2 * mkjournal.c --- make a journal for a filesystem
4 * Copyright (C) 2000 Theodore Ts'o.
7 * This file may be redistributed under the terms of the GNU Library
8 * General Public License, version 2.
26 #include <sys/types.h>
29 #include <sys/ioctl.h>
32 #include <netinet/in.h>
41 * This function automatically sets up the journal superblock and
42 * returns it as an allocated block.
44 errcode_t ext2fs_create_journal_superblock(ext2_filsys fs,
45 __u32 num_blocks, int flags,
49 journal_superblock_t *jsb;
51 if (num_blocks < 1024)
52 return EXT2_ET_JOURNAL_TOO_SMALL;
54 if ((retval = ext2fs_get_mem(fs->blocksize, &jsb)))
57 memset (jsb, 0, fs->blocksize);
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);
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));
71 * If we're creating an external journal device, we need to
72 * adjust these fields.
74 if (fs->super->s_feature_incompat &
75 EXT3_FEATURE_INCOMPAT_JOURNAL_DEV) {
77 if (fs->blocksize == 1024)
78 jsb->s_first = htonl(3);
80 jsb->s_first = htonl(2);
83 *ret_jsb = (char *) jsb;
88 * This function writes a journal using POSIX routines. It is used
89 * for creating external journals and creating journals on live
92 static errcode_t write_journal_file(ext2_filsys fs, char *filename,
93 blk_t num_blocks, int flags)
100 if ((retval = ext2fs_create_journal_superblock(fs, num_blocks, flags,
104 /* Open the device or journal file */
105 if ((fd = open(filename, O_WRONLY)) < 0) {
110 /* Write the superblock out */
111 retval = EXT2_ET_SHORT_WRITE;
112 ret_size = write(fd, buf, fs->blocksize);
117 if (ret_size != (int) fs->blocksize)
119 memset(buf, 0, fs->blocksize);
121 if (flags & EXT2_MKJOURNAL_LAZYINIT)
124 for (i = 1; i < num_blocks; i++) {
125 ret_size = write(fd, buf, fs->blocksize);
130 if (ret_size != (int) fs->blocksize)
139 ext2fs_free_mem(&buf);
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.
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.)
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)
161 /* If fs is null, clean up the static buffer and return */
169 /* Allocate the zeroizing buffer if necessary */
171 buf = malloc(fs->blocksize * STRIDE_LENGTH);
174 memset(buf, 0, fs->blocksize * STRIDE_LENGTH);
176 /* OK, do the write loop */
179 if (blk % STRIDE_LENGTH) {
180 count = STRIDE_LENGTH - (blk % STRIDE_LENGTH);
181 if (count > (num - j))
185 if (count > STRIDE_LENGTH)
186 count = STRIDE_LENGTH;
188 retval = io_channel_write_blk64(fs->io, blk, count, buf);
196 j += count; blk += count;
201 errcode_t ext2fs_zero_blocks(ext2_filsys fs, blk_t blk, int num,
202 blk_t *ret_blk, int *ret_count)
207 retval = ext2fs_zero_blocks2(fs, blk, num, &ret_blk2, ret_count);
209 *ret_blk = (blk_t) ret_blk2;
214 * Helper function for creating the journal using direct I/O routines
216 struct mkjournal_struct {
227 static int mkjournal_proc(ext2_filsys fs,
229 e2_blkcnt_t blockcnt,
230 blk64_t ref_block EXT2FS_ATTR((unused)),
231 int ref_offset EXT2FS_ATTR((unused)),
234 struct mkjournal_struct *es = (struct mkjournal_struct *) priv_data;
243 (EXT2FS_B2C(fs, es->goal) == EXT2FS_B2C(fs, es->goal+1)))
244 new_blk = es->goal+1;
246 es->goal &= ~EXT2FS_CLUSTER_MASK(fs);
247 retval = ext2fs_new_block2(fs, es->goal, 0, &new_blk);
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))
266 retval = ext2fs_zero_blocks2(fs,
273 if (es->zero_count == 0) {
274 es->blk_to_zero = new_blk;
280 memset(es->buf, 0, fs->blocksize);
286 *blocknr = es->goal = new_blk;
287 ext2fs_block_alloc_stats2(fs, new_blk, +1);
289 if (es->num_blocks == 0)
290 return (BLOCK_CHANGED | BLOCK_ABORT);
292 return BLOCK_CHANGED;
297 * This function creates a journal using direct I/O routines.
299 static errcode_t write_journal_inode(ext2_filsys fs, ext2_ino_t journal_ino,
300 blk_t num_blocks, int flags)
303 dgrp_t group, start, end, i, log_flex;
305 struct ext2_inode inode;
306 unsigned long long inode_size;
307 struct mkjournal_struct es;
309 if ((retval = ext2fs_create_journal_superblock(fs, num_blocks, flags,
313 if ((retval = ext2fs_read_bitmaps(fs)))
316 if ((retval = ext2fs_read_inode(fs, journal_ino, &inode)))
319 if (inode.i_blocks > 0)
322 es.num_blocks = num_blocks;
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)))
336 * Set the initial goal block to be roughly at the middle of
337 * the filesystem. Pick a group that has the largest number
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)
348 if (group == fs->group_desc_count)
352 start = (group > 0) ? group-1 : group;
353 end = ((group+1) < fs->group_desc_count) ? group+1 : group;
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))
360 es.goal = (fs->super->s_blocks_per_group * group) +
361 fs->super->s_first_data_block;
363 retval = ext2fs_block_iterate3(fs, journal_ino, BLOCK_FLAG_APPEND,
364 0, mkjournal_proc, &es);
370 retval = ext2fs_zero_blocks2(fs, es.blk_to_zero,
371 es.zero_count, 0, 0);
376 if ((retval = ext2fs_read_inode(fs, journal_ino, &inode)))
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;
390 if ((retval = ext2fs_write_new_inode(fs, journal_ino, &inode)))
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);
401 ext2fs_zero_blocks2(0, 0, 0, 0, 0);
402 ext2fs_free_mem(&buf);
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.
411 int ext2fs_default_journal_size(__u64 num_blocks)
413 if (num_blocks < 2048)
415 if (num_blocks < 32768)
417 if (num_blocks < 256*1024)
419 if (num_blocks < 512*1024)
421 if (num_blocks < 1024*1024)
427 * This function adds a journal device to a filesystem
429 errcode_t ext2fs_add_journal_device(ext2_filsys fs, ext2_filsys journal_dev)
434 journal_superblock_t *jsb;
438 /* Make sure the device exists and is a block device */
439 if (stat(journal_dev->device_name, &st) < 0)
442 if (!S_ISBLK(st.st_mode))
443 return EXT2_ET_JOURNAL_NOT_BLOCK; /* Must be a block device */
445 /* Get the journal superblock */
447 if (journal_dev->blocksize == 1024)
449 if ((retval = io_channel_read_blk64(journal_dev->io, start, -1024,
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;
458 if (ntohl(jsb->s_blocksize) != (unsigned long) fs->blocksize)
459 return EXT2_ET_UNEXPECTED_BLOCK_SIZE;
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)
469 memcpy(&jsb->s_users[nr_users*16],
470 fs->super->s_uuid, 16);
471 jsb->s_nr_users = htonl(nr_users+1);
474 /* Writeback the journal superblock */
475 if ((retval = io_channel_write_blk64(journal_dev->io, start, -1024, buf)))
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);
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.
492 errcode_t ext2fs_add_journal_inode(ext2_filsys fs, blk_t num_blocks, int flags)
495 ext2_ino_t journal_ino;
501 if ((retval = ext2fs_check_mount_point(fs->device_name, &mount_flags,
502 jfile, sizeof(jfile)-10)))
505 if (mount_flags & EXT2_MF_MOUNTED) {
506 strcat(jfile, "/.journal");
509 * If .../.journal already exists, make sure any
510 * immutable or append-only flags are cleared.
512 #if defined(HAVE_CHFLAGS) && defined(UF_NODUMP)
513 (void) chflags (jfile, 0);
516 fd = open(jfile, O_RDONLY);
519 ioctl(fd, EXT2_IOC_SETFLAGS, &f);
525 /* Create the journal file */
526 if ((fd = open(jfile, O_CREAT|O_WRONLY, 0600)) < 0)
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;
536 if ((retval = write_journal_file(fs, jfile, num_blocks, flags)))
539 /* Get inode number of the journal file */
540 if (fstat(fd, &st) < 0) {
545 #if defined(HAVE_CHFLAGS) && defined(UF_NODUMP)
546 retval = fchflags (fd, UF_NODUMP|UF_IMMUTABLE);
549 if (ioctl(fd, EXT2_IOC_GETFLAGS, &f) < 0) {
553 f |= EXT2_NODUMP_FL | EXT2_IMMUTABLE_FL;
554 retval = ioctl(fd, EXT2_IOC_SETFLAGS, &f);
567 journal_ino = st.st_ino;
569 if ((mount_flags & EXT2_MF_BUSY) &&
570 !(fs->flags & EXT2_FLAG_EXCLUSIVE)) {
574 journal_ino = EXT2_JOURNAL_INO;
575 if ((retval = write_journal_inode(fs, journal_ino,
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;
586 ext2fs_mark_super_dirty(fs);
595 main(int argc, char **argv)
602 fprintf(stderr, "Usage: %s filesystem\n", argv[0]);
605 device_name = argv[1];
607 retval = ext2fs_open (device_name, EXT2_FLAG_RW, 0, 0,
608 unix_io_manager, &fs);
610 com_err(argv[0], retval, "while opening %s", device_name);
614 retval = ext2fs_add_journal_inode(fs, 1024);
616 com_err(argv[0], retval, "while adding journal to %s",
620 retval = ext2fs_flush(fs);
622 printf("Warning, had trouble writing out superblocks.\n");