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 #ifdef __linux__
  36 #include <sys/utsname.h>
  37 #define KERNEL_VERSION(a,b,c) (((a) << 16) + ((b) << 8) + (c))
  38 #endif
  39
  40 #include "ext2_fs.h"
  41 #include "e2p/e2p.h"
  42 #include "ext2fs.h"
  43
  44 #include "kernel-jbd.h"
  45
  46 /*
  47  * This function automatically sets up the journal superblock and
  48  * returns it as an allocated block.
  49  */
  50 errcode_t ext2fs_create_journal_superblock(ext2_filsys fs,
  51                                            __u32 num_blocks, int flags,
  52                                            char  **ret_jsb)
  53 {
  54         errcode_t               retval;
  55         journal_superblock_t    *jsb;
  56
  57         if (num_blocks < JFS_MIN_JOURNAL_BLOCKS)
  58                 return EXT2_ET_JOURNAL_TOO_SMALL;
  59
  60         if ((retval = ext2fs_get_mem(fs->blocksize, &jsb)))
  61                 return retval;
  62
  63         memset (jsb, 0, fs->blocksize);
  64
  65         jsb->s_header.h_magic = htonl(JFS_MAGIC_NUMBER);
  66         if (flags & EXT2_MKJOURNAL_V1_SUPER)
  67                 jsb->s_header.h_blocktype = htonl(JFS_SUPERBLOCK_V1);
  68         else
  69                 jsb->s_header.h_blocktype = htonl(JFS_SUPERBLOCK_V2);
  70         jsb->s_blocksize = htonl(fs->blocksize);
  71         jsb->s_maxlen = htonl(num_blocks);
  72         jsb->s_nr_users = htonl(1);
  73         jsb->s_first = htonl(1);
  74         jsb->s_sequence = htonl(1);
  75         memcpy(jsb->s_uuid, fs->super->s_uuid, sizeof(fs->super->s_uuid));
  76         /*
  77          * If we're creating an external journal device, we need to
  78          * adjust these fields.
  79          */
  80         if (ext2fs_has_feature_journal_dev(fs->super)) {
  81                 jsb->s_nr_users = 0;
  82                 jsb->s_first = htonl(ext2fs_journal_sb_start(fs->blocksize) + 1);
  83         }
  84
  85         *ret_jsb = (char *) jsb;
  86         return 0;
  87 }
  88
  89 /*
  90  * This function writes a journal using POSIX routines.  It is used
  91  * for creating external journals and creating journals on live
  92  * filesystems.
  93  */
  94 static errcode_t write_journal_file(ext2_filsys fs, char *filename,
  95                                     blk_t num_blocks, int flags)
  96 {
  97         errcode_t       retval;
  98         char            *buf = 0;
  99         int             fd, ret_size;
 100         blk_t           i;
 101
 102         if ((retval = ext2fs_create_journal_superblock(fs, num_blocks, flags,
 103                                                        &buf)))
 104                 return retval;
 105
 106         /* Open the device or journal file */
 107         if ((fd = open(filename, O_WRONLY)) < 0) {
 108                 retval = errno;
 109                 goto errfree;
 110         }
 111
 112         /* Write the superblock out */
 113         retval = EXT2_ET_SHORT_WRITE;
 114         ret_size = write(fd, buf, fs->blocksize);
 115         if (ret_size < 0) {
 116                 retval = errno;
 117                 goto errout;
 118         }
 119         if (ret_size != (int) fs->blocksize)
 120                 goto errout;
 121         memset(buf, 0, fs->blocksize);
 122
 123         if (flags & EXT2_MKJOURNAL_LAZYINIT)
 124                 goto success;
 125
 126         for (i = 1; i < num_blocks; i++) {
 127                 ret_size = write(fd, buf, fs->blocksize);
 128                 if (ret_size < 0) {
 129                         retval = errno;
 130                         goto errout;
 131                 }
 132                 if (ret_size != (int) fs->blocksize)
 133                         goto errout;
 134         }
 135
 136 success:
 137         retval = 0;
 138 errout:
 139         close(fd);
 140 errfree:
 141         ext2fs_free_mem(&buf);
 142         return retval;
 143 }
 144
 145 /*
 146  * Convenience function which zeros out _num_ blocks starting at
 147  * _blk_.  In case of an error, the details of the error is returned
 148  * via _ret_blk_ and _ret_count_ if they are non-NULL pointers.
 149  * Returns 0 on success, and an error code on an error.
 150  *
 151  * As a special case, if the first argument is NULL, then it will
 152  * attempt to free the static zeroizing buffer.  (This is to keep
 153  * programs that check for memory leaks happy.)
 154  */
 155 #define MAX_STRIDE_LENGTH (4194304 / (int) fs->blocksize)
 156 errcode_t ext2fs_zero_blocks2(ext2_filsys fs, blk64_t blk, int num,
 157                               blk64_t *ret_blk, int *ret_count)
 158 {
 159         int             j, count;
 160         static void     *buf;
 161         static int      stride_length;
 162         errcode_t       retval;
 163
 164         /* If fs is null, clean up the static buffer and return */
 165         if (!fs) {
 166                 if (buf) {
 167                         free(buf);
 168                         buf = 0;
 169                         stride_length = 0;
 170                 }
 171                 return 0;
 172         }
 173
 174         /* Deal with zeroing less than 1 block */
 175         if (num <= 0)
 176                 return 0;
 177
 178         /* Try a zero out command, if supported */
 179         retval = io_channel_zeroout(fs->io, blk, num);
 180         if (retval == 0)
 181                 return 0;
 182
 183         /* Allocate the zeroizing buffer if necessary */
 184         if (num > stride_length && stride_length < MAX_STRIDE_LENGTH) {
 185                 void *p;
 186                 int new_stride = num;
 187
 188                 if (new_stride > MAX_STRIDE_LENGTH)
 189                         new_stride = MAX_STRIDE_LENGTH;
 190                 p = realloc(buf, fs->blocksize * new_stride);
 191                 if (!p)
 192                         return EXT2_ET_NO_MEMORY;
 193                 buf = p;
 194                 stride_length = new_stride;
 195                 memset(buf, 0, fs->blocksize * stride_length);
 196         }
 197         /* OK, do the write loop */
 198         j=0;
 199         while (j < num) {
 200                 if (blk % stride_length) {
 201                         count = stride_length - (blk % stride_length);
 202                         if (count > (num - j))
 203                                 count = num - j;
 204                 } else {
 205                         count = num - j;
 206                         if (count > stride_length)
 207                                 count = stride_length;
 208                 }
 209                 retval = io_channel_write_blk64(fs->io, blk, count, buf);
 210                 if (retval) {
 211                         if (ret_count)
 212                                 *ret_count = count;
 213                         if (ret_blk)
 214                                 *ret_blk = blk;
 215                         return retval;
 216                 }
 217                 j += count; blk += count;
 218         }
 219         return 0;
 220 }
 221
 222 errcode_t ext2fs_zero_blocks(ext2_filsys fs, blk_t blk, int num,
 223                              blk_t *ret_blk, int *ret_count)
 224 {
 225         blk64_t ret_blk2;
 226         errcode_t retval;
 227
 228         retval = ext2fs_zero_blocks2(fs, blk, num, &ret_blk2, ret_count);
 229         if (retval)
 230                 *ret_blk = (blk_t) ret_blk2;
 231         return retval;
 232 }
 233
 234 /*
 235  * Calculate the initial goal block to be roughly at the middle of the
 236  * filesystem.  Pick a group that has the largest number of free
 237  * blocks.
 238  */
 239 static blk64_t get_midpoint_journal_block(ext2_filsys fs)
 240 {
 241         dgrp_t  group, start, end, i, log_flex;
 242
 243         group = ext2fs_group_of_blk2(fs, (ext2fs_blocks_count(fs->super) -
 244                                          fs->super->s_first_data_block) / 2);
 245         log_flex = 1 << fs->super->s_log_groups_per_flex;
 246         if (fs->super->s_log_groups_per_flex && (group > log_flex)) {
 247                 group = group & ~(log_flex - 1);
 248                 while ((group < fs->group_desc_count) &&
 249                        ext2fs_bg_free_blocks_count(fs, group) == 0)
 250                         group++;
 251                 if (group == fs->group_desc_count)
 252                         group = 0;
 253                 start = group;
 254         } else
 255                 start = (group > 0) ? group-1 : group;
 256         end = ((group+1) < fs->group_desc_count) ? group+1 : group;
 257         group = start;
 258         for (i = start + 1; i <= end; i++)
 259                 if (ext2fs_bg_free_blocks_count(fs, i) >
 260                     ext2fs_bg_free_blocks_count(fs, group))
 261                         group = i;
 262         return ext2fs_group_first_block2(fs, group);
 263 }
 264
 265 /*
 266  * This function creates a journal using direct I/O routines.
 267  */
 268 static errcode_t write_journal_inode(ext2_filsys fs, ext2_ino_t journal_ino,
 269                                      blk_t num_blocks, blk64_t goal, int flags)
 270 {
 271         char                    *buf;
 272         errcode_t               retval;
 273         struct ext2_inode       inode;
 274         unsigned long long      inode_size;
 275         int                     falloc_flags = EXT2_FALLOCATE_FORCE_INIT;
 276         blk64_t                 zblk;
 277
 278         if ((retval = ext2fs_create_journal_superblock(fs, num_blocks, flags,
 279                                                        &buf)))
 280                 return retval;
 281
 282         if ((retval = ext2fs_read_bitmaps(fs)))
 283                 goto out2;
 284
 285         if ((retval = ext2fs_read_inode(fs, journal_ino, &inode)))
 286                 goto out2;
 287
 288         if (inode.i_blocks > 0) {
 289                 retval = EEXIST;
 290                 goto out2;
 291         }
 292
 293         if (goal == ~0ULL)
 294                 goal = get_midpoint_journal_block(fs);
 295
 296         if (ext2fs_has_feature_extents(fs->super))
 297                 inode.i_flags |= EXT4_EXTENTS_FL;
 298
 299         if (!(flags & EXT2_MKJOURNAL_LAZYINIT))
 300                 falloc_flags |= EXT2_FALLOCATE_ZERO_BLOCKS;
 301
 302         inode_size = (unsigned long long)fs->blocksize * num_blocks;
 303         inode.i_mtime = inode.i_ctime = fs->now ? fs->now : time(0);
 304         inode.i_links_count = 1;
 305         inode.i_mode = LINUX_S_IFREG | 0600;
 306         retval = ext2fs_inode_size_set(fs, &inode, inode_size);
 307         if (retval)
 308                 goto out2;
 309
 310         retval = ext2fs_fallocate(fs, falloc_flags, journal_ino,
 311                                   &inode, goal, 0, num_blocks);
 312         if (retval)
 313                 goto out2;
 314
 315         if ((retval = ext2fs_write_new_inode(fs, journal_ino, &inode)))
 316                 goto out2;
 317
 318         retval = ext2fs_bmap2(fs, journal_ino, &inode, NULL, 0, 0, NULL, &zblk);
 319         if (retval)
 320                 goto out2;
 321
 322         retval = io_channel_write_blk64(fs->io, zblk, 1, buf);
 323         if (retval)
 324                 goto out2;
 325
 326         memcpy(fs->super->s_jnl_blocks, inode.i_block, EXT2_N_BLOCKS*4);
 327         fs->super->s_jnl_blocks[15] = inode.i_size_high;
 328         fs->super->s_jnl_blocks[16] = inode.i_size;
 329         fs->super->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
 330         ext2fs_mark_super_dirty(fs);
 331
 332 out2:
 333         ext2fs_free_mem(&buf);
 334         return retval;
 335 }
 336
 337 #ifdef __linux__
 338 static int parse_version_number(const char *s)
 339 {
 340         int     major, minor, rev;
 341         char    *endptr;
 342         const char *cp = s;
 343
 344         if (!s)
 345                 return 0;
 346         major = strtol(cp, &endptr, 10);
 347         if (cp == endptr || *endptr != '.')
 348                 return 0;
 349         cp = endptr + 1;
 350         minor = strtol(cp, &endptr, 10);
 351         if (cp == endptr || *endptr != '.')
 352                 return 0;
 353         cp = endptr + 1;
 354         rev = strtol(cp, &endptr, 10);
 355         if (cp == endptr)
 356                 return 0;
 357         return KERNEL_VERSION(major, minor, rev);
 358 }
 359
 360 int ext2fs_is_before_linux_ver(unsigned int major, unsigned int minor,
 361                                unsigned int rev)
 362 {
 363         struct          utsname ut;
 364         static int      linux_version_code = -1;
 365
 366         if (uname(&ut)) {
 367                 perror("uname");
 368                 exit(1);
 369         }
 370         if (linux_version_code < 0)
 371                 linux_version_code = parse_version_number(ut.release);
 372         if (linux_version_code == 0)
 373                 return 0;
 374
 375         return linux_version_code < KERNEL_VERSION(major, minor, rev);
 376 }
 377 #else
 378 int ext2fs_is_before_linux_ver(unsigned int major, unsigned int minor,
 379                                unsigned int rev)
 380 {
 381         return 0;
 382 }
 383 #endif
 384
 385 /*
 386  * Find a reasonable journal file size (in blocks) given the number of blocks
 387  * in the filesystem.  For very small filesystems, it is not reasonable to
 388  * have a journal that fills more than half of the filesystem.
 389  *
 390  * n.b. comments assume 4k blocks
 391  */
 392 int ext2fs_default_journal_size(__u64 num_blocks)
 393 {
 394         if (num_blocks < 2048)
 395                 return -1;
 396         if (num_blocks <= 8192)         /* 32 MB */
 397                 return (1024);                  /* 4 MB */
 398         if (num_blocks < 32768)         /* 128 MB */
 399                 return (2048);                  /* 8 MB */
 400         if (num_blocks < 256*1024)      /* 1 GB */
 401                 return (4096);                  /* 16 MB */
 402         if (num_blocks < 512*1024)      /* 2 GB */
 403                 return (8192);                  /* 32 MB */
 404         if (num_blocks < 4096*1024)     /* 16 GB */
 405                 return (16384);                 /* 64 MB */
 406         if (num_blocks < 8192*1024)     /* 32 GB */
 407                 return (32768);                 /* 128 MB */
 408         if (num_blocks < 16384*1024)    /* 64 GB */
 409                 return (65536);                 /* 256 MB */
 410         if (num_blocks < 32768*1024)    /* 128 GB */
 411                 return (131072);                /* 512 MB */
 412         return 262144;                          /* 1 GB */
 413 }
 414
 415 int ext2fs_journal_sb_start(int blocksize)
 416 {
 417         if (blocksize == EXT2_MIN_BLOCK_SIZE)
 418                 return 2;
 419         return 1;
 420 }
 421
 422 /*
 423  * This function adds a journal device to a filesystem
 424  */
 425 errcode_t ext2fs_add_journal_device(ext2_filsys fs, ext2_filsys journal_dev)
 426 {
 427         struct stat     st;
 428         errcode_t       retval;
 429         char            buf[SUPERBLOCK_SIZE];
 430         journal_superblock_t    *jsb;
 431         int             start;
 432         __u32           i, nr_users;
 433
 434         /* Make sure the device exists and is a block device */
 435         if (stat(journal_dev->device_name, &st) < 0)
 436                 return errno;
 437
 438         if (!S_ISBLK(st.st_mode))
 439                 return EXT2_ET_JOURNAL_NOT_BLOCK; /* Must be a block device */
 440
 441         /* Get the journal superblock */
 442         start = ext2fs_journal_sb_start(journal_dev->blocksize);
 443         if ((retval = io_channel_read_blk64(journal_dev->io, start,
 444                                             -SUPERBLOCK_SIZE,
 445                                             buf)))
 446                 return retval;
 447
 448         jsb = (journal_superblock_t *) buf;
 449         if ((jsb->s_header.h_magic != (unsigned) ntohl(JFS_MAGIC_NUMBER)) ||
 450             (jsb->s_header.h_blocktype != (unsigned) ntohl(JFS_SUPERBLOCK_V2)))
 451                 return EXT2_ET_NO_JOURNAL_SB;
 452
 453         if (ntohl(jsb->s_blocksize) != (unsigned long) fs->blocksize)
 454                 return EXT2_ET_UNEXPECTED_BLOCK_SIZE;
 455
 456         /* Check and see if this filesystem has already been added */
 457         nr_users = ntohl(jsb->s_nr_users);
 458         for (i=0; i < nr_users; i++) {
 459                 if (memcmp(fs->super->s_uuid,
 460                            &jsb->s_users[i*16], 16) == 0)
 461                         break;
 462         }
 463         if (i >= nr_users) {
 464                 memcpy(&jsb->s_users[nr_users*16],
 465                        fs->super->s_uuid, 16);
 466                 jsb->s_nr_users = htonl(nr_users+1);
 467         }
 468
 469         /* Writeback the journal superblock */
 470         if ((retval = io_channel_write_blk64(journal_dev->io, start,
 471                                             -SUPERBLOCK_SIZE, buf)))
 472                 return retval;
 473
 474         fs->super->s_journal_inum = 0;
 475         fs->super->s_journal_dev = st.st_rdev;
 476         memcpy(fs->super->s_journal_uuid, jsb->s_uuid,
 477                sizeof(fs->super->s_journal_uuid));
 478         memset(fs->super->s_jnl_blocks, 0, sizeof(fs->super->s_jnl_blocks));
 479         ext2fs_set_feature_journal(fs->super);
 480         ext2fs_mark_super_dirty(fs);
 481         return 0;
 482 }
 483
 484 /*
 485  * This function adds a journal inode to a filesystem, using either
 486  * POSIX routines if the filesystem is mounted, or using direct I/O
 487  * functions if it is not.
 488  */
 489 errcode_t ext2fs_add_journal_inode2(ext2_filsys fs, blk_t num_blocks,
 490                                     blk64_t goal, int flags)
 491 {
 492         errcode_t               retval;
 493         ext2_ino_t              journal_ino;
 494         struct stat             st;
 495         char                    jfile[1024];
 496         int                     mount_flags;
 497         int                     fd = -1;
 498
 499         if (flags & EXT2_MKJOURNAL_NO_MNT_CHECK)
 500                 mount_flags = 0;
 501         else if ((retval = ext2fs_check_mount_point(fs->device_name,
 502                                                     &mount_flags,
 503                                                     jfile, sizeof(jfile)-10)))
 504                 return retval;
 505
 506         if (mount_flags & EXT2_MF_MOUNTED) {
 507 #if HAVE_EXT2_IOCTLS
 508                 int f = 0;
 509 #endif
 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                         retval = ioctl(fd, EXT2_IOC_SETFLAGS, &f);
 523                         close(fd);
 524                         if (retval)
 525                                 return retval;
 526                 }
 527 #endif
 528 #endif
 529
 530                 /* Create the journal file */
 531                 if ((fd = open(jfile, O_CREAT|O_WRONLY, 0600)) < 0)
 532                         return errno;
 533
 534                 /* Note that we can't do lazy journal initialization for mounted
 535                  * filesystems, since the zero writing is also allocating the
 536                  * journal blocks.  We could use fallocate, but not all kernels
 537                  * support that, and creating a journal on a mounted ext2
 538                  * filesystems is extremely rare these days...  Ignore it. */
 539                 flags &= ~EXT2_MKJOURNAL_LAZYINIT;
 540
 541                 if ((retval = write_journal_file(fs, jfile, num_blocks, flags)))
 542                         goto errout;
 543
 544                 /* Get inode number of the journal file */
 545                 if (fstat(fd, &st) < 0) {
 546                         retval = errno;
 547                         goto errout;
 548                 }
 549
 550 #if defined(HAVE_CHFLAGS) && defined(UF_NODUMP)
 551                 retval = fchflags (fd, UF_NODUMP|UF_IMMUTABLE);
 552 #else
 553 #if HAVE_EXT2_IOCTLS
 554                 if (ioctl(fd, EXT2_IOC_GETFLAGS, &f) < 0) {
 555                         retval = errno;
 556                         goto errout;
 557                 }
 558                 f |= EXT2_NODUMP_FL | EXT2_IMMUTABLE_FL;
 559                 retval = ioctl(fd, EXT2_IOC_SETFLAGS, &f);
 560 #endif
 561 #endif
 562                 if (retval) {
 563                         retval = errno;
 564                         goto errout;
 565                 }
 566
 567                 if (close(fd) < 0) {
 568                         retval = errno;
 569                         fd = -1;
 570                         goto errout;
 571                 }
 572                 journal_ino = st.st_ino;
 573                 memset(fs->super->s_jnl_blocks, 0,
 574                        sizeof(fs->super->s_jnl_blocks));
 575         } else {
 576                 if ((mount_flags & EXT2_MF_BUSY) &&
 577                     !(fs->flags & EXT2_FLAG_EXCLUSIVE)) {
 578                         retval = EBUSY;
 579                         goto errout;
 580                 }
 581                 journal_ino = EXT2_JOURNAL_INO;
 582                 if ((retval = write_journal_inode(fs, journal_ino,
 583                                                   num_blocks, goal, flags)))
 584                         return retval;
 585         }
 586
 587         fs->super->s_journal_inum = journal_ino;
 588         fs->super->s_journal_dev = 0;
 589         memset(fs->super->s_journal_uuid, 0,
 590                sizeof(fs->super->s_journal_uuid));
 591         ext2fs_set_feature_journal(fs->super);
 592
 593         ext2fs_mark_super_dirty(fs);
 594         return 0;
 595 errout:
 596         if (fd >= 0)
 597                 close(fd);
 598         return retval;
 599 }
 600
 601 errcode_t ext2fs_add_journal_inode(ext2_filsys fs, blk_t num_blocks, int flags)
 602 {
 603         return ext2fs_add_journal_inode2(fs, num_blocks, ~0ULL, flags);
 604 }
 605
 606
 607 #ifdef DEBUG
 608 main(int argc, char **argv)
 609 {
 610         errcode_t       retval;
 611         char            *device_name;
 612         ext2_filsys     fs;
 613
 614         if (argc < 2) {
 615                 fprintf(stderr, "Usage: %s filesystem\n", argv[0]);
 616                 exit(1);
 617         }
 618         device_name = argv[1];
 619
 620         retval = ext2fs_open (device_name, EXT2_FLAG_RW, 0, 0,
 621                               unix_io_manager, &fs);
 622         if (retval) {
 623                 com_err(argv[0], retval, "while opening %s", device_name);
 624                 exit(1);
 625         }
 626
 627         retval = ext2fs_add_journal_inode(fs, JFS_MIN_JOURNAL_BLOCKS, 0);
 628         if (retval) {
 629                 com_err(argv[0], retval, "while adding journal to %s",
 630                         device_name);
 631                 exit(1);
 632         }
 633         retval = ext2fs_flush(fs);
 634         if (retval) {
 635                 printf("Warning, had trouble writing out superblocks.\n");
 636         }
 637         ext2fs_close_free(&fs);
 638         exit(0);
 639
 640 }
 641 #endif