE      size of btree nodes\n");
 349         printf("\t-s|--sectorsize SIZE    data block size (may not be mountable by current kernel)\n");
 350         printf("\t-O|--features LIST      comma separated list of filesystem features (use '-O list-all' to list features)\n");
 351         printf("\t-L|--label LABEL        set the filesystem label\n");
 352         printf("\t-U|--uuid UUID          specify the filesystem UUID (must be unique)\n");
 353         printf("  creation:\n");
 354         printf("\t-b|--byte-count SIZE    set filesystem size to SIZE (on the first device)\n");
 355         printf("\t-r|--rootdir DIR        copy files from DIR to the image root directory\n");
 356         printf("\t--shrink                (with --rootdir) shrink the filled filesystem to minimal size\n");
 357         printf("\t-K|--nodiscard          do not perform whole device TRIM\n");
 358         printf("\t-f|--force              force overwrite of existing filesystem\n");
 359         printf("  general:\n");
 360         printf("\t-q|--quiet              no messages except errors\n");
 361         printf("\t-V|--version            print the mkfs.btrfs version and exit\n");
 362         printf("\t--help                  print this help and exit\n");
 363         printf("  deprecated:\n");
 364         printf("\t-A|--alloc-start START  the offset to start the filesystem\n");
 365         printf("\t-l|--leafsize SIZE      deprecated, alias for nodesize\n");
 366         exit(ret);
 367 }
 368
 369 static u64 parse_profile(const char *s)
 370 {
 371         if (strcasecmp(s, "raid0") == 0) {
 372                 return BTRFS_BLOCK_GROUP_RAID0;
 373         } else if (strcasecmp(s, "raid1") == 0) {
 374                 return BTRFS_BLOCK_GROUP_RAID1;
 375         } else if (strcasecmp(s, "raid5") == 0) {
 376                 return BTRFS_BLOCK_GROUP_RAID5;
 377         } else if (strcasecmp(s, "raid6") == 0) {
 378                 return BTRFS_BLOCK_GROUP_RAID6;
 379         } else if (strcasecmp(s, "raid10") == 0) {
 380                 return BTRFS_BLOCK_GROUP_RAID10;
 381         } else if (strcasecmp(s, "dup") == 0) {
 382                 return BTRFS_BLOCK_GROUP_DUP;
 383         } else if (strcasecmp(s, "single") == 0) {
 384                 return 0;
 385         } else {
 386                 error("unknown profile %s", s);
 387                 exit(1);
 388         }
 389         /* not reached */
 390         return 0;
 391 }
 392
 393 static char *parse_label(const char *input)
 394 {
 395         int len = strlen(input);
 396
 397         if (len >= BTRFS_LABEL_SIZE) {
 398                 error("label %s is too long (max %d)", input,
 399                         BTRFS_LABEL_SIZE - 1);
 400                 exit(1);
 401         }
 402         return strdup(input);
 403 }
 404
 405 static int zero_output_file(int out_fd, u64 size)
 406 {
 407         int loop_num;
 408         u64 location = 0;
 409         char buf[SZ_4K];
 410         int ret = 0, i;
 411         ssize_t written;
 412
 413         memset(buf, 0, SZ_4K);
 414
 415         /* Only zero out the first 1M */
 416         loop_num = SZ_1M / SZ_4K;
 417         for (i = 0; i < loop_num; i++) {
 418                 written = pwrite64(out_fd, buf, SZ_4K, location);
 419                 if (written != SZ_4K)
 420                         ret = -EIO;
 421                 location += SZ_4K;
 422         }
 423
 424         /* Then enlarge the file to size */
 425         written = pwrite64(out_fd, buf, 1, size - 1);
 426         if (written < 1)
 427                 ret = -EIO;
 428         return ret;
 429 }
 430
 431 static int is_ssd(const char *file)
 432 {
 433         blkid_probe probe;
 434         char wholedisk[PATH_MAX];
 435         char sysfs_path[PATH_MAX];
 436         dev_t devno;
 437         int fd;
 438         char rotational;
 439         int ret;
 440
 441         probe = blkid_new_probe_from_filename(file);
 442         if (!probe)
 443                 return 0;
 444
 445         /* Device number of this disk (possibly a partition) */
 446         devno = blkid_probe_get_devno(probe);
 447         if (!devno) {
 448                 blkid_free_probe(probe);
 449                 return 0;
 450         }
 451
 452         /* Get whole disk name (not full path) for this devno */
 453         ret = blkid_devno_to_wholedisk(devno,
 454                         wholedisk, sizeof(wholedisk), NULL);
 455         if (ret) {
 456                 blkid_free_probe(probe);
 457                 return 0;
 458         }
 459
 460         snprintf(sysfs_path, PATH_MAX, "/sys/block/%s/queue/rotational",
 461                  wholedisk);
 462
 463         blkid_free_probe(probe);
 464
 465         fd = open(sysfs_path, O_RDONLY);
 466         if (fd < 0) {
 467                 return 0;
 468         }
 469
 470         if (read(fd, &rotational, 1) < 1) {
 471                 close(fd);
 472                 return 0;
 473         }
 474         close(fd);
 475
 476         return rotational == '0';
 477 }
 478
 479 static int _cmp_device_by_id(void *priv, struct list_head *a,
 480                              struct list_head *b)
 481 {
 482         return list_entry(a, struct btrfs_device, dev_list)->devid -
 483                list_entry(b, struct btrfs_device, dev_list)->devid;
 484 }
 485
 486 static void list_all_devices(struct btrfs_root *root)
 487 {
 488         struct btrfs_fs_devices *fs_devices;
 489         struct btrfs_device *device;
 490         int number_of_devices = 0;
 491         u64 total_block_count = 0;
 492
 493         fs_devices = root->fs_info->fs_devices;
 494
 495         list_for_each_entry(device, &fs_devices->devices, dev_list)
 496                 number_of_devices++;
 497
 498         list_sort(NULL, &fs_devices->devices, _cmp_device_by_id);
 499
 500         printf("Number of devices:  %d\n", number_of_devices);
 501         /* printf("Total devices size: %10s\n", */
 502                 /* pretty_size(total_block_count)); */
 503         printf("Devices:\n");
 504         printf("   ID        SIZE  PATH\n");
 505         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 506                 printf("  %3llu  %10s  %s\n",
 507                         device->devid,
 508                         pretty_size(device->total_bytes),
 509                         device->name);
 510                 total_block_count += device->total_bytes;
 511         }
 512
 513         printf("\n");
 514 }
 515
 516 static int is_temp_block_group(struct extent_buffer *node,
 517                                struct btrfs_block_group_item *bgi,
 518                                u64 data_profile, u64 meta_profile,
 519                                u64 sys_profile)
 520 {
 521         u64 flag = btrfs_disk_block_group_flags(node, bgi);
 522         u64 flag_type = flag & BTRFS_BLOCK_GROUP_TYPE_MASK;
 523         u64 flag_profile = flag & BTRFS_BLOCK_GROUP_PROFILE_MASK;
 524         u64 used = btrfs_disk_block_group_used(node, bgi);
 525
 526         /*
 527          * Chunks meets all the following conditions is a temp chunk
 528          * 1) Empty chunk
 529          * Temp chunk is always empty.
 530          *
 531          * 2) profile mismatch with mkfs profile.
 532          * Temp chunk is always in SINGLE
 533          *
 534          * 3) Size differs with mkfs_alloc
 535          * Special case for SINGLE/SINGLE btrfs.
 536          * In that case, temp data chunk and real data chunk are always empty.
 537          * So we need to use mkfs_alloc to be sure which chunk is the newly
 538          * allocated.
 539          *
 540          * Normally, new chunk size is equal to mkfs one (One chunk)
 541          * If it has multiple chunks, we just refuse to delete any one.
 542          * As they are all single, so no real problem will happen.
 543          * So only use condition 1) and 2) to judge them.
 544          */
 545         if (used != 0)
 546                 return 0;
 547         switch (flag_type) {
 548         case BTRFS_BLOCK_GROUP_DATA:
 549         case BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA:
 550                 data_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
 551                 if (flag_profile != data_profile)
 552                         return 1;
 553                 break;
 554         case BTRFS_BLOCK_GROUP_METADATA:
 555                 meta_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
 556                 if (flag_profile != meta_profile)
 557                         return 1;
 558                 break;
 559         case BTRFS_BLOCK_GROUP_SYSTEM:
 560                 sys_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
 561                 if (flag_profile != sys_profile)
 562                         return 1;
 563                 break;
 564         }
 565         return 0;
 566 }
 567
 568 /* Note: if current is a block group, it will skip it anyway */
 569 static int next_block_group(struct btrfs_root *root,
 570                             struct btrfs_path *path)
 571 {
 572         struct btrfs_key key;
 573         int ret = 0;
 574
 575         while (1) {
 576                 ret = btrfs_next_item(root, path);
 577                 if (ret)
 578                         goto out;
 579
 580                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 581                 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
 582                         goto out;
 583         }
 584 out:
 585         return ret;
 586 }
 587
 588 /* This function will cleanup  */
 589 static int cleanup_temp_chunks(struct btrfs_fs_info *fs_info,
 590                                struct mkfs_allocation *alloc,
 591                                u64 data_profile, u64 meta_profile,
 592                                u64 sys_profile)
 593 {
 594         struct btrfs_trans_handle *trans = NULL;
 595         struct btrfs_block_group_item *bgi;
 596         struct btrfs_root *root = fs_info->extent_root;
 597         struct btrfs_key key;
 598         struct btrfs_key found_key;
 599         struct btrfs_path path;
 600         int ret = 0;
 601
 602         btrfs_init_path(&path);
 603         trans = btrfs_start_transaction(root, 1);
 604         BUG_ON(IS_ERR(trans));
 605
 606         key.objectid = 0;
 607         key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
 608         key.offset = 0;
 609
 610         while (1) {
 611                 /*
 612                  * as the rest of the loop may modify the tree, we need to
 613                  * start a new search each time.
 614                  */
 615                 ret = btrfs_search_slot(trans, root, &key, &path, 0, 0);
 616                 if (ret < 0)
 617                         goto out;
 618                 /* Don't pollute ret for >0 case */
 619                 if (ret > 0)
 620                         ret = 0;
 621
 622                 btrfs_item_key_to_cpu(path.nodes[0], &found_key,
 623                                       path.slots[0]);
 624                 if (found_key.objectid < key.objectid)
 625                         goto out;
 626                 if (found_key.type != BTRFS_BLOCK_GROUP_ITEM_KEY) {
 627                         ret = next_block_group(root, &path);
 628                         if (ret < 0)
 629                                 goto out;
 630                         if (ret > 0) {
 631                                 ret = 0;
 632                                 goto out;
 633                         }
 634                         btrfs_item_key_to_cpu(path.nodes[0], &found_key,
 635                                               path.slots[0]);
 636                 }
 637
 638                 bgi = btrfs_item_ptr(path.nodes[0], path.slots[0],
 639                                      struct btrfs_block_group_item);
 640                 if (is_temp_block_group(path.nodes[0], bgi,
 641                                         data_profile, meta_profile,
 642                                         sys_profile)) {
 643                         u64 flags = btrfs_disk_block_group_flags(path.nodes[0],
 644                                                              bgi);
 645
 646                         ret = btrfs_free_block_group(trans, fs_info,
 647                                         found_key.objectid, found_key.offset);
 648                         if (ret < 0)
 649                                 goto out;
 650
 651                         if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
 652                             BTRFS_BLOCK_GROUP_DATA)
 653                                 alloc->data -= found_key.offset;
 654                         else if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
 655                                  BTRFS_BLOCK_GROUP_METADATA)
 656                                 alloc->metadata -= found_key.offset;
 657                         else if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
 658                                  BTRFS_BLOCK_GROUP_SYSTEM)
 659                                 alloc->system -= found_key.offset;
 660                         else if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
 661                                  (BTRFS_BLOCK_GROUP_METADATA |
 662                                   BTRFS_BLOCK_GROUP_DATA))
 663                                 alloc->mixed -= found_key.offset;
 664                 }
 665                 btrfs_release_path(&path);
 666                 key.objectid = found_key.objectid + found_key.offset;
 667         }
 668 out:
 669         if (trans)
 670                 btrfs_commit_transaction(trans, root);
 671         btrfs_release_path(&path);
 672         return ret;
 673 }
 674
 675 /*
 676  * Just update chunk allocation info, since --rootdir may allocate new
 677  * chunks which is not updated in @allocation structure.
 678  */
 679 static void update_chunk_allocation(struct btrfs_fs_info *fs_info,
 680                                     struct mkfs_allocation *allocation)
 681 {
 682         struct btrfs_block_group_cache *bg_cache;
 683         const u64 mixed_flag = BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA;
 684         u64 search_start = 0;
 685
 686         allocation->mixed = 0;
 687         allocation->data = 0;
 688         allocation->metadata = 0;
 689         allocation->system = 0;
 690         while (1) {
 691                 bg_cache = btrfs_lookup_first_block_group(fs_info,
 692                                                           search_start);
 693                 if (!bg_cache)
 694                         break;
 695                 if ((bg_cache->flags & mixed_flag) == mixed_flag)
 696                         allocation->mixed += bg_cache->key.offset;
 697                 else if (bg_cache->flags & BTRFS_BLOCK_GROUP_DATA)
 698                         allocation->data += bg_cache->key.offset;
 699                 else if (bg_cache->flags & BTRFS_BLOCK_GROUP_METADATA)
 700                         allocation->metadata += bg_cache->key.offset;
 701                 else
 702                         allocation->system += bg_cache->key.offset;
 703                 search_start = bg_cache->key.objectid + bg_cache->key.offset;
 704         }
 705 }
 706
 707 int main(int argc, char **argv)
 708 {
 709         char *file;
 710         struct btrfs_root *root;
 711         struct btrfs_fs_info *fs_info;
 712         struct btrfs_trans_handle *trans;
 713         char *label = NULL;
 714         u64 block_count = 0;
 715         u64 dev_block_count = 0;
 716         u64 alloc_start = 0;
 717         u64 metadata_profile = 0;
 718         u64 data_profile = 0;
 719         u32 nodesize = max_t(u32, sysconf(_SC_PAGESIZE),
 720                         BTRFS_MKFS_DEFAULT_NODE_SIZE);
 721         u32 sectorsize = 4096;
 722         u32 stripesize = 4096;
 723         int zero_end = 1;
 724         int fd = -1;
 725         int ret = 0;
 726         int close_ret;
 727         int i;
 728         int mixed = 0;
 729         int nodesize_forced = 0;
 730         int data_profile_opt = 0;
 731         int metadata_profile_opt = 0;
 732         int discard = 1;
 733         int ssd = 0;
 734         int force_overwrite = 0;
 735         char *source_dir = NULL;
 736         bool source_dir_set = false;
 737         bool shrink_rootdir = false;
 738         u64 source_dir_size = 0;
 739         u64 min_dev_size;
 740         u64 shrink_size;
 741         int dev_cnt = 0;
 742         int saved_optind;
 743         char fs_uuid[BTRFS_UUID_UNPARSED_SIZE] = { 0 };
 744         u64 features = BTRFS_MKFS_DEFAULT_FEATURES;
 745         struct mkfs_allocation allocation = { 0 };
 746         struct btrfs_mkfs_config mkfs_cfg;
 747
 748         while(1) {
 749                 int c;
 750                 enum { GETOPT_VAL_SHRINK = 257 };
 751                 static const struct option long_options[] = {
 752                         { "alloc-start", required_argument, NULL, 'A'},
 753                         { "byte-count", required_argument, NULL, 'b' },
 754                         { "force", no_argument, NULL, 'f' },
 755                         { "leafsize", required_argument, NULL, 'l' },
 756                         { "label", required_argument, NULL, 'L'},
 757                         { "metadata", required_argument, NULL, 'm' },
 758                         { "mixed", no_argument, NULL, 'M' },
 759                         { "nodesize", required_argument, NULL, 'n' },
 760                         { "sectorsize", required_argument, NULL, 's' },
 761                         { "data", required_argument, NULL, 'd' },
 762                         { "version", no_argument, NULL, 'V' },
 763                         { "rootdir", required_argument, NULL, 'r' },
 764                         { "nodiscard", no_argument, NULL, 'K' },
 765                         { "features", required_argument, NULL, 'O' },
 766                         { "uuid", required_argument, NULL, 'U' },
 767                         { "quiet", 0, NULL, 'q' },
 768                         { "shrink", no_argument, NULL, GETOPT_VAL_SHRINK },
 769                         { "help", no_argument, NULL, GETOPT_VAL_HELP },
 770                         { NULL, 0, NULL, 0}
 771                 };
 772
 773                 c = getopt_long(argc, argv, "A:b:fl:n:s:m:d:L:O:r:U:VMKq",
 774                                 long_options, NULL);
 775                 if (c < 0)
 776                         break;
 777                 switch(c) {
 778                         case 'A':
 779                                 alloc_start = parse_size(optarg);
 780                                 break;
 781                         case 'f':
 782                                 force_overwrite = 1;
 783                                 break;
 784                         case 'd':
 785                                 data_profile = parse_profile(optarg);
 786                                 data_profile_opt = 1;
 787                                 break;
 788                         case 'l':
 789                                 warning("--leafsize is deprecated, use --nodesize");
 790                                 /* fall through */
 791                         case 'n':
 792                                 nodesize = parse_size(optarg);
 793                                 nodesize_forced = 1;
 794                                 break;
 795                         case 'L':
 796                                 label = parse_label(optarg);
 797                                 break;
 798                         case 'm':
 799                                 metadata_profile = parse_profile(optarg);
 800                                 metadata_profile_opt = 1;
 801                                 break;
 802                         case 'M':
 803                                 mixed = 1;
 804                                 break;
 805                         case 'O': {
 806                                 char *orig = strdup(optarg);
 807                                 char *tmp = orig;
 808
 809                                 tmp = btrfs_parse_fs_features(tmp, &features);
 810                                 if (tmp) {
 811                                         error("unrecognized filesystem feature '%s'",
 812                                                         tmp);
 813                                         free(orig);
 814                                         goto error;
 815                                 }
 816                                 free(orig);
 817                                 if (features & BTRFS_FEATURE_LIST_ALL) {
 818                                         btrfs_list_all_fs_features(0);
 819                                         goto success;
 820                                 }
 821                                 break;
 822                                 }
 823                         case 's':
 824                                 sectorsize = parse_size(optarg);
 825                                 break;
 826                         case 'b':
 827                                 block_count = parse_size(optarg);
 828                                 zero_end = 0;
 829                                 break;
 830                         case 'V':
 831                                 printf("mkfs.btrfs, part of %s\n",
 832                                                 PACKAGE_STRING);
 833                                 goto success;
 834                         case 'r':
 835                                 source_dir = optarg;
 836                                 source_dir_set = true;
 837                                 break;
 838                         case 'U':
 839                                 strncpy(fs_uuid, optarg,
 840                                         BTRFS_UUID_UNPARSED_SIZE - 1);
 841                                 break;
 842                         case 'K':
 843                                 discard = 0;
 844                                 break;
 845                         case 'q':
 846                                 verbose = 0;
 847                                 break;
 848                         case GETOPT_VAL_SHRINK:
 849                                 shrink_rootdir = true;
 850                                 break;
 851                         case GETOPT_VAL_HELP:
 852                         default:
 853                                 print_usage(c != GETOPT_VAL_HELP);
 854                 }
 855         }
 856
 857         if (verbose) {
 858                 printf("%s\n", PACKAGE_STRING);
 859                 printf("See %s for more information.\n\n", PACKAGE_URL);
 860         }
 861
 862         sectorsize = max(sectorsize, (u32)sysconf(_SC_PAGESIZE));
 863         stripesize = sectorsize;
 864         saved_optind = optind;
 865         dev_cnt = argc - optind;
 866         if (dev_cnt == 0)
 867                 print_usage(1);
 868
 869         if (source_dir_set && dev_cnt > 1) {
 870                 error("the option -r is limited to a single device");
 871                 goto error;
 872         }
 873         if (shrink_rootdir && !source_dir_set) {
 874                 error("the option --shrink must be used with --rootdir");
 875                 goto error;
 876         }
 877
 878         if (*fs_uuid) {
 879                 uuid_t dummy_uuid;
 880
 881                 if (uuid_parse(fs_uuid, dummy_uuid) != 0) {
 882                         error("could not parse UUID: %s", fs_uuid);
 883                         goto error;
 884                 }
 885                 if (!test_uuid_unique(fs_uuid)) {
 886                         error("non-unique UUID: %s", fs_uuid);
 887                         goto error;
 888                 }
 889         }
 890
 891         while (dev_cnt-- > 0) {
 892                 file = argv[optind++];
 893                 if (source_dir_set && is_path_exist(file) == 0)
 894                         ret = 0;
 895                 else if (is_block_device(file) == 1)
 896                         ret = test_dev_for_mkfs(file, force_overwrite);
 897                 else
 898                         ret = test_status_for_mkfs(file, force_overwrite);
 899
 900                 if (ret)
 901                         goto error;
 902         }
 903
 904         optind = saved_optind;
 905         dev_cnt = argc - optind;
 906
 907         file = argv[optind++];
 908         ssd = is_ssd(file);
 909
 910         /*
 911         * Set default profiles according to number of added devices.
 912         * For mixed groups defaults are single/single.
 913         */
 914         if (!mixed) {
 915                 if (!metadata_profile_opt) {
 916                         if (dev_cnt == 1 && ssd && verbose)
 917                                 printf("Detected a SSD, turning off metadata "
 918                                 "duplication.  Mkfs with -m dup if you want to "
 919                                 "force metadata duplication.\n");
 920
 921                         metadata_profile = (dev_cnt > 1) ?
 922                                         BTRFS_BLOCK_GROUP_RAID1 : (ssd) ?
 923                                         0: BTRFS_BLOCK_GROUP_DUP;
 924                 }
 925                 if (!data_profile_opt) {
 926                         data_profile = (dev_cnt > 1) ?
 927                                 BTRFS_BLOCK_GROUP_RAID0 : 0; /* raid0 or single */
 928                 }
 929         } else {
 930                 u32 best_nodesize = max_t(u32, sysconf(_SC_PAGESIZE), sectorsize);
 931
 932                 if (metadata_profile_opt || data_profile_opt) {
 933                         if (metadata_profile != data_profile) {
 934                                 error(
 935         "with mixed block groups data and metadata profiles must be the same");
 936                                 goto error;
 937                         }
 938                 }
 939
 940                 if (!nodesize_forced)
 941                         nodesize = best_nodesize;
 942         }
 943
 944         /*
 945          * FS features that can be set by other means than -O
 946          * just set the bit here
 947          */
 948         if (mixed)
 949                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS;
 950
 951         if ((data_profile | metadata_profile) &
 952             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
 953                 features |= BTRFS_FEATURE_INCOMPAT_RAID56;
 954         }
 955
 956         if (btrfs_check_nodesize(nodesize, sectorsize,
 957                                  features))
 958                 goto error;
 959
 960         if (sectorsize < sizeof(struct btrfs_super_block)) {
 961                 error("sectorsize smaller than superblock: %u < %zu",
 962                                 sectorsize, sizeof(struct btrfs_super_block));
 963                 goto error;
 964         }
 965
 966         min_dev_size = btrfs_min_dev_size(nodesize, mixed, metadata_profile,
 967                                           data_profile);
 968         /*
 969          * Enlarge the destination file or create a new one, using the size
 970          * calculated from source dir.
 971          *
 972          * This must be done before minimal device size checks.
 973          */
 974         if (source_dir_set) {
 975                 int oflags = O_RDWR;
 976                 struct stat statbuf;
 977
 978                 if (is_path_exist(file) == 0)
 979                         oflags |= O_CREAT;
 980
 981                 fd = open(file, oflags, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP |
 982                                          S_IROTH);
 983                 if (fd < 0) {
 984                         error("unable to open %s: %m", file);
 985                         goto error;
 986                 }
 987
 988                 ret = fstat(fd, &statbuf);
 989                 if (ret < 0) {
 990                         error("unable to stat %s: %m", file);
 991                         ret = -errno;
 992                         goto error;
 993                 }
 994
 995                 /*
 996                  * Block_count not specified, use file/device size first.
 997                  * Or we will always use source_dir_size calculated for mkfs.
 998                  */
 999                 if (!block_count)
1000                         block_count = btrfs_device_size(fd, &statbuf);
1001                 source_dir_size = btrfs_mkfs_size_dir(source_dir, sectorsize,
1002                                 min_dev_size, metadata_profile, data_profile);
1003                 if (block_count < source_dir_size)
1004                         block_count = source_dir_size;
1005                 ret = zero_output_file(fd, block_count);
1006                 if (ret) {
1007                         error("unable to zero the output file");
1008                         close(fd);
1009                         goto error;
1010                 }
1011                 /* our "device" is the new image file */
1012                 dev_block_count = block_count;
1013                 close(fd);
1014         }
1015         /* Check device/block_count after the nodesize is determined */
1016         if (block_count && block_count < min_dev_size) {
1017                 error("size %llu is too small to make a usable filesystem",
1018                         block_count);
1019                 error("minimum size for btrfs filesystem is %llu",
1020                         min_dev_size);
1021                 goto error;
1022         }
1023         for (i = saved_optind; i < saved_optind + dev_cnt; i++) {
1024                 char *path;
1025
1026                 path = argv[i];
1027                 ret = test_minimum_size(path, min_dev_size);
1028                 if (ret < 0) {
1029                         error("failed to check size for %s: %m", path);
1030                         goto error;
1031                 }
1032                 if (ret > 0) {
1033                         error("'%s' is too small to make a usable filesystem",
1034                                 path);
1035                         error("minimum size for each btrfs device is %llu",
1036                                 min_dev_size);
1037                         goto error;
1038                 }
1039         }
1040         ret = test_num_disk_vs_raid(metadata_profile, data_profile,
1041                         dev_cnt, mixed, ssd);
1042         if (ret)
1043                 goto error;
1044
1045         dev_cnt--;
1046
1047         /*
1048          * Open without O_EXCL so that the problem should not occur by the
1049          * following operation in kernel:
1050          * (btrfs_register_one_device() fails if O_EXCL is on)
1051          */
1052         fd = open(file, O_RDWR);
1053         if (fd < 0) {
1054                 error("unable to open %s: %m", file);
1055                 goto error;
1056         }
1057         ret = btrfs_prepare_device(fd, file, &dev_block_count, block_count,
1058                         (zero_end ? PREP_DEVICE_ZERO_END : 0) |
1059                         (discard ? PREP_DEVICE_DISCARD : 0) |
1060                         (verbose ? PREP_DEVICE_VERBOSE : 0));
1061         if (ret)
1062                 goto error;
1063         if (block_count && block_count > dev_block_count) {
1064                 error("%s is smaller than requested size, expected %llu, found %llu",
1065                       file, (unsigned long long)block_count,
1066                       (unsigned long long)dev_block_count);
1067                 goto error;
1068         }
1069
1070         /* To create the first block group and chunk 0 in make_btrfs */
1071         if (dev_block_count < BTRFS_MKFS_SYSTEM_GROUP_SIZE) {
1072                 error("device is too small to make filesystem, must be at least %llu",
1073                                 (unsigned long long)BTRFS_MKFS_SYSTEM_GROUP_SIZE);
1074                 goto error;
1075         }
1076
1077         if (group_profile_max_safe_loss(metadata_profile) <
1078                 group_profile_max_safe_loss(data_profile)){
1079                 warning("metadata has lower redundancy than data!\n");
1080         }
1081
1082         mkfs_cfg.label = label;
1083         memcpy(mkfs_cfg.fs_uuid, fs_uuid, sizeof(mkfs_cfg.fs_uuid));
1084         mkfs_cfg.num_bytes = dev_block_count;
1085         mkfs_cfg.nodesize = nodesize;
1086         mkfs_cfg.sectorsize = sectorsize;
1087         mkfs_cfg.stripesize = stripesize;
1088         mkfs_cfg.features = features;
1089
1090         ret = make_btrfs(fd, &mkfs_cfg);
1091         if (ret) {
1092                 error("error during mkfs: %s", strerror(-ret));
1093                 goto error;
1094         }
1095
1096         fs_info = open_ctree_fs_info(file, 0, 0, 0,
1097                         OPEN_CTREE_WRITES | OPEN_CTREE_FS_PARTIAL);
1098         if (!fs_info) {
1099                 error("open ctree failed");
1100                 goto error;
1101         }
1102         close(fd);
1103         fd = -1;
1104         root = fs_info->fs_root;
1105         fs_info->alloc_start = alloc_start;
1106
1107         ret = create_metadata_block_groups(root, mixed, &allocation);
1108         if (ret) {
1109                 error("failed to create default block groups: %d", ret);
1110                 goto error;
1111         }
1112
1113         trans = btrfs_start_transaction(root, 1);
1114         if (IS_ERR(trans)) {
1115                 error("failed to start transaction");
1116                 goto error;
1117         }
1118
1119         ret = create_data_block_groups(trans, root, mixed, &allocation);
1120         if (ret) {
1121                 error("failed to create default data block groups: %d", ret);
1122                 goto error;
1123         }
1124
1125         ret = make_root_dir(trans, root);
1126         if (ret) {
1127                 error("failed to setup the root directory: %d", ret);
1128                 goto error;
1129         }
1130
1131         ret = btrfs_commit_transaction(trans, root);
1132         if (ret) {
1133                 error("unable to commit transaction: %d", ret);
1134                 goto out;
1135         }
1136
1137         trans = btrfs_start_transaction(root, 1);
1138         if (IS_ERR(trans)) {
1139                 error("failed to start transaction");
1140                 goto error;
1141         }
1142
1143         if (dev_cnt == 0)
1144                 goto raid_groups;
1145
1146         while (dev_cnt-- > 0) {
1147                 file = argv[optind++];
1148
1149                 /*
1150                  * open without O_EXCL so that the problem should not
1151                  * occur by the following processing.
1152                  * (btrfs_register_one_device() fails if O_EXCL is on)
1153                  */
1154                 fd = open(file, O_RDWR);
1155                 if (fd < 0) {
1156                         error("unable to open %s: %m", file);
1157                         goto error;
1158                 }
1159                 ret = btrfs_device_already_in_root(root, fd,
1160                                                    BTRFS_SUPER_INFO_OFFSET);
1161                 if (ret) {
1162                         error("skipping duplicate device %s in the filesystem",
1163                                 file);
1164                         close(fd);
1165                         continue;
1166                 }
1167                 ret = btrfs_prepare_device(fd, file, &dev_block_count,
1168                                 block_count,
1169                                 (verbose ? PREP_DEVICE_VERBOSE : 0) |
1170                                 (zero_end ? PREP_DEVICE_ZERO_END : 0) |
1171                                 (discard ? PREP_DEVICE_DISCARD : 0));
1172                 if (ret) {
1173                         goto error;
1174                 }
1175
1176                 ret = btrfs_add_to_fsid(trans, root, fd, file, dev_block_count,
1177                                         sectorsize, sectorsize, sectorsize);
1178                 if (ret) {
1179                         error("unable to add %s to filesystem: %d", file, ret);
1180                         goto out;
1181                 }
1182                 if (verbose >= 2) {
1183                         struct btrfs_device *device;
1184
1185                         device = container_of(fs_info->fs_devices->devices.next,
1186                                         struct btrfs_device, dev_list);
1187                         printf("adding device %s id %llu\n", file,
1188                                 (unsigned long long)device->devid);
1189                 }
1190         }
1191
1192 raid_groups:
1193         ret = create_raid_groups(trans, root, data_profile,
1194                          metadata_profile, mixed, &allocation);
1195         if (ret) {
1196                 error("unable to create raid groups: %d", ret);
1197                 goto out;
1198         }
1199
1200         ret = create_tree(trans, root, BTRFS_DATA_RELOC_TREE_OBJECTID);
1201         if (ret) {
1202                 error("unable to create data reloc tree: %d", ret);
1203                 goto out;
1204         }
1205
1206         ret = btrfs_commit_transaction(trans, root);
1207         if (ret) {
1208                 error("unable to commit transaction: %d", ret);
1209                 goto out;
1210         }
1211
1212         ret = cleanup_temp_chunks(fs_info, &allocation, data_profile,
1213                                   metadata_profile, metadata_profile);
1214         if (ret < 0) {
1215                 error("failed to cleanup temporary chunks: %d", ret);
1216                 goto out;
1217         }
1218
1219         if (source_dir_set) {
1220                 ret = btrfs_mkfs_fill_dir(source_dir, root, verbose);
1221                 if (ret) {
1222                         error("error wihle filling filesystem: %d", ret);
1223                         goto out;
1224                 }
1225                 if (shrink_rootdir) {
1226                         ret = btrfs_mkfs_shrink_fs(fs_info, &shrink_size,
1227                                                    shrink_rootdir);
1228                         if (ret < 0) {
1229                                 error("error while shrinking filesystem: %d",
1230                                         ret);
1231                                 goto out;
1232                         }
1233                 }
1234         }
1235
1236         if (verbose) {
1237                 char features_buf[64];
1238
1239                 update_chunk_allocation(fs_info, &allocation);
1240                 printf("Label:              %s\n", label);
1241                 printf("UUID:               %s\n", mkfs_cfg.fs_uuid);
1242                 printf("Node size:          %u\n", nodesize);
1243                 printf("Sector size:        %u\n", sectorsize);
1244                 printf("Filesystem size:    %s\n",
1245                         pretty_size(btrfs_super_total_bytes(fs_info->super_copy)));
1246                 printf("Block group profiles:\n");
1247                 if (allocation.data)
1248                         printf("  Data:             %-8s %16s\n",
1249                                 btrfs_group_profile_str(data_profile),
1250                                 pretty_size(allocation.data));
1251                 if (allocation.metadata)
1252                         printf("  Metadata:         %-8s %16s\n",
1253                                 btrfs_group_profile_str(metadata_profile),
1254                                 pretty_size(allocation.metadata));
1255                 if (allocation.mixed)
1256                         printf("  Data+Metadata:    %-8s %16s\n",
1257                                 btrfs_group_profile_str(data_profile),
1258                                 pretty_size(allocation.mixed));
1259                 printf("  System:           %-8s %16s\n",
1260                         btrfs_group_profile_str(metadata_profile),
1261                         pretty_size(allocation.system));
1262                 printf("SSD detected:       %s\n", ssd ? "yes" : "no");
1263                 btrfs_parse_features_to_string(features_buf, features);
1264                 printf("Incompat features:  %s", features_buf);
1265                 printf("\n");
1266
1267                 list_all_devices(root);
1268         }
1269
1270         /*
1271          * The filesystem is now fully set up, commit the remaining changes and
1272          * fix the signature as the last step before closing the devices.
1273          */
1274         fs_info->finalize_on_close = 1;
1275 out:
1276         close_ret = close_ctree(root);
1277
1278         if (!close_ret) {
1279                 optind = saved_optind;
1280                 dev_cnt = argc - optind;
1281                 while (dev_cnt-- > 0) {
1282                         file = argv[optind++];
1283                         if (is_block_device(file) == 1)
1284                                 btrfs_register_one_device(file);
1285                 }
1286         }
1287
1288         btrfs_close_all_devices();
1289         free(label);
1290
1291         return !!ret;
1292 error:
1293         if (fd > 0)
1294                 close(fd);
1295
1296         free(label);
1297         exit(1);
1298 success:
1299         exit(0);
1300 }