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