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 < JFS_MIN_JOURNAL_BLOCKS)
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 jsb->s_first = htonl(ext2fs_journal_sb_start(fs->blocksize) + 1);
81 *ret_jsb = (char *) jsb;
86 * This function writes a journal using POSIX routines. It is used
87 * for creating external journals and creating journals on live
90 static errcode_t write_journal_file(ext2_filsys fs, char *filename,
91 blk_t num_blocks, int flags)
98 if ((retval = ext2fs_create_journal_superblock(fs, num_blocks, flags,
102 /* Open the device or journal file */
103 if ((fd = open(filename, O_WRONLY)) < 0) {
108 /* Write the superblock out */
109 retval = EXT2_ET_SHORT_WRITE;
110 ret_size = write(fd, buf, fs->blocksize);
115 if (ret_size != (int) fs->blocksize)
117 memset(buf, 0, fs->blocksize);
119 if (flags & EXT2_MKJOURNAL_LAZYINIT)
122 for (i = 1; i < num_blocks; i++) {
123 ret_size = write(fd, buf, fs->blocksize);
128 if (ret_size != (int) fs->blocksize)
137 ext2fs_free_mem(&buf);
142 * Convenience function which zeros out _num_ blocks starting at
143 * _blk_. In case of an error, the details of the error is returned
144 * via _ret_blk_ and _ret_count_ if they are non-NULL pointers.
145 * Returns 0 on success, and an error code on an error.
147 * As a special case, if the first argument is NULL, then it will
148 * attempt to free the static zeroizing buffer. (This is to keep
149 * programs that check for memory leaks happy.)
151 #define STRIDE_LENGTH 8
152 errcode_t ext2fs_zero_blocks2(ext2_filsys fs, blk64_t blk, int num,
153 blk64_t *ret_blk, int *ret_count)
159 /* If fs is null, clean up the static buffer and return */
167 /* Allocate the zeroizing buffer if necessary */
169 buf = malloc(fs->blocksize * STRIDE_LENGTH);
172 memset(buf, 0, fs->blocksize * STRIDE_LENGTH);
174 /* OK, do the write loop */
177 if (blk % STRIDE_LENGTH) {
178 count = STRIDE_LENGTH - (blk % STRIDE_LENGTH);
179 if (count > (num - j))
183 if (count > STRIDE_LENGTH)
184 count = STRIDE_LENGTH;
186 retval = io_channel_write_blk64(fs->io, blk, count, buf);
194 j += count; blk += count;
199 errcode_t ext2fs_zero_blocks(ext2_filsys fs, blk_t blk, int num,
200 blk_t *ret_blk, int *ret_count)
205 retval = ext2fs_zero_blocks2(fs, blk, num, &ret_blk2, ret_count);
207 *ret_blk = (blk_t) ret_blk2;
212 * Helper function for creating the journal using direct I/O routines
214 struct mkjournal_struct {
225 static int mkjournal_proc(ext2_filsys fs,
227 e2_blkcnt_t blockcnt,
228 blk64_t ref_block EXT2FS_ATTR((unused)),
229 int ref_offset EXT2FS_ATTR((unused)),
232 struct mkjournal_struct *es = (struct mkjournal_struct *) priv_data;
241 (EXT2FS_B2C(fs, es->goal) == EXT2FS_B2C(fs, es->goal+1)))
242 new_blk = es->goal+1;
244 es->goal &= ~EXT2FS_CLUSTER_MASK(fs);
245 retval = ext2fs_new_block2(fs, es->goal, 0, &new_blk);
250 ext2fs_block_alloc_stats2(fs, new_blk, +1);
258 retval = io_channel_write_blk64(fs->io, new_blk, 1, es->buf);
259 else if (!(es->flags & EXT2_MKJOURNAL_LAZYINIT)) {
260 if (es->zero_count) {
261 if ((es->blk_to_zero + es->zero_count == new_blk) &&
262 (es->zero_count < 1024))
265 retval = ext2fs_zero_blocks2(fs,
272 if (es->zero_count == 0) {
273 es->blk_to_zero = new_blk;
279 memset(es->buf, 0, fs->blocksize);
285 *blocknr = es->goal = new_blk;
287 if (es->num_blocks == 0)
288 return (BLOCK_CHANGED | BLOCK_ABORT);
290 return BLOCK_CHANGED;
295 * Calculate the initial goal block to be roughly at the middle of the
296 * filesystem. Pick a group that has the largest number of free
299 static blk64_t get_midpoint_journal_block(ext2_filsys fs)
301 dgrp_t group, start, end, i, log_flex;
303 group = ext2fs_group_of_blk2(fs, (ext2fs_blocks_count(fs->super) -
304 fs->super->s_first_data_block) / 2);
305 log_flex = 1 << fs->super->s_log_groups_per_flex;
306 if (fs->super->s_log_groups_per_flex && (group > log_flex)) {
307 group = group & ~(log_flex - 1);
308 while ((group < fs->group_desc_count) &&
309 ext2fs_bg_free_blocks_count(fs, group) == 0)
311 if (group == fs->group_desc_count)
315 start = (group > 0) ? group-1 : group;
316 end = ((group+1) < fs->group_desc_count) ? group+1 : group;
318 for (i = start + 1; i <= end; i++)
319 if (ext2fs_bg_free_blocks_count(fs, i) >
320 ext2fs_bg_free_blocks_count(fs, group))
322 return ext2fs_group_first_block2(fs, group);
326 * This function creates a journal using direct I/O routines.
328 static errcode_t write_journal_inode(ext2_filsys fs, ext2_ino_t journal_ino,
329 blk_t num_blocks, blk64_t goal, int flags)
333 struct ext2_inode inode;
334 unsigned long long inode_size;
335 struct mkjournal_struct es;
337 if ((retval = ext2fs_create_journal_superblock(fs, num_blocks, flags,
341 if ((retval = ext2fs_read_bitmaps(fs)))
344 if ((retval = ext2fs_read_inode(fs, journal_ino, &inode)))
347 if (inode.i_blocks > 0) {
352 es.num_blocks = num_blocks;
358 es.goal = (goal != ~0ULL) ? goal : get_midpoint_journal_block(fs);
360 if (fs->super->s_feature_incompat & EXT3_FEATURE_INCOMPAT_EXTENTS) {
361 inode.i_flags |= EXT4_EXTENTS_FL;
362 if ((retval = ext2fs_write_inode(fs, journal_ino, &inode)))
366 retval = ext2fs_block_iterate3(fs, journal_ino, BLOCK_FLAG_APPEND,
367 0, mkjournal_proc, &es);
373 retval = ext2fs_zero_blocks2(fs, es.blk_to_zero,
374 es.zero_count, 0, 0);
379 if ((retval = ext2fs_read_inode(fs, journal_ino, &inode)))
382 inode_size = (unsigned long long)fs->blocksize * num_blocks;
383 ext2fs_iblk_add_blocks(fs, &inode, es.newblocks);
384 inode.i_mtime = inode.i_ctime = fs->now ? fs->now : time(0);
385 inode.i_links_count = 1;
386 inode.i_mode = LINUX_S_IFREG | 0600;
387 retval = ext2fs_inode_size_set(fs, &inode, inode_size);
391 if ((retval = ext2fs_write_new_inode(fs, journal_ino, &inode)))
395 memcpy(fs->super->s_jnl_blocks, inode.i_block, EXT2_N_BLOCKS*4);
396 fs->super->s_jnl_blocks[15] = inode.i_size_high;
397 fs->super->s_jnl_blocks[16] = inode.i_size;
398 fs->super->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
399 ext2fs_mark_super_dirty(fs);
402 ext2fs_zero_blocks2(0, 0, 0, 0, 0);
404 ext2fs_free_mem(&buf);
409 * Find a reasonable journal file size (in blocks) given the number of blocks
410 * in the filesystem. For very small filesystems, it is not reasonable to
411 * have a journal that fills more than half of the filesystem.
413 int ext2fs_default_journal_size(__u64 num_blocks)
415 if (num_blocks < 2048)
417 if (num_blocks < 32768)
419 if (num_blocks < 256*1024)
421 if (num_blocks < 512*1024)
423 if (num_blocks < 1024*1024)
428 int ext2fs_journal_sb_start(int blocksize)
430 if (blocksize == EXT2_MIN_BLOCK_SIZE)
436 * This function adds a journal device to a filesystem
438 errcode_t ext2fs_add_journal_device(ext2_filsys fs, ext2_filsys journal_dev)
442 char buf[SUPERBLOCK_SIZE];
443 journal_superblock_t *jsb;
447 /* Make sure the device exists and is a block device */
448 if (stat(journal_dev->device_name, &st) < 0)
451 if (!S_ISBLK(st.st_mode))
452 return EXT2_ET_JOURNAL_NOT_BLOCK; /* Must be a block device */
454 /* Get the journal superblock */
455 start = ext2fs_journal_sb_start(journal_dev->blocksize);
456 if ((retval = io_channel_read_blk64(journal_dev->io, start,
461 jsb = (journal_superblock_t *) buf;
462 if ((jsb->s_header.h_magic != (unsigned) ntohl(JFS_MAGIC_NUMBER)) ||
463 (jsb->s_header.h_blocktype != (unsigned) ntohl(JFS_SUPERBLOCK_V2)))
464 return EXT2_ET_NO_JOURNAL_SB;
466 if (ntohl(jsb->s_blocksize) != (unsigned long) fs->blocksize)
467 return EXT2_ET_UNEXPECTED_BLOCK_SIZE;
469 /* Check and see if this filesystem has already been added */
470 nr_users = ntohl(jsb->s_nr_users);
471 for (i=0; i < nr_users; i++) {
472 if (memcmp(fs->super->s_uuid,
473 &jsb->s_users[i*16], 16) == 0)
477 memcpy(&jsb->s_users[nr_users*16],
478 fs->super->s_uuid, 16);
479 jsb->s_nr_users = htonl(nr_users+1);
482 /* Writeback the journal superblock */
483 if ((retval = io_channel_write_blk64(journal_dev->io, start,
484 -SUPERBLOCK_SIZE, buf)))
487 fs->super->s_journal_inum = 0;
488 fs->super->s_journal_dev = st.st_rdev;
489 memcpy(fs->super->s_journal_uuid, jsb->s_uuid,
490 sizeof(fs->super->s_journal_uuid));
491 fs->super->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
492 ext2fs_mark_super_dirty(fs);
497 * This function adds a journal inode to a filesystem, using either
498 * POSIX routines if the filesystem is mounted, or using direct I/O
499 * functions if it is not.
501 errcode_t ext2fs_add_journal_inode2(ext2_filsys fs, blk_t num_blocks,
502 blk64_t goal, int flags)
505 ext2_ino_t journal_ino;
511 if (flags & EXT2_MKJOURNAL_NO_MNT_CHECK)
513 else if ((retval = ext2fs_check_mount_point(fs->device_name,
515 jfile, sizeof(jfile)-10)))
518 if (mount_flags & EXT2_MF_MOUNTED) {
522 strcat(jfile, "/.journal");
525 * If .../.journal already exists, make sure any
526 * immutable or append-only flags are cleared.
528 #if defined(HAVE_CHFLAGS) && defined(UF_NODUMP)
529 (void) chflags (jfile, 0);
532 fd = open(jfile, O_RDONLY);
534 retval = ioctl(fd, EXT2_IOC_SETFLAGS, &f);
542 /* Create the journal file */
543 if ((fd = open(jfile, O_CREAT|O_WRONLY, 0600)) < 0)
546 /* Note that we can't do lazy journal initialization for mounted
547 * filesystems, since the zero writing is also allocating the
548 * journal blocks. We could use fallocate, but not all kernels
549 * support that, and creating a journal on a mounted ext2
550 * filesystems is extremely rare these days... Ignore it. */
551 flags &= ~EXT2_MKJOURNAL_LAZYINIT;
553 if ((retval = write_journal_file(fs, jfile, num_blocks, flags)))
556 /* Get inode number of the journal file */
557 if (fstat(fd, &st) < 0) {
562 #if defined(HAVE_CHFLAGS) && defined(UF_NODUMP)
563 retval = fchflags (fd, UF_NODUMP|UF_IMMUTABLE);
566 if (ioctl(fd, EXT2_IOC_GETFLAGS, &f) < 0) {
570 f |= EXT2_NODUMP_FL | EXT2_IMMUTABLE_FL;
571 retval = ioctl(fd, EXT2_IOC_SETFLAGS, &f);
584 journal_ino = st.st_ino;
585 memset(fs->super->s_jnl_blocks, 0,
586 sizeof(fs->super->s_jnl_blocks));
588 if ((mount_flags & EXT2_MF_BUSY) &&
589 !(fs->flags & EXT2_FLAG_EXCLUSIVE)) {
593 journal_ino = EXT2_JOURNAL_INO;
594 if ((retval = write_journal_inode(fs, journal_ino,
595 num_blocks, goal, flags)))
599 fs->super->s_journal_inum = journal_ino;
600 fs->super->s_journal_dev = 0;
601 memset(fs->super->s_journal_uuid, 0,
602 sizeof(fs->super->s_journal_uuid));
603 fs->super->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
605 ext2fs_mark_super_dirty(fs);
613 errcode_t ext2fs_add_journal_inode(ext2_filsys fs, blk_t num_blocks, int flags)
615 return ext2fs_add_journal_inode2(fs, num_blocks, ~0ULL, flags);
620 main(int argc, char **argv)
627 fprintf(stderr, "Usage: %s filesystem\n", argv[0]);
630 device_name = argv[1];
632 retval = ext2fs_open (device_name, EXT2_FLAG_RW, 0, 0,
633 unix_io_manager, &fs);
635 com_err(argv[0], retval, "while opening %s", device_name);
639 retval = ext2fs_add_journal_inode(fs, JFS_MIN_JOURNAL_BLOCKS, 0);
641 com_err(argv[0], retval, "while adding journal to %s",
645 retval = ext2fs_flush(fs);
647 printf("Warning, had trouble writing out superblocks.\n");
649 ext2fs_close_free(&fs);