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
  40 #include "kernel-jbd.h"
  41
  42 /*
  43  * This function automatically sets up the journal superblock and
  44  * returns it as an allocated block.
  45  */
  46 errcode_t ext2fs_create_journal_superblock(ext2_filsys fs,
  47                                            __u32 num_blocks, int flags,
  48                                            char  **ret_jsb)
  49 {
  50         errcode_t               retval;
  51         journal_superblock_t    *jsb;
  52
  53         if (num_blocks < JFS_MIN_JOURNAL_BLOCKS)
  54                 return EXT2_ET_JOURNAL_TOO_SMALL;
  55
  56         if ((retval = ext2fs_get_mem(fs->blocksize, &jsb)))
  57                 return retval;
  58
  59         memset (jsb, 0, fs->blocksize);
  60
  61         jsb->s_header.h_magic = htonl(JFS_MAGIC_NUMBER);
  62         if (flags & EXT2_MKJOURNAL_V1_SUPER)
  63                 jsb->s_header.h_blocktype = htonl(JFS_SUPERBLOCK_V1);
  64         else
  65                 jsb->s_header.h_blocktype = htonl(JFS_SUPERBLOCK_V2);
  66         jsb->s_blocksize = htonl(fs->blocksize);
  67         jsb->s_maxlen = htonl(num_blocks);
  68         jsb->s_nr_users = htonl(1);
  69         jsb->s_first = htonl(1);
  70         jsb->s_sequence = htonl(1);
  71         memcpy(jsb->s_uuid, fs->super->s_uuid, sizeof(fs->super->s_uuid));
  72         /*
  73          * If we're creating an external journal device, we need to
  74          * adjust these fields.
  75          */
  76         if (fs->super->s_feature_incompat &
  77             EXT3_FEATURE_INCOMPAT_JOURNAL_DEV) {
  78                 jsb->s_nr_users = 0;
  79                 jsb->s_first = htonl(ext2fs_journal_sb_start(fs->blocksize) + 1);
  80         }
  81
  82         *ret_jsb = (char *) jsb;
  83         return 0;
  84 }
  85
  86 /*
  87  * This function writes a journal using POSIX routines.  It is used
  88  * for creating external journals and creating journals on live
  89  * filesystems.
  90  */
  91 static errcode_t write_journal_file(ext2_filsys fs, char *filename,
  92                                     blk_t num_blocks, int flags)
  93 {
  94         errcode_t       retval;
  95         char            *buf = 0;
  96         int             fd, ret_size;
  97         blk_t           i;
  98
  99         if ((retval = ext2fs_create_journal_superblock(fs, num_blocks, flags,
 100                                                        &buf)))
 101                 return retval;
 102
 103         /* Open the device or journal file */
 104         if ((fd = open(filename, O_WRONLY)) < 0) {
 105                 retval = errno;
 106                 goto errfree;
 107         }
 108
 109         /* Write the superblock out */
 110         retval = EXT2_ET_SHORT_WRITE;
 111         ret_size = write(fd, buf, fs->blocksize);
 112         if (ret_size < 0) {
 113                 retval = errno;
 114                 goto errout;
 115         }
 116         if (ret_size != (int) fs->blocksize)
 117                 goto errout;
 118         memset(buf, 0, fs->blocksize);
 119
 120         if (flags & EXT2_MKJOURNAL_LAZYINIT)
 121                 goto success;
 122
 123         for (i = 1; i < num_blocks; i++) {
 124                 ret_size = write(fd, buf, fs->blocksize);
 125                 if (ret_size < 0) {
 126                         retval = errno;
 127                         goto errout;
 128                 }
 129                 if (ret_size != (int) fs->blocksize)
 130                         goto errout;
 131         }
 132
 133 success:
 134         retval = 0;
 135 errout:
 136         close(fd);
 137 errfree:
 138         ext2fs_free_mem(&buf);
 139         return retval;
 140 }
 141
 142 /*
 143  * Convenience function which zeros out _num_ blocks starting at
 144  * _blk_.  In case of an error, the details of the error is returned
 145  * via _ret_blk_ and _ret_count_ if they are non-NULL pointers.
 146  * Returns 0 on success, and an error code on an error.
 147  *
 148  * As a special case, if the first argument is NULL, then it will
 149  * attempt to free the static zeroizing buffer.  (This is to keep
 150  * programs that check for memory leaks happy.)
 151  */
 152 #define MAX_STRIDE_LENGTH (4194304 / (int) fs->blocksize)
 153 errcode_t ext2fs_zero_blocks2(ext2_filsys fs, blk64_t blk, int num,
 154                               blk64_t *ret_blk, int *ret_count)
 155 {
 156         int             j, count;
 157         static void     *buf;
 158         static int      stride_length;
 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
 170         /* Deal with zeroing less than 1 block */
 171         if (num <= 0)
 172                 return 0;
 173
 174         /* Try a zero out command, if supported */
 175         retval = io_channel_zeroout(fs->io, blk, num);
 176         if (retval == 0)
 177                 return 0;
 178
 179         /* Allocate the zeroizing buffer if necessary */
 180         if (num > stride_length && stride_length < MAX_STRIDE_LENGTH) {
 181                 void *p;
 182                 int new_stride = num;
 183
 184                 if (new_stride > MAX_STRIDE_LENGTH)
 185                         new_stride = MAX_STRIDE_LENGTH;
 186                 p = realloc(buf, fs->blocksize * new_stride);
 187                 if (!p)
 188                         return EXT2_ET_NO_MEMORY;
 189                 buf = p;
 190                 stride_length = new_stride;
 191                 memset(buf, 0, fs->blocksize * stride_length);
 192         }
 193         /* OK, do the write loop */
 194         j=0;
 195         while (j < num) {
 196                 if (blk % stride_length) {
 197                         count = stride_length - (blk % stride_length);
 198                         if (count > (num - j))
 199                                 count = num - j;
 200                 } else {
 201                         count = num - j;
 202                         if (count > stride_length)
 203                                 count = stride_length;
 204                 }
 205                 retval = io_channel_write_blk64(fs->io, blk, count, buf);
 206                 if (retval) {
 207                         if (ret_count)
 208                                 *ret_count = count;
 209                         if (ret_blk)
 210                                 *ret_blk = blk;
 211                         return retval;
 212                 }
 213                 j += count; blk += count;
 214         }
 215         return 0;
 216 }
 217
 218 errcode_t ext2fs_zero_blocks(ext2_filsys fs, blk_t blk, int num,
 219                              blk_t *ret_blk, int *ret_count)
 220 {
 221         blk64_t ret_blk2;
 222         errcode_t retval;
 223
 224         retval = ext2fs_zero_blocks2(fs, blk, num, &ret_blk2, ret_count);
 225         if (retval)
 226                 *ret_blk = (blk_t) ret_blk2;
 227         return retval;
 228 }
 229
 230 /*
 231  * Calculate the initial goal block to be roughly at the middle of the
 232  * filesystem.  Pick a group that has the largest number of free
 233  * blocks.
 234  */
 235 static blk64_t get_midpoint_journal_block(ext2_filsys fs)
 236 {
 237         dgrp_t  group, start, end, i, log_flex;
 238
 239         group = ext2fs_group_of_blk2(fs, (ext2fs_blocks_count(fs->super) -
 240                                          fs->super->s_first_data_block) / 2);
 241         log_flex = 1 << fs->super->s_log_groups_per_flex;
 242         if (fs->super->s_log_groups_per_flex && (group > log_flex)) {
 243                 group = group & ~(log_flex - 1);
 244                 while ((group < fs->group_desc_count) &&
 245                        ext2fs_bg_free_blocks_count(fs, group) == 0)
 246                         group++;
 247                 if (group == fs->group_desc_count)
 248                         group = 0;
 249                 start = group;
 250         } else
 251                 start = (group > 0) ? group-1 : group;
 252         end = ((group+1) < fs->group_desc_count) ? group+1 : group;
 253         group = start;
 254         for (i = start + 1; i <= end; i++)
 255                 if (ext2fs_bg_free_blocks_count(fs, i) >
 256                     ext2fs_bg_free_blocks_count(fs, group))
 257                         group = i;
 258         return ext2fs_group_first_block2(fs, group);
 259 }
 260
 261 /*
 262  * This function creates a journal using direct I/O routines.
 263  */
 264 static errcode_t write_journal_inode(ext2_filsys fs, ext2_ino_t journal_ino,
 265                                      blk_t num_blocks, blk64_t goal, int flags)
 266 {
 267         char                    *buf;
 268         errcode_t               retval;
 269         struct ext2_inode       inode;
 270         unsigned long long      inode_size;
 271         int                     falloc_flags = EXT2_FALLOCATE_FORCE_INIT;
 272         blk64_t                 zblk;
 273
 274         if ((retval = ext2fs_create_journal_superblock(fs, num_blocks, flags,
 275                                                        &buf)))
 276                 return retval;
 277
 278         if ((retval = ext2fs_read_bitmaps(fs)))
 279                 goto out2;
 280
 281         if ((retval = ext2fs_read_inode(fs, journal_ino, &inode)))
 282                 goto out2;
 283
 284         if (inode.i_blocks > 0) {
 285                 retval = EEXIST;
 286                 goto out2;
 287         }
 288
 289         if (goal == ~0ULL)
 290                 goal = get_midpoint_journal_block(fs);
 291
 292         if (fs->super->s_feature_incompat & EXT3_FEATURE_INCOMPAT_EXTENTS)
 293                 inode.i_flags |= EXT4_EXTENTS_FL;
 294
 295         if (!(flags & EXT2_MKJOURNAL_LAZYINIT))
 296                 falloc_flags |= EXT2_FALLOCATE_ZERO_BLOCKS;
 297
 298         inode_size = (unsigned long long)fs->blocksize * num_blocks;
 299         inode.i_mtime = inode.i_ctime = fs->now ? fs->now : time(0);
 300         inode.i_links_count = 1;
 301         inode.i_mode = LINUX_S_IFREG | 0600;
 302         retval = ext2fs_inode_size_set(fs, &inode, inode_size);
 303         if (retval)
 304                 goto out2;
 305
 306         retval = ext2fs_fallocate(fs, falloc_flags, journal_ino,
 307                                   &inode, goal, 0, num_blocks);
 308         if (retval)
 309                 goto out2;
 310
 311         if ((retval = ext2fs_write_new_inode(fs, journal_ino, &inode)))
 312                 goto out2;
 313
 314         retval = ext2fs_bmap2(fs, journal_ino, &inode, NULL, 0, 0, NULL, &zblk);
 315         if (retval)
 316                 goto out2;
 317
 318         retval = io_channel_write_blk64(fs->io, zblk, 1, buf);
 319         if (retval)
 320                 goto out2;
 321
 322         memcpy(fs->super->s_jnl_blocks, inode.i_block, EXT2_N_BLOCKS*4);
 323         fs->super->s_jnl_blocks[15] = inode.i_size_high;
 324         fs->super->s_jnl_blocks[16] = inode.i_size;
 325         fs->super->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
 326         ext2fs_mark_super_dirty(fs);
 327
 328 out2:
 329         ext2fs_free_mem(&buf);
 330         return retval;
 331 }
 332
 333 /*
 334  * Find a reasonable journal file size (in blocks) given the number of blocks
 335  * in the filesystem.  For very small filesystems, it is not reasonable to
 336  * have a journal that fills more than half of the filesystem.
 337  */
 338 int ext2fs_default_journal_size(__u64 num_blocks)
 339 {
 340         if (num_blocks < 2048)
 341                 return -1;
 342         if (num_blocks < 32768)
 343                 return (1024);
 344         if (num_blocks < 256*1024)
 345                 return (4096);
 346         if (num_blocks < 512*1024)
 347                 return (8192);
 348         if (num_blocks < 1024*1024)
 349                 return (16384);
 350         return 32768;
 351 }
 352
 353 int ext2fs_journal_sb_start(int blocksize)
 354 {
 355         if (blocksize == EXT2_MIN_BLOCK_SIZE)
 356                 return 2;
 357         return 1;
 358 }
 359
 360 /*
 361  * This function adds a journal device to a filesystem
 362  */
 363 errcode_t ext2fs_add_journal_device(ext2_filsys fs, ext2_filsys journal_dev)
 364 {
 365         struct stat     st;
 366         errcode_t       retval;
 367         char            buf[SUPERBLOCK_SIZE];
 368         journal_superblock_t    *jsb;
 369         int             start;
 370         __u32           i, nr_users;
 371
 372         /* Make sure the device exists and is a block device */
 373         if (stat(journal_dev->device_name, &st) < 0)
 374                 return errno;
 375
 376         if (!S_ISBLK(st.st_mode))
 377                 return EXT2_ET_JOURNAL_NOT_BLOCK; /* Must be a block device */
 378
 379         /* Get the journal superblock */
 380         start = ext2fs_journal_sb_start(journal_dev->blocksize);
 381         if ((retval = io_channel_read_blk64(journal_dev->io, start,
 382                                             -SUPERBLOCK_SIZE,
 383                                             buf)))
 384                 return retval;
 385
 386         jsb = (journal_superblock_t *) buf;
 387         if ((jsb->s_header.h_magic != (unsigned) ntohl(JFS_MAGIC_NUMBER)) ||
 388             (jsb->s_header.h_blocktype != (unsigned) ntohl(JFS_SUPERBLOCK_V2)))
 389                 return EXT2_ET_NO_JOURNAL_SB;
 390
 391         if (ntohl(jsb->s_blocksize) != (unsigned long) fs->blocksize)
 392                 return EXT2_ET_UNEXPECTED_BLOCK_SIZE;
 393
 394         /* Check and see if this filesystem has already been added */
 395         nr_users = ntohl(jsb->s_nr_users);
 396         for (i=0; i < nr_users; i++) {
 397                 if (memcmp(fs->super->s_uuid,
 398                            &jsb->s_users[i*16], 16) == 0)
 399                         break;
 400         }
 401         if (i >= nr_users) {
 402                 memcpy(&jsb->s_users[nr_users*16],
 403                        fs->super->s_uuid, 16);
 404                 jsb->s_nr_users = htonl(nr_users+1);
 405         }
 406
 407         /* Writeback the journal superblock */
 408         if ((retval = io_channel_write_blk64(journal_dev->io, start,
 409                                             -SUPERBLOCK_SIZE, buf)))
 410                 return retval;
 411
 412         fs->super->s_journal_inum = 0;
 413         fs->super->s_journal_dev = st.st_rdev;
 414         memcpy(fs->super->s_journal_uuid, jsb->s_uuid,
 415                sizeof(fs->super->s_journal_uuid));
 416         memset(fs->super->s_jnl_blocks, 0, sizeof(fs->super->s_jnl_blocks));
 417         fs->super->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 418         ext2fs_mark_super_dirty(fs);
 419         return 0;
 420 }
 421
 422 /*
 423  * This function adds a journal inode to a filesystem, using either
 424  * POSIX routines if the filesystem is mounted, or using direct I/O
 425  * functions if it is not.
 426  */
 427 errcode_t ext2fs_add_journal_inode2(ext2_filsys fs, blk_t num_blocks,
 428                                     blk64_t goal, int flags)
 429 {
 430         errcode_t               retval;
 431         ext2_ino_t              journal_ino;
 432         struct stat             st;
 433         char                    jfile[1024];
 434         int                     mount_flags;
 435         int                     fd = -1;
 436
 437         if (flags & EXT2_MKJOURNAL_NO_MNT_CHECK)
 438                 mount_flags = 0;
 439         else if ((retval = ext2fs_check_mount_point(fs->device_name,
 440                                                     &mount_flags,
 441                                                     jfile, sizeof(jfile)-10)))
 442                 return retval;
 443
 444         if (mount_flags & EXT2_MF_MOUNTED) {
 445 #if HAVE_EXT2_IOCTLS
 446                 int f = 0;
 447 #endif
 448                 strcat(jfile, "/.journal");
 449
 450                 /*
 451                  * If .../.journal already exists, make sure any
 452                  * immutable or append-only flags are cleared.
 453                  */
 454 #if defined(HAVE_CHFLAGS) && defined(UF_NODUMP)
 455                 (void) chflags (jfile, 0);
 456 #else
 457 #if HAVE_EXT2_IOCTLS
 458                 fd = open(jfile, O_RDONLY);
 459                 if (fd >= 0) {
 460                         retval = ioctl(fd, EXT2_IOC_SETFLAGS, &f);
 461                         close(fd);
 462                         if (retval)
 463                                 return retval;
 464                 }
 465 #endif
 466 #endif
 467
 468                 /* Create the journal file */
 469                 if ((fd = open(jfile, O_CREAT|O_WRONLY, 0600)) < 0)
 470                         return errno;
 471
 472                 /* Note that we can't do lazy journal initialization for mounted
 473                  * filesystems, since the zero writing is also allocating the
 474                  * journal blocks.  We could use fallocate, but not all kernels
 475                  * support that, and creating a journal on a mounted ext2
 476                  * filesystems is extremely rare these days...  Ignore it. */
 477                 flags &= ~EXT2_MKJOURNAL_LAZYINIT;
 478
 479                 if ((retval = write_journal_file(fs, jfile, num_blocks, flags)))
 480                         goto errout;
 481
 482                 /* Get inode number of the journal file */
 483                 if (fstat(fd, &st) < 0) {
 484                         retval = errno;
 485                         goto errout;
 486                 }
 487
 488 #if defined(HAVE_CHFLAGS) && defined(UF_NODUMP)
 489                 retval = fchflags (fd, UF_NODUMP|UF_IMMUTABLE);
 490 #else
 491 #if HAVE_EXT2_IOCTLS
 492                 if (ioctl(fd, EXT2_IOC_GETFLAGS, &f) < 0) {
 493                         retval = errno;
 494                         goto errout;
 495                 }
 496                 f |= EXT2_NODUMP_FL | EXT2_IMMUTABLE_FL;
 497                 retval = ioctl(fd, EXT2_IOC_SETFLAGS, &f);
 498 #endif
 499 #endif
 500                 if (retval) {
 501                         retval = errno;
 502                         goto errout;
 503                 }
 504
 505                 if (close(fd) < 0) {
 506                         retval = errno;
 507                         fd = -1;
 508                         goto errout;
 509                 }
 510                 journal_ino = st.st_ino;
 511                 memset(fs->super->s_jnl_blocks, 0,
 512                        sizeof(fs->super->s_jnl_blocks));
 513         } else {
 514                 if ((mount_flags & EXT2_MF_BUSY) &&
 515                     !(fs->flags & EXT2_FLAG_EXCLUSIVE)) {
 516                         retval = EBUSY;
 517                         goto errout;
 518                 }
 519                 journal_ino = EXT2_JOURNAL_INO;
 520                 if ((retval = write_journal_inode(fs, journal_ino,
 521                                                   num_blocks, goal, flags)))
 522                         return retval;
 523         }
 524
 525         fs->super->s_journal_inum = journal_ino;
 526         fs->super->s_journal_dev = 0;
 527         memset(fs->super->s_journal_uuid, 0,
 528                sizeof(fs->super->s_journal_uuid));
 529         fs->super->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 530
 531         ext2fs_mark_super_dirty(fs);
 532         return 0;
 533 errout:
 534         if (fd >= 0)
 535                 close(fd);
 536         return retval;
 537 }
 538
 539 errcode_t ext2fs_add_journal_inode(ext2_filsys fs, blk_t num_blocks, int flags)
 540 {
 541         return ext2fs_add_journal_inode2(fs, num_blocks, ~0ULL, flags);
 542 }
 543
 544
 545 #ifdef DEBUG
 546 main(int argc, char **argv)
 547 {
 548         errcode_t       retval;
 549         char            *device_name;
 550         ext2_filsys     fs;
 551
 552         if (argc < 2) {
 553                 fprintf(stderr, "Usage: %s filesystem\n", argv[0]);
 554                 exit(1);
 555         }
 556         device_name = argv[1];
 557
 558         retval = ext2fs_open (device_name, EXT2_FLAG_RW, 0, 0,
 559                               unix_io_manager, &fs);
 560         if (retval) {
 561                 com_err(argv[0], retval, "while opening %s", device_name);
 562                 exit(1);
 563         }
 564
 565         retval = ext2fs_add_journal_inode(fs, JFS_MIN_JOURNAL_BLOCKS, 0);
 566         if (retval) {
 567                 com_err(argv[0], retval, "while adding journal to %s",
 568                         device_name);
 569                 exit(1);
 570         }
 571         retval = ext2fs_flush(fs);
 572         if (retval) {
 573                 printf("Warning, had trouble writing out superblocks.\n");
 574         }
 575         ext2fs_close_free(&fs);
 576         exit(0);
 577
 578 }
 579 #endif