lib/ext2fs/mkjournal.c

   1 /*
   2  * mkjournal.c --- make a journal for a filesystem
   3  *
   4  * Copyright (C) 2000 Theodore Ts'o.
   5  *
   6  * %Begin-Header%
   7  * This file may be redistributed under the terms of the GNU Library
   8  * General Public License, version 2.
   9  * %End-Header%
  10  */
  11
  12 #include "config.h"
  13 #include <stdio.h>
  14 #include <string.h>
  15 #if HAVE_UNISTD_H
  16 #include <unistd.h>
  17 #endif
  18 #if HAVE_ERRNO_H
  19 #include <errno.h>
  20 #endif
  21 #include <fcntl.h>
  22 #include <time.h>
  23 #if HAVE_SYS_STAT_H
  24 #include <sys/stat.h>
  25 #endif
  26 #if HAVE_SYS_TYPES_H
  27 #include <sys/types.h>
  28 #endif
  29 #if HAVE_SYS_IOCTL_H
  30 #include <sys/ioctl.h>
  31 #endif
  32 #if HAVE_NETINET_IN_H
  33 #include <netinet/in.h>
  34 #endif
  35
  36 #include "ext2_fs.h"
  37 #include "e2p/e2p.h"
  38 #include "ext2fs.h"
  39 #include "jfs_user.h"
  40
  41 /*
  42  * This function automatically sets up the journal superblock and
  43  * returns it as an allocated block.
  44  */
  45 errcode_t ext2fs_create_journal_superblock(ext2_filsys fs,
  46                                            __u32 num_blocks, int flags,
  47                                            char  **ret_jsb)
  48 {
  49         errcode_t               retval;
  50         journal_superblock_t    *jsb;
  51
  52         if (num_blocks < 1024)
  53                 return EXT2_ET_JOURNAL_TOO_SMALL;
  54
  55         if ((retval = ext2fs_get_mem(fs->blocksize, &jsb)))
  56                 return retval;
  57
  58         memset (jsb, 0, fs->blocksize);
  59
  60         jsb->s_header.h_magic = htonl(JFS_MAGIC_NUMBER);
  61         if (flags & EXT2_MKJOURNAL_V1_SUPER)
  62                 jsb->s_header.h_blocktype = htonl(JFS_SUPERBLOCK_V1);
  63         else
  64                 jsb->s_header.h_blocktype = htonl(JFS_SUPERBLOCK_V2);
  65         jsb->s_blocksize = htonl(fs->blocksize);
  66         jsb->s_maxlen = htonl(num_blocks);
  67         jsb->s_nr_users = htonl(1);
  68         jsb->s_first = htonl(1);
  69         jsb->s_sequence = htonl(1);
  70         memcpy(jsb->s_uuid, fs->super->s_uuid, sizeof(fs->super->s_uuid));
  71         /*
  72          * If we're creating an external journal device, we need to
  73          * adjust these fields.
  74          */
  75         if (fs->super->s_feature_incompat &
  76             EXT3_FEATURE_INCOMPAT_JOURNAL_DEV) {
  77                 jsb->s_nr_users = 0;
  78                 if (fs->blocksize == 1024)
  79                         jsb->s_first = htonl(3);
  80                 else
  81                         jsb->s_first = htonl(2);
  82         }
  83
  84         *ret_jsb = (char *) jsb;
  85         return 0;
  86 }
  87
  88 /*
  89  * This function writes a journal using POSIX routines.  It is used
  90  * for creating external journals and creating journals on live
  91  * filesystems.
  92  */
  93 static errcode_t write_journal_file(ext2_filsys fs, char *filename,
  94                                     blk_t num_blocks, int flags)
  95 {
  96         errcode_t       retval;
  97         char            *buf = 0;
  98         int             fd, ret_size;
  99         blk_t           i;
 100
 101         if ((retval = ext2fs_create_journal_superblock(fs, num_blocks, flags,
 102                                                        &buf)))
 103                 return retval;
 104
 105         /* Open the device or journal file */
 106         if ((fd = open(filename, O_WRONLY)) < 0) {
 107                 retval = errno;
 108                 goto errfree;
 109         }
 110
 111         /* Write the superblock out */
 112         retval = EXT2_ET_SHORT_WRITE;
 113         ret_size = write(fd, buf, fs->blocksize);
 114         if (ret_size < 0) {
 115                 retval = errno;
 116                 goto errout;
 117         }
 118         if (ret_size != (int) fs->blocksize)
 119                 goto errout;
 120         memset(buf, 0, fs->blocksize);
 121
 122         if (flags & EXT2_MKJOURNAL_LAZYINIT)
 123                 goto success;
 124
 125         for (i = 1; i < num_blocks; i++) {
 126                 ret_size = write(fd, buf, fs->blocksize);
 127                 if (ret_size < 0) {
 128                         retval = errno;
 129                         goto errout;
 130                 }
 131                 if (ret_size != (int) fs->blocksize)
 132                         goto errout;
 133         }
 134
 135 success:
 136         retval = 0;
 137 errout:
 138         close(fd);
 139 errfree:
 140         ext2fs_free_mem(&buf);
 141         return retval;
 142 }
 143
 144 /*
 145  * Convenience function which zeros out _num_ blocks starting at
 146  * _blk_.  In case of an error, the details of the error is returned
 147  * via _ret_blk_ and _ret_count_ if they are non-NULL pointers.
 148  * Returns 0 on success, and an error code on an error.
 149  *
 150  * As a special case, if the first argument is NULL, then it will
 151  * attempt to free the static zeroizing buffer.  (This is to keep
 152  * programs that check for memory leaks happy.)
 153  */
 154 #define STRIDE_LENGTH 8
 155 errcode_t ext2fs_zero_blocks2(ext2_filsys fs, blk64_t blk, int num,
 156                               blk64_t *ret_blk, int *ret_count)
 157 {
 158         int             j, count;
 159         static char     *buf;
 160         errcode_t       retval;
 161
 162         /* If fs is null, clean up the static buffer and return */
 163         if (!fs) {
 164                 if (buf) {
 165                         free(buf);
 166                         buf = 0;
 167                 }
 168                 return 0;
 169         }
 170         /* Allocate the zeroizing buffer if necessary */
 171         if (!buf) {
 172                 buf = malloc(fs->blocksize * STRIDE_LENGTH);
 173                 if (!buf)
 174                         return ENOMEM;
 175                 memset(buf, 0, fs->blocksize * STRIDE_LENGTH);
 176         }
 177         /* OK, do the write loop */
 178         j=0;
 179         while (j < num) {
 180                 if (blk % STRIDE_LENGTH) {
 181                         count = STRIDE_LENGTH - (blk % STRIDE_LENGTH);
 182                         if (count > (num - j))
 183                                 count = num - j;
 184                 } else {
 185                         count = num - j;
 186                         if (count > STRIDE_LENGTH)
 187                                 count = STRIDE_LENGTH;
 188                 }
 189                 retval = io_channel_write_blk64(fs->io, blk, count, buf);
 190                 if (retval) {
 191                         if (ret_count)
 192                                 *ret_count = count;
 193                         if (ret_blk)
 194                                 *ret_blk = blk;
 195                         return retval;
 196                 }
 197                 j += count; blk += count;
 198         }
 199         return 0;
 200 }
 201
 202 errcode_t ext2fs_zero_blocks(ext2_filsys fs, blk_t blk, int num,
 203                              blk_t *ret_blk, int *ret_count)
 204 {
 205         blk64_t ret_blk2;
 206         errcode_t retval;
 207
 208         retval = ext2fs_zero_blocks2(fs, blk, num, &ret_blk2, ret_count);
 209         if (retval)
 210                 *ret_blk = (blk_t) ret_blk2;
 211         return retval;
 212 }
 213
 214 /*
 215  * Helper function for creating the journal using direct I/O routines
 216  */
 217 struct mkjournal_struct {
 218         int             num_blocks;
 219         int             newblocks;
 220         blk64_t         goal;
 221         blk64_t         blk_to_zero;
 222         int             zero_count;
 223         int             flags;
 224         char            *buf;
 225         errcode_t       err;
 226 };
 227
 228 static int mkjournal_proc(ext2_filsys   fs,
 229                           blk64_t       *blocknr,
 230                           e2_blkcnt_t   blockcnt,
 231                           blk64_t       ref_block EXT2FS_ATTR((unused)),
 232                           int           ref_offset EXT2FS_ATTR((unused)),
 233                           void          *priv_data)
 234 {
 235         struct mkjournal_struct *es = (struct mkjournal_struct *) priv_data;
 236         blk64_t new_blk;
 237         errcode_t       retval;
 238
 239         if (*blocknr) {
 240                 es->goal = *blocknr;
 241                 return 0;
 242         }
 243         if (blockcnt &&
 244             (EXT2FS_B2C(fs, es->goal) == EXT2FS_B2C(fs, es->goal+1)))
 245                 new_blk = es->goal+1;
 246         else {
 247                 es->goal &= ~EXT2FS_CLUSTER_MASK(fs);
 248                 retval = ext2fs_new_block2(fs, es->goal, 0, &new_blk);
 249                 if (retval) {
 250                         es->err = retval;
 251                         return BLOCK_ABORT;
 252                 }
 253                 es->newblocks++;
 254         }
 255         if (blockcnt >= 0)
 256                 es->num_blocks--;
 257
 258         retval = 0;
 259         if (blockcnt <= 0)
 260                 retval = io_channel_write_blk64(fs->io, new_blk, 1, es->buf);
 261         else if (!(es->flags & EXT2_MKJOURNAL_LAZYINIT)) {
 262                 if (es->zero_count) {
 263                         if ((es->blk_to_zero + es->zero_count == new_blk) &&
 264                             (es->zero_count < 1024))
 265                                 es->zero_count++;
 266                         else {
 267                                 retval = ext2fs_zero_blocks2(fs,
 268                                                              es->blk_to_zero,
 269                                                              es->zero_count,
 270                                                              0, 0);
 271                                 es->zero_count = 0;
 272                         }
 273                 }
 274                 if (es->zero_count == 0) {
 275                         es->blk_to_zero = new_blk;
 276                         es->zero_count = 1;
 277                 }
 278         }
 279
 280         if (blockcnt == 0)
 281                 memset(es->buf, 0, fs->blocksize);
 282
 283         if (retval) {
 284                 es->err = retval;
 285                 return BLOCK_ABORT;
 286         }
 287         *blocknr = es->goal = new_blk;
 288         ext2fs_block_alloc_stats2(fs, new_blk, +1);
 289
 290         if (es->num_blocks == 0)
 291                 return (BLOCK_CHANGED | BLOCK_ABORT);
 292         else
 293                 return BLOCK_CHANGED;
 294
 295 }
 296
 297 /*
 298  * This function creates a journal using direct I/O routines.
 299  */
 300 static errcode_t write_journal_inode(ext2_filsys fs, ext2_ino_t journal_ino,
 301                                      blk_t num_blocks, int flags)
 302 {
 303         char                    *buf;
 304         dgrp_t                  group, start, end, i, log_flex;
 305         errcode_t               retval;
 306         struct ext2_inode       inode;
 307         unsigned long long      inode_size;
 308         struct mkjournal_struct es;
 309
 310         if ((retval = ext2fs_create_journal_superblock(fs, num_blocks, flags,
 311                                                        &buf)))
 312                 return retval;
 313
 314         if ((retval = ext2fs_read_bitmaps(fs)))
 315                 return retval;
 316
 317         if ((retval = ext2fs_read_inode(fs, journal_ino, &inode)))
 318                 return retval;
 319
 320         if (inode.i_blocks > 0)
 321                 return EEXIST;
 322
 323         es.num_blocks = num_blocks;
 324         es.newblocks = 0;
 325         es.buf = buf;
 326         es.err = 0;
 327         es.flags = flags;
 328         es.zero_count = 0;
 329
 330         if (fs->super->s_feature_incompat & EXT3_FEATURE_INCOMPAT_EXTENTS) {
 331                 inode.i_flags |= EXT4_EXTENTS_FL;
 332                 if ((retval = ext2fs_write_inode(fs, journal_ino, &inode)))
 333                         return retval;
 334         }
 335
 336         /*
 337          * Set the initial goal block to be roughly at the middle of
 338          * the filesystem.  Pick a group that has the largest number
 339          * of free blocks.
 340          */
 341         group = ext2fs_group_of_blk2(fs, (ext2fs_blocks_count(fs->super) -
 342                                          fs->super->s_first_data_block) / 2);
 343         log_flex = 1 << fs->super->s_log_groups_per_flex;
 344         if (fs->super->s_log_groups_per_flex && (group > log_flex)) {
 345                 group = group & ~(log_flex - 1);
 346                 while ((group < fs->group_desc_count) &&
 347                        ext2fs_bg_free_blocks_count(fs, group) == 0)
 348                         group++;
 349                 if (group == fs->group_desc_count)
 350                         group = 0;
 351                 start = group;
 352         } else
 353                 start = (group > 0) ? group-1 : group;
 354         end = ((group+1) < fs->group_desc_count) ? group+1 : group;
 355         group = start;
 356         for (i=start+1; i <= end; i++)
 357                 if (ext2fs_bg_free_blocks_count(fs, i) >
 358                     ext2fs_bg_free_blocks_count(fs, group))
 359                         group = i;
 360
 361         es.goal = (fs->super->s_blocks_per_group * group) +
 362                 fs->super->s_first_data_block;
 363
 364         retval = ext2fs_block_iterate3(fs, journal_ino, BLOCK_FLAG_APPEND,
 365                                        0, mkjournal_proc, &es);
 366         if (es.err) {
 367                 retval = es.err;
 368                 goto errout;
 369         }
 370         if (es.zero_count) {
 371                 retval = ext2fs_zero_blocks2(fs, es.blk_to_zero,
 372                                             es.zero_count, 0, 0);
 373                 if (retval)
 374                         goto errout;
 375         }
 376
 377         if ((retval = ext2fs_read_inode(fs, journal_ino, &inode)))
 378                 goto errout;
 379
 380         inode_size = (unsigned long long)fs->blocksize * num_blocks;
 381         inode.i_size = inode_size & 0xFFFFFFFF;
 382         inode.i_size_high = (inode_size >> 32) & 0xFFFFFFFF;
 383         if (inode.i_size_high)
 384                 fs->super->s_feature_ro_compat |=
 385                         EXT2_FEATURE_RO_COMPAT_LARGE_FILE;
 386         ext2fs_iblk_add_blocks(fs, &inode, es.newblocks);
 387         inode.i_mtime = inode.i_ctime = fs->now ? fs->now : time(0);
 388         inode.i_links_count = 1;
 389         inode.i_mode = LINUX_S_IFREG | 0600;
 390
 391         if ((retval = ext2fs_write_new_inode(fs, journal_ino, &inode)))
 392                 goto errout;
 393         retval = 0;
 394
 395         memcpy(fs->super->s_jnl_blocks, inode.i_block, EXT2_N_BLOCKS*4);
 396         fs->super->s_jnl_blocks[15] = inode.i_size_high;
 397         fs->super->s_jnl_blocks[16] = inode.i_size;
 398         fs->super->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
 399         ext2fs_mark_super_dirty(fs);
 400
 401 errout:
 402         ext2fs_zero_blocks2(0, 0, 0, 0, 0);
 403         ext2fs_free_mem(&buf);
 404         return retval;
 405 }
 406
 407 /*
 408  * Find a reasonable journal file size (in blocks) given the number of blocks
 409  * in the filesystem.  For very small filesystems, it is not reasonable to
 410  * have a journal that fills more than half of the filesystem.
 411  */
 412 int ext2fs_default_journal_size(__u64 num_blocks)
 413 {
 414         if (num_blocks < 2048)
 415                 return -1;
 416         if (num_blocks < 32768)
 417                 return (1024);
 418         if (num_blocks < 256*1024)
 419                 return (4096);
 420         if (num_blocks < 512*1024)
 421                 return (8192);
 422         if (num_blocks < 1024*1024)
 423                 return (16384);
 424         return 32768;
 425 }
 426
 427 /*
 428  * This function adds a journal device to a filesystem
 429  */
 430 errcode_t ext2fs_add_journal_device(ext2_filsys fs, ext2_filsys journal_dev)
 431 {
 432         struct stat     st;
 433         errcode_t       retval;
 434         char            buf[1024];
 435         journal_superblock_t    *jsb;
 436         int             start;
 437         __u32           i, nr_users;
 438
 439         /* Make sure the device exists and is a block device */
 440         if (stat(journal_dev->device_name, &st) < 0)
 441                 return errno;
 442
 443         if (!S_ISBLK(st.st_mode))
 444                 return EXT2_ET_JOURNAL_NOT_BLOCK; /* Must be a block device */
 445
 446         /* Get the journal superblock */
 447         start = 1;
 448         if (journal_dev->blocksize == 1024)
 449                 start++;
 450         if ((retval = io_channel_read_blk64(journal_dev->io, start, -1024,
 451                                             buf)))
 452                 return retval;
 453
 454         jsb = (journal_superblock_t *) buf;
 455         if ((jsb->s_header.h_magic != (unsigned) ntohl(JFS_MAGIC_NUMBER)) ||
 456             (jsb->s_header.h_blocktype != (unsigned) ntohl(JFS_SUPERBLOCK_V2)))
 457                 return EXT2_ET_NO_JOURNAL_SB;
 458
 459         if (ntohl(jsb->s_blocksize) != (unsigned long) fs->blocksize)
 460                 return EXT2_ET_UNEXPECTED_BLOCK_SIZE;
 461
 462         /* Check and see if this filesystem has already been added */
 463         nr_users = ntohl(jsb->s_nr_users);
 464         for (i=0; i < nr_users; i++) {
 465                 if (memcmp(fs->super->s_uuid,
 466                            &jsb->s_users[i*16], 16) == 0)
 467                         break;
 468         }
 469         if (i >= nr_users) {
 470                 memcpy(&jsb->s_users[nr_users*16],
 471                        fs->super->s_uuid, 16);
 472                 jsb->s_nr_users = htonl(nr_users+1);
 473         }
 474
 475         /* Writeback the journal superblock */
 476         if ((retval = io_channel_write_blk64(journal_dev->io, start, -1024, buf)))
 477                 return retval;
 478
 479         fs->super->s_journal_inum = 0;
 480         fs->super->s_journal_dev = st.st_rdev;
 481         memcpy(fs->super->s_journal_uuid, jsb->s_uuid,
 482                sizeof(fs->super->s_journal_uuid));
 483         fs->super->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 484         ext2fs_mark_super_dirty(fs);
 485         return 0;
 486 }
 487
 488 /*
 489  * This function adds a journal inode to a filesystem, using either
 490  * POSIX routines if the filesystem is mounted, or using direct I/O
 491  * functions if it is not.
 492  */
 493 errcode_t ext2fs_add_journal_inode(ext2_filsys fs, blk_t num_blocks, int flags)
 494 {
 495         errcode_t               retval;
 496         ext2_ino_t              journal_ino;
 497         struct stat             st;
 498         char                    jfile[1024];
 499         int                     mount_flags, f;
 500         int                     fd = -1;
 501
 502         if (flags & EXT2_MKJOURNAL_NO_MNT_CHECK)
 503                 mount_flags = 0;
 504         else if ((retval = ext2fs_check_mount_point(fs->device_name,
 505                                                     &mount_flags,
 506                                                     jfile, sizeof(jfile)-10)))
 507                 return retval;
 508
 509         if (mount_flags & EXT2_MF_MOUNTED) {
 510                 strcat(jfile, "/.journal");
 511
 512                 /*
 513                  * If .../.journal already exists, make sure any
 514                  * immutable or append-only flags are cleared.
 515                  */
 516 #if defined(HAVE_CHFLAGS) && defined(UF_NODUMP)
 517                 (void) chflags (jfile, 0);
 518 #else
 519 #if HAVE_EXT2_IOCTLS
 520                 fd = open(jfile, O_RDONLY);
 521                 if (fd >= 0) {
 522                         f = 0;
 523                         ioctl(fd, EXT2_IOC_SETFLAGS, &f);
 524                         close(fd);
 525                 }
 526 #endif
 527 #endif
 528
 529                 /* Create the journal file */
 530                 if ((fd = open(jfile, O_CREAT|O_WRONLY, 0600)) < 0)
 531                         return errno;
 532
 533                 /* Note that we can't do lazy journal initialization for mounted
 534                  * filesystems, since the zero writing is also allocating the
 535                  * journal blocks.  We could use fallocate, but not all kernels
 536                  * support that, and creating a journal on a mounted ext2
 537                  * filesystems is extremely rare these days...  Ignore it. */
 538                 flags &= ~EXT2_MKJOURNAL_LAZYINIT;
 539
 540                 if ((retval = write_journal_file(fs, jfile, num_blocks, flags)))
 541                         goto errout;
 542
 543                 /* Get inode number of the journal file */
 544                 if (fstat(fd, &st) < 0) {
 545                         retval = errno;
 546                         goto errout;
 547                 }
 548
 549 #if defined(HAVE_CHFLAGS) && defined(UF_NODUMP)
 550                 retval = fchflags (fd, UF_NODUMP|UF_IMMUTABLE);
 551 #else
 552 #if HAVE_EXT2_IOCTLS
 553                 if (ioctl(fd, EXT2_IOC_GETFLAGS, &f) < 0) {
 554                         retval = errno;
 555                         goto errout;
 556                 }
 557                 f |= EXT2_NODUMP_FL | EXT2_IMMUTABLE_FL;
 558                 retval = ioctl(fd, EXT2_IOC_SETFLAGS, &f);
 559 #endif
 560 #endif
 561                 if (retval) {
 562                         retval = errno;
 563                         goto errout;
 564                 }
 565
 566                 if (close(fd) < 0) {
 567                         retval = errno;
 568                         fd = -1;
 569                         goto errout;
 570                 }
 571                 journal_ino = st.st_ino;
 572         } else {
 573                 if ((mount_flags & EXT2_MF_BUSY) &&
 574                     !(fs->flags & EXT2_FLAG_EXCLUSIVE)) {
 575                         retval = EBUSY;
 576                         goto errout;
 577                 }
 578                 journal_ino = EXT2_JOURNAL_INO;
 579                 if ((retval = write_journal_inode(fs, journal_ino,
 580                                                   num_blocks, flags)))
 581                         return retval;
 582         }
 583
 584         fs->super->s_journal_inum = journal_ino;
 585         fs->super->s_journal_dev = 0;
 586         memset(fs->super->s_journal_uuid, 0,
 587                sizeof(fs->super->s_journal_uuid));
 588         fs->super->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 589
 590         ext2fs_mark_super_dirty(fs);
 591         return 0;
 592 errout:
 593         if (fd > 0)
 594                 close(fd);
 595         return retval;
 596 }
 597
 598 #ifdef DEBUG
 599 main(int argc, char **argv)
 600 {
 601         errcode_t       retval;
 602         char            *device_name;
 603         ext2_filsys     fs;
 604
 605         if (argc < 2) {
 606                 fprintf(stderr, "Usage: %s filesystem\n", argv[0]);
 607                 exit(1);
 608         }
 609         device_name = argv[1];
 610
 611         retval = ext2fs_open (device_name, EXT2_FLAG_RW, 0, 0,
 612                               unix_io_manager, &fs);
 613         if (retval) {
 614                 com_err(argv[0], retval, "while opening %s", device_name);
 615                 exit(1);
 616         }
 617
 618         retval = ext2fs_add_journal_inode(fs, 1024, 0);
 619         if (retval) {
 620                 com_err(argv[0], retval, "while adding journal to %s",
 621                         device_name);
 622                 exit(1);
 623         }
 624         retval = ext2fs_flush(fs);
 625         if (retval) {
 626                 printf("Warning, had trouble writing out superblocks.\n");
 627         }
 628         ext2fs_close(fs);
 629         exit(0);
 630
 631 }
 632 #endif