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