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 create_chunks(struct btrfs_trans_handle *trans,
 404                          struct btrfs_root *root, u64 num_of_meta_chunks,
 405                          u64 size_of_data,
 406                          struct mkfs_allocation *allocation)
 407 {
 408         struct btrfs_fs_info *fs_info = root->fs_info;
 409         u64 chunk_start;
 410         u64 chunk_size;
 411         u64 meta_type = BTRFS_BLOCK_GROUP_METADATA;
 412         u64 data_type = BTRFS_BLOCK_GROUP_DATA;
 413         u64 minimum_data_chunk_size = SZ_8M;
 414         u64 i;
 415         int ret;
 416
 417         for (i = 0; i < num_of_meta_chunks; i++) {
 418                 ret = btrfs_alloc_chunk(trans, fs_info,
 419                                         &chunk_start, &chunk_size, meta_type);
 420                 if (ret)
 421                         return ret;
 422                 ret = btrfs_make_block_group(trans, fs_info, 0,
 423                                              meta_type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 424                                              chunk_start, chunk_size);
 425                 allocation->metadata += chunk_size;
 426                 if (ret)
 427                         return ret;
 428                 set_extent_dirty(&root->fs_info->free_space_cache,
 429                                  chunk_start, chunk_start + chunk_size - 1);
 430         }
 431
 432         if (size_of_data < minimum_data_chunk_size)
 433                 size_of_data = minimum_data_chunk_size;
 434
 435         ret = btrfs_alloc_data_chunk(trans, fs_info,
 436                                      &chunk_start, size_of_data, data_type, 0);
 437         if (ret)
 438                 return ret;
 439         ret = btrfs_make_block_group(trans, fs_info, 0,
 440                                      data_type, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 441                                      chunk_start, size_of_data);
 442         allocation->data += size_of_data;
 443         if (ret)
 444                 return ret;
 445         set_extent_dirty(&root->fs_info->free_space_cache,
 446                          chunk_start, chunk_start + size_of_data - 1);
 447         return ret;
 448 }
 449
 450 static int zero_output_file(int out_fd, u64 size)
 451 {
 452         int loop_num;
 453         u64 location = 0;
 454         char buf[SZ_4K];
 455         int ret = 0, i;
 456         ssize_t written;
 457
 458         memset(buf, 0, SZ_4K);
 459
 460         /* Only zero out the first 1M */
 461         loop_num = SZ_1M / SZ_4K;
 462         for (i = 0; i < loop_num; i++) {
 463                 written = pwrite64(out_fd, buf, SZ_4K, location);
 464                 if (written != SZ_4K)
 465                         ret = -EIO;
 466                 location += SZ_4K;
 467         }
 468
 469         /* Then enlarge the file to size */
 470         written = pwrite64(out_fd, buf, 1, size - 1);
 471         if (written < 1)
 472                 ret = -EIO;
 473         return ret;
 474 }
 475
 476 static int is_ssd(const char *file)
 477 {
 478         blkid_probe probe;
 479         char wholedisk[PATH_MAX];
 480         char sysfs_path[PATH_MAX];
 481         dev_t devno;
 482         int fd;
 483         char rotational;
 484         int ret;
 485
 486         probe = blkid_new_probe_from_filename(file);
 487         if (!probe)
 488                 return 0;
 489
 490         /* Device number of this disk (possibly a partition) */
 491         devno = blkid_probe_get_devno(probe);
 492         if (!devno) {
 493                 blkid_free_probe(probe);
 494                 return 0;
 495         }
 496
 497         /* Get whole disk name (not full path) for this devno */
 498         ret = blkid_devno_to_wholedisk(devno,
 499                         wholedisk, sizeof(wholedisk), NULL);
 500         if (ret) {
 501                 blkid_free_probe(probe);
 502                 return 0;
 503         }
 504
 505         snprintf(sysfs_path, PATH_MAX, "/sys/block/%s/queue/rotational",
 506                  wholedisk);
 507
 508         blkid_free_probe(probe);
 509
 510         fd = open(sysfs_path, O_RDONLY);
 511         if (fd < 0) {
 512                 return 0;
 513         }
 514
 515         if (read(fd, &rotational, 1) < 1) {
 516                 close(fd);
 517                 return 0;
 518         }
 519         close(fd);
 520
 521         return rotational == '0';
 522 }
 523
 524 static int _cmp_device_by_id(void *priv, struct list_head *a,
 525                              struct list_head *b)
 526 {
 527         return list_entry(a, struct btrfs_device, dev_list)->devid -
 528                list_entry(b, struct btrfs_device, dev_list)->devid;
 529 }
 530
 531 static void list_all_devices(struct btrfs_root *root)
 532 {
 533         struct btrfs_fs_devices *fs_devices;
 534         struct btrfs_device *device;
 535         int number_of_devices = 0;
 536         u64 total_block_count = 0;
 537
 538         fs_devices = root->fs_info->fs_devices;
 539
 540         list_for_each_entry(device, &fs_devices->devices, dev_list)
 541                 number_of_devices++;
 542
 543         list_sort(NULL, &fs_devices->devices, _cmp_device_by_id);
 544
 545         printf("Number of devices:  %d\n", number_of_devices);
 546         /* printf("Total devices size: %10s\n", */
 547                 /* pretty_size(total_block_count)); */
 548         printf("Devices:\n");
 549         printf("   ID        SIZE  PATH\n");
 550         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 551                 printf("  %3llu  %10s  %s\n",
 552                         device->devid,
 553                         pretty_size(device->total_bytes),
 554                         device->name);
 555                 total_block_count += device->total_bytes;
 556         }
 557
 558         printf("\n");
 559 }
 560
 561 static int is_temp_block_group(struct extent_buffer *node,
 562                                struct btrfs_block_group_item *bgi,
 563                                u64 data_profile, u64 meta_profile,
 564                                u64 sys_profile)
 565 {
 566         u64 flag = btrfs_disk_block_group_flags(node, bgi);
 567         u64 flag_type = flag & BTRFS_BLOCK_GROUP_TYPE_MASK;
 568         u64 flag_profile = flag & BTRFS_BLOCK_GROUP_PROFILE_MASK;
 569         u64 used = btrfs_disk_block_group_used(node, bgi);
 570
 571         /*
 572          * Chunks meets all the following conditions is a temp chunk
 573          * 1) Empty chunk
 574          * Temp chunk is always empty.
 575          *
 576          * 2) profile mismatch with mkfs profile.
 577          * Temp chunk is always in SINGLE
 578          *
 579          * 3) Size differs with mkfs_alloc
 580          * Special case for SINGLE/SINGLE btrfs.
 581          * In that case, temp data chunk and real data chunk are always empty.
 582          * So we need to use mkfs_alloc to be sure which chunk is the newly
 583          * allocated.
 584          *
 585          * Normally, new chunk size is equal to mkfs one (One chunk)
 586          * If it has multiple chunks, we just refuse to delete any one.
 587          * As they are all single, so no real problem will happen.
 588          * So only use condition 1) and 2) to judge them.
 589          */
 590         if (used != 0)
 591                 return 0;
 592         switch (flag_type) {
 593         case BTRFS_BLOCK_GROUP_DATA:
 594         case BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA:
 595                 data_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
 596                 if (flag_profile != data_profile)
 597                         return 1;
 598                 break;
 599         case BTRFS_BLOCK_GROUP_METADATA:
 600                 meta_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
 601                 if (flag_profile != meta_profile)
 602                         return 1;
 603                 break;
 604         case BTRFS_BLOCK_GROUP_SYSTEM:
 605                 sys_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
 606                 if (flag_profile != sys_profile)
 607                         return 1;
 608                 break;
 609         }
 610         return 0;
 611 }
 612
 613 /* Note: if current is a block group, it will skip it anyway */
 614 static int next_block_group(struct btrfs_root *root,
 615                             struct btrfs_path *path)
 616 {
 617         struct btrfs_key key;
 618         int ret = 0;
 619
 620         while (1) {
 621                 ret = btrfs_next_item(root, path);
 622                 if (ret)
 623                         goto out;
 624
 625                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 626                 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
 627                         goto out;
 628         }
 629 out:
 630         return ret;
 631 }
 632
 633 /* This function will cleanup  */
 634 static int cleanup_temp_chunks(struct btrfs_fs_info *fs_info,
 635                                struct mkfs_allocation *alloc,
 636                                u64 data_profile, u64 meta_profile,
 637                                u64 sys_profile)
 638 {
 639         struct btrfs_trans_handle *trans = NULL;
 640         struct btrfs_block_group_item *bgi;
 641         struct btrfs_root *root = fs_info->extent_root;
 642         struct btrfs_key key;
 643         struct btrfs_key found_key;
 644         struct btrfs_path path;
 645         int ret = 0;
 646
 647         btrfs_init_path(&path);
 648         trans = btrfs_start_transaction(root, 1);
 649         BUG_ON(IS_ERR(trans));
 650
 651         key.objectid = 0;
 652         key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
 653         key.offset = 0;
 654
 655         while (1) {
 656                 /*
 657                  * as the rest of the loop may modify the tree, we need to
 658                  * start a new search each time.
 659                  */
 660                 ret = btrfs_search_slot(trans, root, &key, &path, 0, 0);
 661                 if (ret < 0)
 662                         goto out;
 663                 /* Don't pollute ret for >0 case */
 664                 if (ret > 0)
 665                         ret = 0;
 666
 667                 btrfs_item_key_to_cpu(path.nodes[0], &found_key,
 668                                       path.slots[0]);
 669                 if (found_key.objectid < key.objectid)
 670                         goto out;
 671                 if (found_key.type != BTRFS_BLOCK_GROUP_ITEM_KEY) {
 672                         ret = next_block_group(root, &path);
 673                         if (ret < 0)
 674                                 goto out;
 675                         if (ret > 0) {
 676                                 ret = 0;
 677                                 goto out;
 678                         }
 679                         btrfs_item_key_to_cpu(path.nodes[0], &found_key,
 680                                               path.slots[0]);
 681                 }
 682
 683                 bgi = btrfs_item_ptr(path.nodes[0], path.slots[0],
 684                                      struct btrfs_block_group_item);
 685                 if (is_temp_block_group(path.nodes[0], bgi,
 686                                         data_profile, meta_profile,
 687                                         sys_profile)) {
 688                         u64 flags = btrfs_disk_block_group_flags(path.nodes[0],
 689                                                              bgi);
 690
 691                         ret = btrfs_free_block_group(trans, fs_info,
 692                                         found_key.objectid, found_key.offset);
 693                         if (ret < 0)
 694                                 goto out;
 695
 696                         if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
 697                             BTRFS_BLOCK_GROUP_DATA)
 698                                 alloc->data -= found_key.offset;
 699                         else if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
 700                                  BTRFS_BLOCK_GROUP_METADATA)
 701                                 alloc->metadata -= found_key.offset;
 702                         else if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
 703                                  BTRFS_BLOCK_GROUP_SYSTEM)
 704                                 alloc->system -= found_key.offset;
 705                         else if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) ==
 706                                  (BTRFS_BLOCK_GROUP_METADATA |
 707                                   BTRFS_BLOCK_GROUP_DATA))
 708                                 alloc->mixed -= found_key.offset;
 709                 }
 710                 btrfs_release_path(&path);
 711                 key.objectid = found_key.objectid + found_key.offset;
 712         }
 713 out:
 714         if (trans)
 715                 btrfs_commit_transaction(trans, root);
 716         btrfs_release_path(&path);
 717         return ret;
 718 }
 719
 720 int main(int argc, char **argv)
 721 {
 722         char *file;
 723         struct btrfs_root *root;
 724         struct btrfs_fs_info *fs_info;
 725         struct btrfs_trans_handle *trans;
 726         char *label = NULL;
 727         u64 block_count = 0;
 728         u64 dev_block_count = 0;
 729         u64 alloc_start = 0;
 730         u64 metadata_profile = 0;
 731         u64 data_profile = 0;
 732         u32 nodesize = max_t(u32, sysconf(_SC_PAGESIZE),
 733                         BTRFS_MKFS_DEFAULT_NODE_SIZE);
 734         u32 sectorsize = 4096;
 735         u32 stripesize = 4096;
 736         int zero_end = 1;
 737         int fd = -1;
 738         int ret;
 739         int close_ret;
 740         int i;
 741         int mixed = 0;
 742         int nodesize_forced = 0;
 743         int data_profile_opt = 0;
 744         int metadata_profile_opt = 0;
 745         int discard = 1;
 746         int ssd = 0;
 747         int force_overwrite = 0;
 748         char *source_dir = NULL;
 749         int source_dir_set = 0;
 750         u64 num_of_meta_chunks = 0;
 751         u64 size_of_data = 0;
 752         u64 source_dir_size = 0;
 753         u64 min_dev_size;
 754         int dev_cnt = 0;
 755         int saved_optind;
 756         char fs_uuid[BTRFS_UUID_UNPARSED_SIZE] = { 0 };
 757         u64 features = BTRFS_MKFS_DEFAULT_FEATURES;
 758         struct mkfs_allocation allocation = { 0 };
 759         struct btrfs_mkfs_config mkfs_cfg;
 760
 761         while(1) {
 762                 int c;
 763                 static const struct option long_options[] = {
 764                         { "alloc-start", required_argument, NULL, 'A'},
 765                         { "byte-count", required_argument, NULL, 'b' },
 766                         { "force", no_argument, NULL, 'f' },
 767                         { "leafsize", required_argument, NULL, 'l' },
 768                         { "label", required_argument, NULL, 'L'},
 769                         { "metadata", required_argument, NULL, 'm' },
 770                         { "mixed", no_argument, NULL, 'M' },
 771                         { "nodesize", required_argument, NULL, 'n' },
 772                         { "sectorsize", required_argument, NULL, 's' },
 773                         { "data", required_argument, NULL, 'd' },
 774                         { "version", no_argument, NULL, 'V' },
 775                         { "rootdir", required_argument, NULL, 'r' },
 776                         { "nodiscard", no_argument, NULL, 'K' },
 777                         { "features", required_argument, NULL, 'O' },
 778                         { "uuid", required_argument, NULL, 'U' },
 779                         { "quiet", 0, NULL, 'q' },
 780                         { "help", no_argument, NULL, GETOPT_VAL_HELP },
 781                         { NULL, 0, NULL, 0}
 782                 };
 783
 784                 c = getopt_long(argc, argv, "A:b:fl:n:s:m:d:L:O:r:U:VMKq",
 785                                 long_options, NULL);
 786                 if (c < 0)
 787                         break;
 788                 switch(c) {
 789                         case 'A':
 790                                 alloc_start = parse_size(optarg);
 791                                 break;
 792                         case 'f':
 793                                 force_overwrite = 1;
 794                                 break;
 795                         case 'd':
 796                                 data_profile = parse_profile(optarg);
 797                                 data_profile_opt = 1;
 798                                 break;
 799                         case 'l':
 800                                 warning("--leafsize is deprecated, use --nodesize");
 801                                 /* fall through */
 802                         case 'n':
 803                                 nodesize = parse_size(optarg);
 804                                 nodesize_forced = 1;
 805                                 break;
 806                         case 'L':
 807                                 label = parse_label(optarg);
 808                                 break;
 809                         case 'm':
 810                                 metadata_profile = parse_profile(optarg);
 811                                 metadata_profile_opt = 1;
 812                                 break;
 813                         case 'M':
 814                                 mixed = 1;
 815                                 break;
 816                         case 'O': {
 817                                 char *orig = strdup(optarg);
 818                                 char *tmp = orig;
 819
 820                                 tmp = btrfs_parse_fs_features(tmp, &features);
 821                                 if (tmp) {
 822                                         error("unrecognized filesystem feature '%s'",
 823                                                         tmp);
 824                                         free(orig);
 825                                         goto error;
 826                                 }
 827                                 free(orig);
 828                                 if (features & BTRFS_FEATURE_LIST_ALL) {
 829                                         btrfs_list_all_fs_features(0);
 830                                         goto success;
 831                                 }
 832                                 break;
 833                                 }
 834                         case 's':
 835                                 sectorsize = parse_size(optarg);
 836                                 break;
 837                         case 'b':
 838                                 block_count = parse_size(optarg);
 839                                 zero_end = 0;
 840                                 break;
 841                         case 'V':
 842                                 printf("mkfs.btrfs, part of %s\n",
 843                                                 PACKAGE_STRING);
 844                                 goto success;
 845                         case 'r':
 846                                 source_dir = optarg;
 847                                 source_dir_set = 1;
 848                                 break;
 849                         case 'U':
 850                                 strncpy(fs_uuid, optarg,
 851                                         BTRFS_UUID_UNPARSED_SIZE - 1);
 852                                 break;
 853                         case 'K':
 854                                 discard = 0;
 855                                 break;
 856                         case 'q':
 857                                 verbose = 0;
 858                                 break;
 859                         case GETOPT_VAL_HELP:
 860                         default:
 861                                 print_usage(c != GETOPT_VAL_HELP);
 862                 }
 863         }
 864
 865         if (verbose) {
 866                 printf("%s\n", PACKAGE_STRING);
 867                 printf("See %s for more information.\n\n", PACKAGE_URL);
 868         }
 869
 870         sectorsize = max(sectorsize, (u32)sysconf(_SC_PAGESIZE));
 871         stripesize = sectorsize;
 872         saved_optind = optind;
 873         dev_cnt = argc - optind;
 874         if (dev_cnt == 0)
 875                 print_usage(1);
 876
 877         if (source_dir_set && dev_cnt > 1) {
 878                 error("the option -r is limited to a single device");
 879                 goto error;
 880         }
 881
 882         if (*fs_uuid) {
 883                 uuid_t dummy_uuid;
 884
 885                 if (uuid_parse(fs_uuid, dummy_uuid) != 0) {
 886                         error("could not parse UUID: %s", fs_uuid);
 887                         goto error;
 888                 }
 889                 if (!test_uuid_unique(fs_uuid)) {
 890                         error("non-unique UUID: %s", fs_uuid);
 891                         goto error;
 892                 }
 893         }
 894
 895         while (dev_cnt-- > 0) {
 896                 file = argv[optind++];
 897                 if (is_block_device(file) == 1)
 898                         ret = test_dev_for_mkfs(file, force_overwrite);
 899                 else
 900                         ret = test_status_for_mkfs(file, force_overwrite);
 901
 902                 if (ret)
 903                         goto error;
 904         }
 905
 906         optind = saved_optind;
 907         dev_cnt = argc - optind;
 908
 909         file = argv[optind++];
 910         ssd = is_ssd(file);
 911
 912         /*
 913         * Set default profiles according to number of added devices.
 914         * For mixed groups defaults are single/single.
 915         */
 916         if (!mixed) {
 917                 if (!metadata_profile_opt) {
 918                         if (dev_cnt == 1 && ssd && verbose)
 919                                 printf("Detected a SSD, turning off metadata "
 920                                 "duplication.  Mkfs with -m dup if you want to "
 921                                 "force metadata duplication.\n");
 922
 923                         metadata_profile = (dev_cnt > 1) ?
 924                                         BTRFS_BLOCK_GROUP_RAID1 : (ssd) ?
 925                                         0: BTRFS_BLOCK_GROUP_DUP;
 926                 }
 927                 if (!data_profile_opt) {
 928                         data_profile = (dev_cnt > 1) ?
 929                                 BTRFS_BLOCK_GROUP_RAID0 : 0; /* raid0 or single */
 930                 }
 931         } else {
 932                 u32 best_nodesize = max_t(u32, sysconf(_SC_PAGESIZE), sectorsize);
 933
 934                 if (metadata_profile_opt || data_profile_opt) {
 935                         if (metadata_profile != data_profile) {
 936                                 error(
 937         "with mixed block groups data and metadata profiles must be the same");
 938                                 goto error;
 939                         }
 940                 }
 941
 942                 if (!nodesize_forced)
 943                         nodesize = best_nodesize;
 944         }
 945
 946         /*
 947          * FS features that can be set by other means than -O
 948          * just set the bit here
 949          */
 950         if (mixed)
 951                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS;
 952
 953         if ((data_profile | metadata_profile) &
 954             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
 955                 features |= BTRFS_FEATURE_INCOMPAT_RAID56;
 956         }
 957
 958         if (btrfs_check_nodesize(nodesize, sectorsize,
 959                                  features))
 960                 goto error;
 961
 962         if (sectorsize < sizeof(struct btrfs_super_block)) {
 963                 error("sectorsize smaller than superblock: %u < %zu",
 964                                 sectorsize, sizeof(struct btrfs_super_block));
 965                 goto error;
 966         }
 967
 968         min_dev_size = btrfs_min_dev_size(nodesize, mixed, metadata_profile,
 969                                           data_profile);
 970         /* Check device/block_count after the nodesize is determined */
 971         if (block_count && block_count < min_dev_size) {
 972                 error("size %llu is too small to make a usable filesystem",
 973                         block_count);
 974                 error("minimum size for btrfs filesystem is %llu",
 975                         min_dev_size);
 976                 goto error;
 977         }
 978         for (i = saved_optind; i < saved_optind + dev_cnt; i++) {
 979                 char *path;
 980
 981                 path = argv[i];
 982                 ret = test_minimum_size(path, min_dev_size);
 983                 if (ret < 0) {
 984                         error("failed to check size for %s: %s",
 985                                 path, strerror(-ret));
 986                         goto error;
 987                 }
 988                 if (ret > 0) {
 989                         error("'%s' is too small to make a usable filesystem",
 990                                 path);
 991                         error("minimum size for each btrfs device is %llu",
 992                                 min_dev_size);
 993                         goto error;
 994                 }
 995         }
 996         ret = test_num_disk_vs_raid(metadata_profile, data_profile,
 997                         dev_cnt, mixed, ssd);
 998         if (ret)
 999                 goto error;
1000
1001         dev_cnt--;
1002
1003         if (!source_dir_set) {
1004                 /*
1005                  * open without O_EXCL so that the problem should not
1006                  * occur by the following processing.
1007                  * (btrfs_register_one_device() fails if O_EXCL is on)
1008                  */
1009                 fd = open(file, O_RDWR);
1010                 if (fd < 0) {
1011                         error("unable to open %s: %s", file, strerror(errno));
1012                         goto error;
1013                 }
1014                 ret = btrfs_prepare_device(fd, file, &dev_block_count,
1015                                 block_count,
1016                                 (zero_end ? PREP_DEVICE_ZERO_END : 0) |
1017                                 (discard ? PREP_DEVICE_DISCARD : 0) |
1018                                 (verbose ? PREP_DEVICE_VERBOSE : 0));
1019                 if (ret) {
1020                         goto error;
1021                 }
1022                 if (block_count && block_count > dev_block_count) {
1023                         error("%s is smaller than requested size, expected %llu, found %llu",
1024                                         file,
1025                                         (unsigned long long)block_count,
1026                                         (unsigned long long)dev_block_count);
1027                         goto error;
1028                 }
1029         } else {
1030                 fd = open(file, O_CREAT | O_RDWR,
1031                                 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH);
1032                 if (fd < 0) {
1033                         error("unable to open %s: %s", file, strerror(errno));
1034                         goto error;
1035                 }
1036
1037                 source_dir_size = btrfs_mkfs_size_dir(source_dir, sectorsize,
1038                                         &num_of_meta_chunks, &size_of_data);
1039                 if(block_count < source_dir_size)
1040                         block_count = source_dir_size;
1041                 ret = zero_output_file(fd, block_count);
1042                 if (ret) {
1043                         error("unable to zero the output file");
1044                         goto error;
1045                 }
1046                 /* our "device" is the new image file */
1047                 dev_block_count = block_count;
1048         }
1049
1050         /* To create the first block group and chunk 0 in make_btrfs */
1051         if (dev_block_count < BTRFS_MKFS_SYSTEM_GROUP_SIZE) {
1052                 error("device is too small to make filesystem, must be at least %llu",
1053                                 (unsigned long long)BTRFS_MKFS_SYSTEM_GROUP_SIZE);
1054                 goto error;
1055         }
1056
1057         if (group_profile_max_safe_loss(metadata_profile) <
1058                 group_profile_max_safe_loss(data_profile)){
1059                 warning("metadata has lower redundancy than data!\n");
1060         }
1061
1062         mkfs_cfg.label = label;
1063         memcpy(mkfs_cfg.fs_uuid, fs_uuid, sizeof(mkfs_cfg.fs_uuid));
1064         mkfs_cfg.num_bytes = dev_block_count;
1065         mkfs_cfg.nodesize = nodesize;
1066         mkfs_cfg.sectorsize = sectorsize;
1067         mkfs_cfg.stripesize = stripesize;
1068         mkfs_cfg.features = features;
1069
1070         ret = make_btrfs(fd, &mkfs_cfg);
1071         if (ret) {
1072                 error("error during mkfs: %s", strerror(-ret));
1073                 goto error;
1074         }
1075
1076         fs_info = open_ctree_fs_info(file, 0, 0, 0,
1077                         OPEN_CTREE_WRITES | OPEN_CTREE_FS_PARTIAL);
1078         if (!fs_info) {
1079                 error("open ctree failed");
1080                 goto error;
1081         }
1082         close(fd);
1083         fd = -1;
1084         root = fs_info->fs_root;
1085         fs_info->alloc_start = alloc_start;
1086
1087         ret = create_metadata_block_groups(root, mixed, &allocation);
1088         if (ret) {
1089                 error("failed to create default block groups: %d", ret);
1090                 goto error;
1091         }
1092
1093         trans = btrfs_start_transaction(root, 1);
1094         if (IS_ERR(trans)) {
1095                 error("failed to start transaction");
1096                 goto error;
1097         }
1098
1099         ret = create_data_block_groups(trans, root, mixed, &allocation);
1100         if (ret) {
1101                 error("failed to create default data block groups: %d", ret);
1102                 goto error;
1103         }
1104
1105         ret = make_root_dir(trans, root);
1106         if (ret) {
1107                 error("failed to setup the root directory: %d", ret);
1108                 goto error;
1109         }
1110
1111         ret = btrfs_commit_transaction(trans, root);
1112         if (ret) {
1113                 error("unable to commit transaction: %d", ret);
1114                 goto out;
1115         }
1116
1117         trans = btrfs_start_transaction(root, 1);
1118         if (IS_ERR(trans)) {
1119                 error("failed to start transaction");
1120                 goto error;
1121         }
1122
1123         if (dev_cnt == 0)
1124                 goto raid_groups;
1125
1126         while (dev_cnt-- > 0) {
1127                 file = argv[optind++];
1128
1129                 /*
1130                  * open without O_EXCL so that the problem should not
1131                  * occur by the following processing.
1132                  * (btrfs_register_one_device() fails if O_EXCL is on)
1133                  */
1134                 fd = open(file, O_RDWR);
1135                 if (fd < 0) {
1136                         error("unable to open %s: %s", file, strerror(errno));
1137                         goto error;
1138                 }
1139                 ret = btrfs_device_already_in_root(root, fd,
1140                                                    BTRFS_SUPER_INFO_OFFSET);
1141                 if (ret) {
1142                         error("skipping duplicate device %s in the filesystem",
1143                                 file);
1144                         close(fd);
1145                         continue;
1146                 }
1147                 ret = btrfs_prepare_device(fd, file, &dev_block_count,
1148                                 block_count,
1149                                 (verbose ? PREP_DEVICE_VERBOSE : 0) |
1150                                 (zero_end ? PREP_DEVICE_ZERO_END : 0) |
1151                                 (discard ? PREP_DEVICE_DISCARD : 0));
1152                 if (ret) {
1153                         goto error;
1154                 }
1155
1156                 ret = btrfs_add_to_fsid(trans, root, fd, file, dev_block_count,
1157                                         sectorsize, sectorsize, sectorsize);
1158                 if (ret) {
1159                         error("unable to add %s to filesystem: %d", file, ret);
1160                         goto out;
1161                 }
1162                 if (verbose >= 2) {
1163                         struct btrfs_device *device;
1164
1165                         device = container_of(fs_info->fs_devices->devices.next,
1166                                         struct btrfs_device, dev_list);
1167                         printf("adding device %s id %llu\n", file,
1168                                 (unsigned long long)device->devid);
1169                 }
1170         }
1171
1172 raid_groups:
1173         if (!source_dir_set) {
1174                 ret = create_raid_groups(trans, root, data_profile,
1175                                  metadata_profile, mixed, &allocation);
1176                 if (ret) {
1177                         error("unable to create raid groups: %d", ret);
1178                         goto out;
1179                 }
1180         }
1181
1182         ret = create_tree(trans, root, BTRFS_DATA_RELOC_TREE_OBJECTID);
1183         if (ret) {
1184                 error("unable to create data reloc tree: %d", ret);
1185                 goto out;
1186         }
1187
1188         ret = btrfs_commit_transaction(trans, root);
1189         if (ret) {
1190                 error("unable to commit transaction: %d", ret);
1191                 goto out;
1192         }
1193
1194         if (source_dir_set) {
1195                 trans = btrfs_start_transaction(root, 1);
1196                 BUG_ON(IS_ERR(trans));
1197                 ret = create_chunks(trans, root,
1198                                     num_of_meta_chunks, size_of_data,
1199                                     &allocation);
1200                 if (ret) {
1201                         error("unable to create chunks: %d", ret);
1202                         goto out;
1203                 }
1204                 ret = btrfs_commit_transaction(trans, root);
1205                 if (ret) {
1206                         error("transaction commit failed: %d", ret);
1207                         goto out;
1208                 }
1209
1210                 ret = btrfs_mkfs_fill_dir(source_dir, root, verbose);
1211                 if (ret) {
1212                         error("error wihle filling filesystem: %d", ret);
1213                         goto out;
1214                 }
1215         }
1216         ret = cleanup_temp_chunks(fs_info, &allocation, data_profile,
1217                                   metadata_profile, metadata_profile);
1218         if (ret < 0) {
1219                 error("failed to cleanup temporary chunks: %d", ret);
1220                 goto out;
1221         }
1222
1223         if (verbose) {
1224                 char features_buf[64];
1225
1226                 printf("Label:              %s\n", label);
1227                 printf("UUID:               %s\n", mkfs_cfg.fs_uuid);
1228                 printf("Node size:          %u\n", nodesize);
1229                 printf("Sector size:        %u\n", sectorsize);
1230                 printf("Filesystem size:    %s\n",
1231                         pretty_size(btrfs_super_total_bytes(fs_info->super_copy)));
1232                 printf("Block group profiles:\n");
1233                 if (allocation.data)
1234                         printf("  Data:             %-8s %16s\n",
1235                                 btrfs_group_profile_str(data_profile),
1236                                 pretty_size(allocation.data));
1237                 if (allocation.metadata)
1238                         printf("  Metadata:         %-8s %16s\n",
1239                                 btrfs_group_profile_str(metadata_profile),
1240                                 pretty_size(allocation.metadata));
1241                 if (allocation.mixed)
1242                         printf("  Data+Metadata:    %-8s %16s\n",
1243                                 btrfs_group_profile_str(data_profile),
1244                                 pretty_size(allocation.mixed));
1245                 printf("  System:           %-8s %16s\n",
1246                         btrfs_group_profile_str(metadata_profile),
1247                         pretty_size(allocation.system));
1248                 printf("SSD detected:       %s\n", ssd ? "yes" : "no");
1249                 btrfs_parse_features_to_string(features_buf, features);
1250                 printf("Incompat features:  %s", features_buf);
1251                 printf("\n");
1252
1253                 list_all_devices(root);
1254         }
1255
1256         /*
1257          * The filesystem is now fully set up, commit the remaining changes and
1258          * fix the signature as the last step before closing the devices.
1259          */
1260         fs_info->finalize_on_close = 1;
1261 out:
1262         close_ret = close_ctree(root);
1263
1264         if (!close_ret) {
1265                 optind = saved_optind;
1266                 dev_cnt = argc - optind;
1267                 while (dev_cnt-- > 0) {
1268                         file = argv[optind++];
1269                         if (is_block_device(file) == 1)
1270                                 btrfs_register_one_device(file);
1271                 }
1272         }
1273
1274         btrfs_close_all_devices();
1275         free(label);
1276
1277         return !!ret;
1278 error:
1279         if (fd > 0)
1280                 close(fd);
1281
1282         free(label);
1283         exit(1);
1284 success:
1285         exit(0);
1286 }