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.
27 #include <sys/types.h>
30 #include <sys/ioctl.h>
33 #include <netinet/in.h>
42 * This function automatically sets up the journal superblock and
43 * returns it as an allocated block.
45 errcode_t ext2fs_create_journal_superblock(ext2_filsys fs,
46 __u32 num_blocks, int flags,
50 journal_superblock_t *jsb;
52 if (num_blocks < 1024)
53 return EXT2_ET_JOURNAL_TOO_SMALL;
55 if ((retval = ext2fs_get_mem(fs->blocksize, &jsb)))
58 memset (jsb, 0, fs->blocksize);
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);
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));
72 * If we're creating an external journal device, we need to
73 * adjust these fields.
75 if (fs->super->s_feature_incompat &
76 EXT3_FEATURE_INCOMPAT_JOURNAL_DEV) {
78 if (fs->blocksize == 1024)
79 jsb->s_first = htonl(3);
81 jsb->s_first = htonl(2);
84 *ret_jsb = (char *) jsb;
89 * This function writes a journal using POSIX routines. It is used
90 * for creating external journals and creating journals on live
93 static errcode_t write_journal_file(ext2_filsys fs, char *filename,
94 blk_t num_blocks, int flags)
101 if ((retval = ext2fs_create_journal_superblock(fs, num_blocks, flags,
105 /* Open the device or journal file */
106 if ((fd = open(filename, O_WRONLY)) < 0) {
111 /* Write the superblock out */
112 retval = EXT2_ET_SHORT_WRITE;
113 ret_size = write(fd, buf, fs->blocksize);
118 if (ret_size != (int) fs->blocksize)
120 memset(buf, 0, fs->blocksize);
122 if (flags & EXT2_MKJOURNAL_LAZYINIT)
125 for (i = 1; i < num_blocks; i++) {
126 ret_size = write(fd, buf, fs->blocksize);
131 if (ret_size != (int) fs->blocksize)
140 ext2fs_free_mem(&buf);
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.
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.)
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)
162 /* If fs is null, clean up the static buffer and return */
170 /* Allocate the zeroizing buffer if necessary */
172 buf = malloc(fs->blocksize * STRIDE_LENGTH);
175 memset(buf, 0, fs->blocksize * STRIDE_LENGTH);
177 /* OK, do the write loop */
180 if (blk % STRIDE_LENGTH) {
181 count = STRIDE_LENGTH - (blk % STRIDE_LENGTH);
182 if (count > (num - j))
186 if (count > STRIDE_LENGTH)
187 count = STRIDE_LENGTH;
189 retval = io_channel_write_blk64(fs->io, blk, count, buf);
197 j += count; blk += count;
202 errcode_t ext2fs_zero_blocks(ext2_filsys fs, blk_t blk, int num,
203 blk_t *ret_blk, int *ret_count)
208 retval = ext2fs_zero_blocks2(fs, blk, num, &ret_blk2, ret_count);
210 *ret_blk = (blk_t) ret_blk2;
215 * Helper function for creating the journal using direct I/O routines
217 struct mkjournal_struct {
228 static int mkjournal_proc(ext2_filsys fs,
230 e2_blkcnt_t blockcnt,
231 blk64_t ref_block EXT2FS_ATTR((unused)),
232 int ref_offset EXT2FS_ATTR((unused)),
235 struct mkjournal_struct *es = (struct mkjournal_struct *) priv_data;
244 (EXT2FS_B2C(fs, es->goal) == EXT2FS_B2C(fs, es->goal+1)))
245 new_blk = es->goal+1;
247 es->goal &= ~EXT2FS_CLUSTER_MASK(fs);
248 retval = ext2fs_new_block2(fs, es->goal, 0, &new_blk);
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))
267 retval = ext2fs_zero_blocks2(fs,
274 if (es->zero_count == 0) {
275 es->blk_to_zero = new_blk;
281 memset(es->buf, 0, fs->blocksize);
287 *blocknr = es->goal = new_blk;
288 ext2fs_block_alloc_stats2(fs, new_blk, +1);
290 if (es->num_blocks == 0)
291 return (BLOCK_CHANGED | BLOCK_ABORT);
293 return BLOCK_CHANGED;
298 * This function creates a journal using direct I/O routines.
300 static errcode_t write_journal_inode(ext2_filsys fs, ext2_ino_t journal_ino,
301 blk_t num_blocks, int flags)
304 dgrp_t group, start, end, i, log_flex;
306 struct ext2_inode inode;
307 unsigned long long inode_size;
308 struct mkjournal_struct es;
310 if ((retval = ext2fs_create_journal_superblock(fs, num_blocks, flags,
314 if ((retval = ext2fs_read_bitmaps(fs)))
317 if ((retval = ext2fs_read_inode(fs, journal_ino, &inode)))
320 if (inode.i_blocks > 0)
323 es.num_blocks = num_blocks;
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)))
337 * Set the initial goal block to be roughly at the middle of
338 * the filesystem. Pick a group that has the largest number
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)
349 if (group == fs->group_desc_count)
353 start = (group > 0) ? group-1 : group;
354 end = ((group+1) < fs->group_desc_count) ? group+1 : group;
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))
361 es.goal = ext2fs_group_first_block2(fs, group);
362 retval = ext2fs_block_iterate3(fs, journal_ino, BLOCK_FLAG_APPEND,
363 0, mkjournal_proc, &es);
369 retval = ext2fs_zero_blocks2(fs, es.blk_to_zero,
370 es.zero_count, 0, 0);
375 if ((retval = ext2fs_read_inode(fs, journal_ino, &inode)))
378 inode_size = (unsigned long long)fs->blocksize * num_blocks;
379 inode.i_size = inode_size & 0xFFFFFFFF;
380 inode.i_size_high = (inode_size >> 32) & 0xFFFFFFFF;
381 if (ext2fs_needs_large_file_feature(inode_size))
382 fs->super->s_feature_ro_compat |=
383 EXT2_FEATURE_RO_COMPAT_LARGE_FILE;
384 ext2fs_iblk_add_blocks(fs, &inode, es.newblocks);
385 inode.i_mtime = inode.i_ctime = fs->now ? fs->now : time(0);
386 inode.i_links_count = 1;
387 inode.i_mode = LINUX_S_IFREG | 0600;
389 if ((retval = ext2fs_write_new_inode(fs, journal_ino, &inode)))
393 memcpy(fs->super->s_jnl_blocks, inode.i_block, EXT2_N_BLOCKS*4);
394 fs->super->s_jnl_blocks[15] = inode.i_size_high;
395 fs->super->s_jnl_blocks[16] = inode.i_size;
396 fs->super->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
397 ext2fs_mark_super_dirty(fs);
400 ext2fs_zero_blocks2(0, 0, 0, 0, 0);
401 ext2fs_free_mem(&buf);
406 * Find a reasonable journal file size (in blocks) given the number of blocks
407 * in the filesystem. For very small filesystems, it is not reasonable to
408 * have a journal that fills more than half of the filesystem.
410 int ext2fs_default_journal_size(__u64 num_blocks)
412 if (num_blocks < 2048)
414 if (num_blocks < 32768)
416 if (num_blocks < 256*1024)
418 if (num_blocks < 512*1024)
420 if (num_blocks < 1024*1024)
426 * This function adds a journal device to a filesystem
428 errcode_t ext2fs_add_journal_device(ext2_filsys fs, ext2_filsys journal_dev)
433 journal_superblock_t *jsb;
437 /* Make sure the device exists and is a block device */
438 if (stat(journal_dev->device_name, &st) < 0)
441 if (!S_ISBLK(st.st_mode))
442 return EXT2_ET_JOURNAL_NOT_BLOCK; /* Must be a block device */
444 /* Get the journal superblock */
446 if (journal_dev->blocksize == 1024)
448 if ((retval = io_channel_read_blk64(journal_dev->io, start, -1024,
452 jsb = (journal_superblock_t *) buf;
453 if ((jsb->s_header.h_magic != (unsigned) ntohl(JFS_MAGIC_NUMBER)) ||
454 (jsb->s_header.h_blocktype != (unsigned) ntohl(JFS_SUPERBLOCK_V2)))
455 return EXT2_ET_NO_JOURNAL_SB;
457 if (ntohl(jsb->s_blocksize) != (unsigned long) fs->blocksize)
458 return EXT2_ET_UNEXPECTED_BLOCK_SIZE;
460 /* Check and see if this filesystem has already been added */
461 nr_users = ntohl(jsb->s_nr_users);
462 for (i=0; i < nr_users; i++) {
463 if (memcmp(fs->super->s_uuid,
464 &jsb->s_users[i*16], 16) == 0)
468 memcpy(&jsb->s_users[nr_users*16],
469 fs->super->s_uuid, 16);
470 jsb->s_nr_users = htonl(nr_users+1);
473 /* Writeback the journal superblock */
474 if ((retval = io_channel_write_blk64(journal_dev->io, start, -1024, buf)))
477 fs->super->s_journal_inum = 0;
478 fs->super->s_journal_dev = st.st_rdev;
479 memcpy(fs->super->s_journal_uuid, jsb->s_uuid,
480 sizeof(fs->super->s_journal_uuid));
481 fs->super->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
482 ext2fs_mark_super_dirty(fs);
487 * This function adds a journal inode to a filesystem, using either
488 * POSIX routines if the filesystem is mounted, or using direct I/O
489 * functions if it is not.
491 errcode_t ext2fs_add_journal_inode(ext2_filsys fs, blk_t num_blocks, int flags)
494 ext2_ino_t journal_ino;
500 if (flags & EXT2_MKJOURNAL_NO_MNT_CHECK)
502 else if ((retval = ext2fs_check_mount_point(fs->device_name,
504 jfile, sizeof(jfile)-10)))
507 if (mount_flags & EXT2_MF_MOUNTED) {
511 strcat(jfile, "/.journal");
514 * If .../.journal already exists, make sure any
515 * immutable or append-only flags are cleared.
517 #if defined(HAVE_CHFLAGS) && defined(UF_NODUMP)
518 (void) chflags (jfile, 0);
521 fd = open(jfile, O_RDONLY);
523 ioctl(fd, EXT2_IOC_SETFLAGS, &f);
529 /* Create the journal file */
530 if ((fd = open(jfile, O_CREAT|O_WRONLY, 0600)) < 0)
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;
540 if ((retval = write_journal_file(fs, jfile, num_blocks, flags)))
543 /* Get inode number of the journal file */
544 if (fstat(fd, &st) < 0) {
549 #if defined(HAVE_CHFLAGS) && defined(UF_NODUMP)
550 retval = fchflags (fd, UF_NODUMP|UF_IMMUTABLE);
553 if (ioctl(fd, EXT2_IOC_GETFLAGS, &f) < 0) {
557 f |= EXT2_NODUMP_FL | EXT2_IMMUTABLE_FL;
558 retval = ioctl(fd, EXT2_IOC_SETFLAGS, &f);
571 journal_ino = st.st_ino;
573 if ((mount_flags & EXT2_MF_BUSY) &&
574 !(fs->flags & EXT2_FLAG_EXCLUSIVE)) {
578 journal_ino = EXT2_JOURNAL_INO;
579 if ((retval = write_journal_inode(fs, journal_ino,
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;
590 ext2fs_mark_super_dirty(fs);
599 main(int argc, char **argv)
606 fprintf(stderr, "Usage: %s filesystem\n", argv[0]);
609 device_name = argv[1];
611 retval = ext2fs_open (device_name, EXT2_FLAG_RW, 0, 0,
612 unix_io_manager, &fs);
614 com_err(argv[0], retval, "while opening %s", device_name);
618 retval = ext2fs_add_journal_inode(fs, 1024, 0);
620 com_err(argv[0], retval, "while adding journal to %s",
624 retval = ext2fs_flush(fs);
626 printf("Warning, had trouble writing out superblocks.\n");