mkfs/rootdir.c

   1 /*
   2  * Copyright (C) 2017 SUSE.  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.
  15  */
  16
  17 #include "kerncompat.h"
  18 #include "androidcompat.h"
  19
  20 #include <sys/stat.h>
  21 #include <sys/types.h>
  22 #include <sys/xattr.h>
  23 #include <linux/limits.h>
  24 #include <dirent.h>
  25 #include <unistd.h>
  26 #include <fcntl.h>
  27 #include <ftw.h>
  28 #include "ctree.h"
  29 #include "volumes.h"
  30 #include "internal.h"
  31 #include "disk-io.h"
  32 #include "messages.h"
  33 #include "transaction.h"
  34 #include "utils.h"
  35 #include "mkfs/rootdir.h"
  36 #include "mkfs/common.h"
  37 #include "send-utils.h"
  38
  39 static u32 fs_block_size;
  40
  41 static u64 index_cnt = 2;
  42
  43 /*
  44  * Size estimate will be done using the following data:
  45  * 1) Number of inodes
  46  *    Since we will later shrink the fs, over-estimate is completely fine here
  47  *    as long as our estimate ensures we can populate the image without ENOSPC.
  48  *    So we only record how many inodes there are, and account the maximum
  49  *    space for each inode.
  50  *
  51  * 2) Data space for each (regular) inode
  52  *    To estimate data chunk size.
  53  *    Don't care if it can fit as an inline extent.
  54  *    Always round them up to sectorsize.
  55  */
  56 static u64 ftw_meta_nr_inode;
  57 static u64 ftw_data_size;
  58
  59 static int add_directory_items(struct btrfs_trans_handle *trans,
  60                                struct btrfs_root *root, u64 objectid,
  61                                ino_t parent_inum, const char *name,
  62                                struct stat *st, int *dir_index_cnt)
  63 {
  64         int ret;
  65         int name_len;
  66         struct btrfs_key location;
  67         u8 filetype = 0;
  68
  69         name_len = strlen(name);
  70
  71         location.objectid = objectid;
  72         location.offset = 0;
  73         location.type = BTRFS_INODE_ITEM_KEY;
  74
  75         if (S_ISDIR(st->st_mode))
  76                 filetype = BTRFS_FT_DIR;
  77         if (S_ISREG(st->st_mode))
  78                 filetype = BTRFS_FT_REG_FILE;
  79         if (S_ISLNK(st->st_mode))
  80                 filetype = BTRFS_FT_SYMLINK;
  81         if (S_ISSOCK(st->st_mode))
  82                 filetype = BTRFS_FT_SOCK;
  83         if (S_ISCHR(st->st_mode))
  84                 filetype = BTRFS_FT_CHRDEV;
  85         if (S_ISBLK(st->st_mode))
  86                 filetype = BTRFS_FT_BLKDEV;
  87         if (S_ISFIFO(st->st_mode))
  88                 filetype = BTRFS_FT_FIFO;
  89
  90         ret = btrfs_insert_dir_item(trans, root, name, name_len,
  91                                     parent_inum, &location,
  92                                     filetype, index_cnt);
  93         if (ret)
  94                 return ret;
  95         ret = btrfs_insert_inode_ref(trans, root, name, name_len,
  96                                      objectid, parent_inum, index_cnt);
  97         *dir_index_cnt = index_cnt;
  98         index_cnt++;
  99
 100         return ret;
 101 }
 102
 103 static int fill_inode_item(struct btrfs_trans_handle *trans,
 104                            struct btrfs_root *root,
 105                            struct btrfs_inode_item *dst, struct stat *src)
 106 {
 107         u64 blocks = 0;
 108         u64 sectorsize = root->fs_info->sectorsize;
 109
 110         /*
 111          * btrfs_inode_item has some reserved fields
 112          * and represents on-disk inode entry, so
 113          * zero everything to prevent information leak
 114          */
 115         memset(dst, 0, sizeof(*dst));
 116
 117         btrfs_set_stack_inode_generation(dst, trans->transid);
 118         btrfs_set_stack_inode_size(dst, src->st_size);
 119         btrfs_set_stack_inode_nbytes(dst, 0);
 120         btrfs_set_stack_inode_block_group(dst, 0);
 121         btrfs_set_stack_inode_nlink(dst, src->st_nlink);
 122         btrfs_set_stack_inode_uid(dst, src->st_uid);
 123         btrfs_set_stack_inode_gid(dst, src->st_gid);
 124         btrfs_set_stack_inode_mode(dst, src->st_mode);
 125         btrfs_set_stack_inode_rdev(dst, 0);
 126         btrfs_set_stack_inode_flags(dst, 0);
 127         btrfs_set_stack_timespec_sec(&dst->atime, src->st_atime);
 128         btrfs_set_stack_timespec_nsec(&dst->atime, 0);
 129         btrfs_set_stack_timespec_sec(&dst->ctime, src->st_ctime);
 130         btrfs_set_stack_timespec_nsec(&dst->ctime, 0);
 131         btrfs_set_stack_timespec_sec(&dst->mtime, src->st_mtime);
 132         btrfs_set_stack_timespec_nsec(&dst->mtime, 0);
 133         btrfs_set_stack_timespec_sec(&dst->otime, 0);
 134         btrfs_set_stack_timespec_nsec(&dst->otime, 0);
 135
 136         if (S_ISDIR(src->st_mode)) {
 137                 btrfs_set_stack_inode_size(dst, 0);
 138                 btrfs_set_stack_inode_nlink(dst, 1);
 139         }
 140         if (S_ISREG(src->st_mode)) {
 141                 btrfs_set_stack_inode_size(dst, (u64)src->st_size);
 142                 if (src->st_size <= BTRFS_MAX_INLINE_DATA_SIZE(root->fs_info) &&
 143                     src->st_size < sectorsize)
 144                         btrfs_set_stack_inode_nbytes(dst, src->st_size);
 145                 else {
 146                         blocks = src->st_size / sectorsize;
 147                         if (src->st_size % sectorsize)
 148                                 blocks += 1;
 149                         blocks *= sectorsize;
 150                         btrfs_set_stack_inode_nbytes(dst, blocks);
 151                 }
 152         }
 153         if (S_ISLNK(src->st_mode))
 154                 btrfs_set_stack_inode_nbytes(dst, src->st_size + 1);
 155
 156         return 0;
 157 }
 158
 159 static int directory_select(const struct direct *entry)
 160 {
 161         if (entry->d_name[0] == '.' &&
 162                 (entry->d_name[1] == 0 ||
 163                  (entry->d_name[1] == '.' && entry->d_name[2] == 0)))
 164                 return 0;
 165         return 1;
 166 }
 167
 168 static void free_namelist(struct direct **files, int count)
 169 {
 170         int i;
 171
 172         if (count < 0)
 173                 return;
 174
 175         for (i = 0; i < count; ++i)
 176                 free(files[i]);
 177         free(files);
 178 }
 179
 180 static u64 calculate_dir_inode_size(const char *dirname)
 181 {
 182         int count, i;
 183         struct direct **files, *cur_file;
 184         u64 dir_inode_size = 0;
 185
 186         count = scandir(dirname, &files, directory_select, NULL);
 187
 188         for (i = 0; i < count; i++) {
 189                 cur_file = files[i];
 190                 dir_inode_size += strlen(cur_file->d_name);
 191         }
 192
 193         free_namelist(files, count);
 194
 195         dir_inode_size *= 2;
 196         return dir_inode_size;
 197 }
 198
 199 static int add_inode_items(struct btrfs_trans_handle *trans,
 200                            struct btrfs_root *root,
 201                            struct stat *st, const char *name,
 202                            u64 self_objectid,
 203                            struct btrfs_inode_item *inode_ret)
 204 {
 205         int ret;
 206         struct btrfs_inode_item btrfs_inode;
 207         u64 objectid;
 208         u64 inode_size = 0;
 209
 210         fill_inode_item(trans, root, &btrfs_inode, st);
 211         objectid = self_objectid;
 212
 213         if (S_ISDIR(st->st_mode)) {
 214                 inode_size = calculate_dir_inode_size(name);
 215                 btrfs_set_stack_inode_size(&btrfs_inode, inode_size);
 216         }
 217
 218         ret = btrfs_insert_inode(trans, root, objectid, &btrfs_inode);
 219
 220         *inode_ret = btrfs_inode;
 221         return ret;
 222 }
 223
 224 static int add_xattr_item(struct btrfs_trans_handle *trans,
 225                           struct btrfs_root *root, u64 objectid,
 226                           const char *file_name)
 227 {
 228         int ret;
 229         int cur_name_len;
 230         char xattr_list[XATTR_LIST_MAX];
 231         char *cur_name;
 232         char cur_value[XATTR_SIZE_MAX];
 233         char delimiter = '\0';
 234         char *next_location = xattr_list;
 235
 236         ret = llistxattr(file_name, xattr_list, XATTR_LIST_MAX);
 237         if (ret < 0) {
 238                 if (errno == ENOTSUP)
 239                         return 0;
 240                 error("getting a list of xattr failed for %s: %s", file_name,
 241                                 strerror(errno));
 242                 return ret;
 243         }
 244         if (ret == 0)
 245                 return ret;
 246
 247         cur_name = strtok(xattr_list, &delimiter);
 248         while (cur_name != NULL) {
 249                 cur_name_len = strlen(cur_name);
 250                 next_location += cur_name_len + 1;
 251
 252                 ret = lgetxattr(file_name, cur_name, cur_value, XATTR_SIZE_MAX);
 253                 if (ret < 0) {
 254                         if (errno == ENOTSUP)
 255                                 return 0;
 256                         error("gettig a xattr value failed for %s attr %s: %s",
 257                                 file_name, cur_name, strerror(errno));
 258                         return ret;
 259                 }
 260
 261                 ret = btrfs_insert_xattr_item(trans, root, cur_name,
 262                                               cur_name_len, cur_value,
 263                                               ret, objectid);
 264                 if (ret) {
 265                         error("inserting a xattr item failed for %s: %s",
 266                                         file_name, strerror(-ret));
 267                 }
 268
 269                 cur_name = strtok(next_location, &delimiter);
 270         }
 271
 272         return ret;
 273 }
 274
 275 static int add_symbolic_link(struct btrfs_trans_handle *trans,
 276                              struct btrfs_root *root,
 277                              u64 objectid, const char *path_name)
 278 {
 279         int ret;
 280         char buf[PATH_MAX];
 281
 282         ret = readlink(path_name, buf, sizeof(buf));
 283         if (ret <= 0) {
 284                 error("readlink failed for %s: %s", path_name, strerror(errno));
 285                 goto fail;
 286         }
 287         if (ret >= sizeof(buf)) {
 288                 error("symlink too long for %s", path_name);
 289                 ret = -1;
 290                 goto fail;
 291         }
 292
 293         buf[ret] = '\0'; /* readlink does not do it for us */
 294         ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
 295                                          buf, ret + 1);
 296 fail:
 297         return ret;
 298 }
 299
 300 static int add_file_items(struct btrfs_trans_handle *trans,
 301                           struct btrfs_root *root,
 302                           struct btrfs_inode_item *btrfs_inode, u64 objectid,
 303                           struct stat *st, const char *path_name)
 304 {
 305         int ret = -1;
 306         ssize_t ret_read;
 307         u64 bytes_read = 0;
 308         struct btrfs_key key;
 309         int blocks;
 310         u32 sectorsize = root->fs_info->sectorsize;
 311         u64 first_block = 0;
 312         u64 file_pos = 0;
 313         u64 cur_bytes;
 314         u64 total_bytes;
 315         struct extent_buffer *eb = NULL;
 316         int fd;
 317
 318         if (st->st_size == 0)
 319                 return 0;
 320
 321         fd = open(path_name, O_RDONLY);
 322         if (fd == -1) {
 323                 error("cannot open %s: %s", path_name, strerror(errno));
 324                 return ret;
 325         }
 326
 327         blocks = st->st_size / sectorsize;
 328         if (st->st_size % sectorsize)
 329                 blocks += 1;
 330
 331         if (st->st_size <= BTRFS_MAX_INLINE_DATA_SIZE(root->fs_info) &&
 332             st->st_size < sectorsize) {
 333                 char *buffer = malloc(st->st_size);
 334
 335                 if (!buffer) {
 336                         ret = -ENOMEM;
 337                         goto end;
 338                 }
 339
 340                 ret_read = pread64(fd, buffer, st->st_size, bytes_read);
 341                 if (ret_read == -1) {
 342                         error("cannot read %s at offset %llu length %llu: %s",
 343                                 path_name, (unsigned long long)bytes_read,
 344                                 (unsigned long long)st->st_size,
 345                                 strerror(errno));
 346                         free(buffer);
 347                         goto end;
 348                 }
 349
 350                 ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
 351                                                  buffer, st->st_size);
 352                 free(buffer);
 353                 goto end;
 354         }
 355
 356         /* round up our st_size to the FS blocksize */
 357         total_bytes = (u64)blocks * sectorsize;
 358
 359         /*
 360          * do our IO in extent buffers so it can work
 361          * against any raid type
 362          */
 363         eb = calloc(1, sizeof(*eb) + sectorsize);
 364         if (!eb) {
 365                 ret = -ENOMEM;
 366                 goto end;
 367         }
 368
 369 again:
 370
 371         /*
 372          * keep our extent size at 1MB max, this makes it easier to work inside
 373          * the tiny block groups created during mkfs
 374          */
 375         cur_bytes = min(total_bytes, (u64)SZ_1M);
 376         ret = btrfs_reserve_extent(trans, root, cur_bytes, 0, 0, (u64)-1,
 377                                    &key, 1);
 378         if (ret)
 379                 goto end;
 380
 381         first_block = key.objectid;
 382         bytes_read = 0;
 383
 384         while (bytes_read < cur_bytes) {
 385
 386                 memset(eb->data, 0, sectorsize);
 387
 388                 ret_read = pread64(fd, eb->data, sectorsize, file_pos +
 389                                    bytes_read);
 390                 if (ret_read == -1) {
 391                         error("cannot read %s at offset %llu length %llu: %s",
 392                                 path_name,
 393                                 (unsigned long long)file_pos + bytes_read,
 394                                 (unsigned long long)sectorsize,
 395                                 strerror(errno));
 396                         goto end;
 397                 }
 398
 399                 eb->start = first_block + bytes_read;
 400                 eb->len = sectorsize;
 401
 402                 /*
 403                  * we're doing the csum before we record the extent, but
 404                  * that's ok
 405                  */
 406                 ret = btrfs_csum_file_block(trans, root->fs_info->csum_root,
 407                                 first_block + bytes_read + sectorsize,
 408                                 first_block + bytes_read,
 409                                 eb->data, sectorsize);
 410                 if (ret)
 411                         goto end;
 412
 413                 ret = write_and_map_eb(root->fs_info, eb);
 414                 if (ret) {
 415                         error("failed to write %s", path_name);
 416                         goto end;
 417                 }
 418
 419                 bytes_read += sectorsize;
 420         }
 421
 422         if (bytes_read) {
 423                 ret = btrfs_record_file_extent(trans, root, objectid,
 424                                 btrfs_inode, file_pos, first_block, cur_bytes);
 425                 if (ret)
 426                         goto end;
 427
 428         }
 429
 430         file_pos += cur_bytes;
 431         total_bytes -= cur_bytes;
 432
 433         if (total_bytes)
 434                 goto again;
 435
 436 end:
 437         free(eb);
 438         close(fd);
 439         return ret;
 440 }
 441
 442 static int traverse_directory(struct btrfs_trans_handle *trans,
 443                               struct btrfs_root *root, const char *dir_name,
 444                               struct directory_name_entry *dir_head)
 445 {
 446         int ret = 0;
 447
 448         struct btrfs_inode_item cur_inode;
 449         struct btrfs_inode_item *inode_item;
 450         int count, i, dir_index_cnt;
 451         struct direct **files;
 452         struct stat st;
 453         struct directory_name_entry *dir_entry, *parent_dir_entry;
 454         struct direct *cur_file;
 455         ino_t parent_inum, cur_inum;
 456         ino_t highest_inum = 0;
 457         const char *parent_dir_name;
 458         struct btrfs_path path;
 459         struct extent_buffer *leaf;
 460         struct btrfs_key root_dir_key;
 461         u64 root_dir_inode_size = 0;
 462
 463         /* Add list for source directory */
 464         dir_entry = malloc(sizeof(struct directory_name_entry));
 465         if (!dir_entry)
 466                 return -ENOMEM;
 467         dir_entry->dir_name = dir_name;
 468         dir_entry->path = realpath(dir_name, NULL);
 469         if (!dir_entry->path) {
 470                 error("realpath  failed for %s: %s", dir_name, strerror(errno));
 471                 ret = -1;
 472                 goto fail_no_dir;
 473         }
 474
 475         parent_inum = highest_inum + BTRFS_FIRST_FREE_OBJECTID;
 476         dir_entry->inum = parent_inum;
 477         list_add_tail(&dir_entry->list, &dir_head->list);
 478
 479         btrfs_init_path(&path);
 480
 481         root_dir_key.objectid = btrfs_root_dirid(&root->root_item);
 482         root_dir_key.offset = 0;
 483         root_dir_key.type = BTRFS_INODE_ITEM_KEY;
 484         ret = btrfs_lookup_inode(trans, root, &path, &root_dir_key, 1);
 485         if (ret) {
 486                 error("failed to lookup root dir: %d", ret);
 487                 goto fail_no_dir;
 488         }
 489
 490         leaf = path.nodes[0];
 491         inode_item = btrfs_item_ptr(leaf, path.slots[0],
 492                                     struct btrfs_inode_item);
 493
 494         root_dir_inode_size = calculate_dir_inode_size(dir_name);
 495         btrfs_set_inode_size(leaf, inode_item, root_dir_inode_size);
 496         btrfs_mark_buffer_dirty(leaf);
 497
 498         btrfs_release_path(&path);
 499
 500         do {
 501                 parent_dir_entry = list_entry(dir_head->list.next,
 502                                               struct directory_name_entry,
 503                                               list);
 504                 list_del(&parent_dir_entry->list);
 505
 506                 parent_inum = parent_dir_entry->inum;
 507                 parent_dir_name = parent_dir_entry->dir_name;
 508                 if (chdir(parent_dir_entry->path)) {
 509                         error("chdir failed for %s: %s",
 510                                 parent_dir_name, strerror(errno));
 511                         ret = -1;
 512                         goto fail_no_files;
 513                 }
 514
 515                 count = scandir(parent_dir_entry->path, &files,
 516                                 directory_select, NULL);
 517                 if (count == -1) {
 518                         error("scandir failed for %s: %s",
 519                                 parent_dir_name, strerror(errno));
 520                         ret = -1;
 521                         goto fail;
 522                 }
 523
 524                 for (i = 0; i < count; i++) {
 525                         cur_file = files[i];
 526
 527                         if (lstat(cur_file->d_name, &st) == -1) {
 528                                 error("lstat failed for %s: %s",
 529                                         cur_file->d_name, strerror(errno));
 530                                 ret = -1;
 531                                 goto fail;
 532                         }
 533
 534                         cur_inum = st.st_ino;
 535                         ret = add_directory_items(trans, root,
 536                                                   cur_inum, parent_inum,
 537                                                   cur_file->d_name,
 538                                                   &st, &dir_index_cnt);
 539                         if (ret) {
 540                                 error("unable to add directory items for %s: %d",
 541                                         cur_file->d_name, ret);
 542                                 goto fail;
 543                         }
 544
 545                         ret = add_inode_items(trans, root, &st,
 546                                               cur_file->d_name, cur_inum,
 547                                               &cur_inode);
 548                         if (ret == -EEXIST) {
 549                                 if (st.st_nlink <= 1) {
 550                                         error(
 551                         "item %s already exists but has wrong st_nlink %lu <= 1",
 552                                                 cur_file->d_name,
 553                                                 (unsigned long)st.st_nlink);
 554                                         goto fail;
 555                                 }
 556                                 continue;
 557                         }
 558                         if (ret) {
 559                                 error("unable to add inode items for %s: %d",
 560                                         cur_file->d_name, ret);
 561                                 goto fail;
 562                         }
 563
 564                         ret = add_xattr_item(trans, root,
 565                                              cur_inum, cur_file->d_name);
 566                         if (ret) {
 567                                 error("unable to add xattr items for %s: %d",
 568                                         cur_file->d_name, ret);
 569                                 if (ret != -ENOTSUP)
 570                                         goto fail;
 571                         }
 572
 573                         if (S_ISDIR(st.st_mode)) {
 574                                 char tmp[PATH_MAX];
 575
 576                                 dir_entry = malloc(sizeof(*dir_entry));
 577                                 if (!dir_entry) {
 578                                         ret = -ENOMEM;
 579                                         goto fail;
 580                                 }
 581                                 dir_entry->dir_name = cur_file->d_name;
 582                                 if (path_cat_out(tmp, parent_dir_entry->path,
 583                                                         cur_file->d_name)) {
 584                                         error("invalid path: %s/%s",
 585                                                         parent_dir_entry->path,
 586                                                         cur_file->d_name);
 587                                         ret = -EINVAL;
 588                                         goto fail;
 589                                 }
 590                                 dir_entry->path = strdup(tmp);
 591                                 if (!dir_entry->path) {
 592                                         error("not enough memory to store path");
 593                                         ret = -ENOMEM;
 594                                         goto fail;
 595                                 }
 596                                 dir_entry->inum = cur_inum;
 597                                 list_add_tail(&dir_entry->list,
 598                                               &dir_head->list);
 599                         } else if (S_ISREG(st.st_mode)) {
 600                                 ret = add_file_items(trans, root, &cur_inode,
 601                                                      cur_inum, &st,
 602                                                      cur_file->d_name);
 603                                 if (ret) {
 604                                         error("unable to add file items for %s: %d",
 605                                                 cur_file->d_name, ret);
 606                                         goto fail;
 607                                 }
 608                         } else if (S_ISLNK(st.st_mode)) {
 609                                 ret = add_symbolic_link(trans, root,
 610                                                 cur_inum, cur_file->d_name);
 611                                 if (ret) {
 612                                         error("unable to add symlink for %s: %d",
 613                                                 cur_file->d_name, ret);
 614                                         goto fail;
 615                                 }
 616                         }
 617                 }
 618
 619                 free_namelist(files, count);
 620                 free(parent_dir_entry->path);
 621                 free(parent_dir_entry);
 622
 623                 index_cnt = 2;
 624
 625         } while (!list_empty(&dir_head->list));
 626
 627 out:
 628         return !!ret;
 629 fail:
 630         free_namelist(files, count);
 631 fail_no_files:
 632         free(parent_dir_entry);
 633         goto out;
 634 fail_no_dir:
 635         free(dir_entry);
 636         goto out;
 637 }
 638
 639 int btrfs_mkfs_fill_dir(const char *source_dir, struct btrfs_root *root,
 640                         bool verbose)
 641 {
 642         int ret;
 643         struct btrfs_trans_handle *trans;
 644         struct stat root_st;
 645         struct directory_name_entry dir_head;
 646         struct directory_name_entry *dir_entry = NULL;
 647
 648         ret = lstat(source_dir, &root_st);
 649         if (ret) {
 650                 error("unable to lstat %s: %s", source_dir, strerror(errno));
 651                 ret = -errno;
 652                 goto out;
 653         }
 654
 655         INIT_LIST_HEAD(&dir_head.list);
 656
 657         trans = btrfs_start_transaction(root, 1);
 658         BUG_ON(IS_ERR(trans));
 659         ret = traverse_directory(trans, root, source_dir, &dir_head);
 660         if (ret) {
 661                 error("unable to traverse directory %s: %d", source_dir, ret);
 662                 goto fail;
 663         }
 664         ret = btrfs_commit_transaction(trans, root);
 665         if (ret) {
 666                 error("transaction commit failed: %d", ret);
 667                 goto out;
 668         }
 669
 670         if (verbose)
 671                 printf("Making image is completed.\n");
 672         return 0;
 673 fail:
 674         /*
 675          * Since we don't have btrfs_abort_transaction() yet, uncommitted trans
 676          * will trigger a BUG_ON().
 677          *
 678          * However before mkfs is fully finished, the magic number is invalid,
 679          * so even we commit transaction here, the fs still can't be mounted.
 680          *
 681          * To do a graceful error out, here we commit transaction as a
 682          * workaround.
 683          * Since we have already hit some problem, the return value doesn't
 684          * matter now.
 685          */
 686         btrfs_commit_transaction(trans, root);
 687         while (!list_empty(&dir_head.list)) {
 688                 dir_entry = list_entry(dir_head.list.next,
 689                                        struct directory_name_entry, list);
 690                 list_del(&dir_entry->list);
 691                 free(dir_entry->path);
 692                 free(dir_entry);
 693         }
 694 out:
 695         return ret;
 696 }
 697
 698 static int ftw_add_entry_size(const char *fpath, const struct stat *st,
 699                               int type, struct FTW *ftwbuf)
 700 {
 701         /*
 702          * Failed to read the directory, mostly due to EPERM.  Abort ASAP, so
 703          * we don't need to populate the fs.
 704          */
 705         if (type == FTW_DNR || type == FTW_NS)
 706                 return -EPERM;
 707
 708         if (S_ISREG(st->st_mode))
 709                 ftw_data_size += round_up(st->st_size, fs_block_size);
 710         ftw_meta_nr_inode++;
 711
 712         return 0;
 713 }
 714
 715 u64 btrfs_mkfs_size_dir(const char *dir_name, u32 sectorsize, u64 min_dev_size,
 716                         u64 meta_profile, u64 data_profile)
 717 {
 718         u64 total_size = 0;
 719         int ret;
 720
 721         u64 meta_size = 0;              /* Based on @ftw_meta_nr_inode */
 722         u64 meta_chunk_size = 0;        /* Based on @meta_size */
 723         u64 data_chunk_size = 0;        /* Based on @ftw_data_size */
 724
 725         u64 meta_threshold = SZ_8M;
 726         u64 data_threshold = SZ_8M;
 727
 728         float data_multipler = 1;
 729         float meta_multipler = 1;
 730
 731         fs_block_size = sectorsize;
 732         ftw_data_size = 0;
 733         ftw_meta_nr_inode = 0;
 734
 735         /*
 736          * Symbolic link is not followed when creating files, so no need to
 737          * follow them here.
 738          */
 739         ret = nftw(dir_name, ftw_add_entry_size, 10, FTW_PHYS);
 740         if (ret < 0) {
 741                 error("ftw subdir walk of %s failed: %s", dir_name,
 742                         strerror(errno));
 743                 exit(1);
 744         }
 745
 746
 747         /*
 748          * Maximum metadata useage for every inode, which will be PATH_MAX
 749          * for the following items:
 750          * 1) DIR_ITEM
 751          * 2) DIR_INDEX
 752          * 3) INODE_REF
 753          *
 754          * Plus possible inline extent size, which is sectorsize.
 755          *
 756          * And finally, allow metadata usage to increase with data size.
 757          * Follow the old kernel 8:1 data:meta ratio.
 758          * This is especially important for --rootdir, as the file extent size
 759          * upper limit is 1M, instead of 128M in kernel.
 760          * This can bump meta usage easily.
 761          */
 762         meta_size = ftw_meta_nr_inode * (PATH_MAX * 3 + sectorsize) +
 763                     ftw_data_size / 8;
 764
 765         /* Minimal chunk size from btrfs_alloc_chunk(). */
 766         if (meta_profile & BTRFS_BLOCK_GROUP_DUP) {
 767                 meta_threshold = SZ_32M;
 768                 meta_multipler = 2;
 769         }
 770         if (data_profile & BTRFS_BLOCK_GROUP_DUP) {
 771                 data_threshold = SZ_64M;
 772                 data_multipler = 2;
 773         }
 774
 775         /*
 776          * Only when the usage is larger than the minimal chunk size (threshold)
 777          * we need to allocate new chunk, or the initial chunk in the image is
 778          * large enough.
 779          */
 780         if (meta_size > meta_threshold)
 781                 meta_chunk_size = (round_up(meta_size, meta_threshold) -
 782                                    meta_threshold) * meta_multipler;
 783         if (ftw_data_size > data_threshold)
 784                 data_chunk_size = (round_up(ftw_data_size, data_threshold) -
 785                                    data_threshold) * data_multipler;
 786
 787         total_size = data_chunk_size + meta_chunk_size + min_dev_size;
 788         return total_size;
 789 }
 790
 791 /*
 792  * Get the end position of the last device extent for given @devid;
 793  * @size_ret is exclsuive (means it should be aligned to sectorsize)
 794  */
 795 static int get_device_extent_end(struct btrfs_fs_info *fs_info,
 796                                  u64 devid, u64 *size_ret)
 797 {
 798         struct btrfs_root *dev_root = fs_info->dev_root;
 799         struct btrfs_key key;
 800         struct btrfs_path path;
 801         struct btrfs_dev_extent *de;
 802         int ret;
 803
 804         key.objectid = devid;
 805         key.type = BTRFS_DEV_EXTENT_KEY;
 806         key.offset = (u64)-1;
 807
 808         btrfs_init_path(&path);
 809         ret = btrfs_search_slot(NULL, dev_root, &key, &path, 0, 0);
 810         /* Not really possible */
 811         BUG_ON(ret == 0);
 812
 813         ret = btrfs_previous_item(dev_root, &path, devid, BTRFS_DEV_EXTENT_KEY);
 814         if (ret < 0)
 815                 goto out;
 816
 817         /* No dev_extent at all, not really possible for rootdir case */
 818         if (ret > 0) {
 819                 *size_ret = 0;
 820                 ret = -EUCLEAN;
 821                 goto out;
 822         }
 823
 824         btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
 825         de = btrfs_item_ptr(path.nodes[0], path.slots[0],
 826                             struct btrfs_dev_extent);
 827         *size_ret = key.offset + btrfs_dev_extent_length(path.nodes[0], de);
 828 out:
 829         btrfs_release_path(&path);
 830
 831         return ret;
 832 }
 833
 834 /*
 835  * Set device size to @new_size.
 836  *
 837  * Only used for --rootdir option.
 838  * We will need to reset the following values:
 839  * 1) dev item in chunk tree
 840  * 2) super->dev_item
 841  * 3) super->total_bytes
 842  */
 843 static int set_device_size(struct btrfs_fs_info *fs_info,
 844                            struct btrfs_device *device, u64 new_size)
 845 {
 846         struct btrfs_root *chunk_root = fs_info->chunk_root;
 847         struct btrfs_trans_handle *trans;
 848         struct btrfs_dev_item *di;
 849         struct btrfs_path path;
 850         struct btrfs_key key;
 851         int ret;
 852
 853         /*
 854          * Update in-meory device->total_bytes, so that at trans commit time,
 855          * super->dev_item will also get updated
 856          */
 857         device->total_bytes = new_size;
 858         btrfs_init_path(&path);
 859
 860         /* Update device item in chunk tree */
 861         trans = btrfs_start_transaction(chunk_root, 1);
 862         if (IS_ERR(trans)) {
 863                 ret = PTR_ERR(trans);
 864                 error("failed to start transaction: %d (%s)", ret,
 865                         strerror(-ret));
 866                 return ret;
 867         }
 868         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
 869         key.type = BTRFS_DEV_ITEM_KEY;
 870         key.offset = device->devid;
 871
 872         ret = btrfs_search_slot(trans, chunk_root, &key, &path, 0, 1);
 873         if (ret < 0)
 874                 goto err;
 875         if (ret > 0)
 876                 ret = -ENOENT;
 877         di = btrfs_item_ptr(path.nodes[0], path.slots[0],
 878                             struct btrfs_dev_item);
 879         btrfs_set_device_total_bytes(path.nodes[0], di, new_size);
 880         btrfs_mark_buffer_dirty(path.nodes[0]);
 881
 882         /*
 883          * Update super->total_bytes, since it's only used for --rootdir,
 884          * there is only one device, just use the @new_size.
 885          */
 886         btrfs_set_super_total_bytes(fs_info->super_copy, new_size);
 887
 888         /*
 889          * Commit transaction to reflect the updated super->total_bytes and
 890          * super->dev_item
 891          */
 892         ret = btrfs_commit_transaction(trans, chunk_root);
 893         if (ret < 0)
 894                 error("failed to commit current transaction: %d (%s)",
 895                         ret, strerror(-ret));
 896         btrfs_release_path(&path);
 897         return ret;
 898
 899 err:
 900         btrfs_release_path(&path);
 901         /*
 902          * Committing the transaction here won't cause problems since the fs
 903          * still has an invalid magic number, and something wrong already
 904          * happened, we don't care the return value anyway.
 905          */
 906         btrfs_commit_transaction(trans, chunk_root);
 907         return ret;
 908 }
 909
 910 int btrfs_mkfs_shrink_fs(struct btrfs_fs_info *fs_info, u64 *new_size_ret,
 911                          bool shrink_file_size)
 912 {
 913         u64 new_size;
 914         struct btrfs_device *device;
 915         struct list_head *cur;
 916         struct stat64 file_stat;
 917         int nr_devs = 0;
 918         int ret;
 919
 920         list_for_each(cur, &fs_info->fs_devices->devices)
 921                 nr_devs++;
 922
 923         if (nr_devs > 1) {
 924                 error("cannot shrink fs with more than 1 device");
 925                 return -ENOTTY;
 926         }
 927
 928         ret = get_device_extent_end(fs_info, 1, &new_size);
 929         if (ret < 0) {
 930                 error("failed to get minimal device size: %d (%s)",
 931                         ret, strerror(-ret));
 932                 return ret;
 933         }
 934
 935         BUG_ON(!IS_ALIGNED(new_size, fs_info->sectorsize));
 936
 937         device = list_entry(fs_info->fs_devices->devices.next,
 938                            struct btrfs_device, dev_list);
 939         ret = set_device_size(fs_info, device, new_size);
 940         if (ret < 0)
 941                 return ret;
 942         if (new_size_ret)
 943                 *new_size_ret = new_size;
 944
 945         if (shrink_file_size) {
 946                 ret = fstat64(device->fd, &file_stat);
 947                 if (ret < 0) {
 948                         error("failed to stat devid %llu: %s", device->devid,
 949                                 strerror(errno));
 950                         return ret;
 951                 }
 952                 if (!S_ISREG(file_stat.st_mode))
 953                         return ret;
 954                 ret = ftruncate64(device->fd, new_size);
 955                 if (ret < 0) {
 956                         error("failed to truncate device file of devid %llu: %s",
 957                                 device->devid, strerror(errno));
 958                         return ret;
 959                 }
 960         }
 961         return ret;
 962 }