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 MAX_STRIDE_LENGTH (4194304 / (int) fs->blocksize)
152 errcode_t ext2fs_zero_blocks2(ext2_filsys fs, blk64_t blk, int num,
153 blk64_t *ret_blk, int *ret_count)
157 static int stride_length;
160 /* If fs is null, clean up the static buffer and return */
169 /* Deal with zeroing less than 1 block */
173 /* Try a zero out command, if supported */
174 retval = io_channel_zeroout(fs->io, blk, num);
178 /* Allocate the zeroizing buffer if necessary */
179 if (num > stride_length && stride_length < MAX_STRIDE_LENGTH) {
181 int new_stride = num;
183 if (new_stride > MAX_STRIDE_LENGTH)
184 new_stride = MAX_STRIDE_LENGTH;
185 p = realloc(buf, fs->blocksize * new_stride);
187 return EXT2_ET_NO_MEMORY;
189 stride_length = new_stride;
190 memset(buf, 0, fs->blocksize * stride_length);
192 /* OK, do the write loop */
195 if (blk % stride_length) {
196 count = stride_length - (blk % stride_length);
197 if (count > (num - j))
201 if (count > stride_length)
202 count = stride_length;
204 retval = io_channel_write_blk64(fs->io, blk, count, buf);
212 j += count; blk += count;
217 errcode_t ext2fs_zero_blocks(ext2_filsys fs, blk_t blk, int num,
218 blk_t *ret_blk, int *ret_count)
223 retval = ext2fs_zero_blocks2(fs, blk, num, &ret_blk2, ret_count);
225 *ret_blk = (blk_t) ret_blk2;
230 * Helper function for creating the journal using direct I/O routines
232 struct mkjournal_struct {
243 static int mkjournal_proc(ext2_filsys fs,
245 e2_blkcnt_t blockcnt,
246 blk64_t ref_block EXT2FS_ATTR((unused)),
247 int ref_offset EXT2FS_ATTR((unused)),
250 struct mkjournal_struct *es = (struct mkjournal_struct *) priv_data;
259 (EXT2FS_B2C(fs, es->goal) == EXT2FS_B2C(fs, es->goal+1)))
260 new_blk = es->goal+1;
262 es->goal &= ~EXT2FS_CLUSTER_MASK(fs);
263 retval = ext2fs_new_block2(fs, es->goal, 0, &new_blk);
268 ext2fs_block_alloc_stats2(fs, new_blk, +1);
276 retval = io_channel_write_blk64(fs->io, new_blk, 1, es->buf);
277 else if (!(es->flags & EXT2_MKJOURNAL_LAZYINIT)) {
278 if (es->zero_count) {
279 if ((es->blk_to_zero + es->zero_count == new_blk) &&
280 (es->zero_count < 1024))
283 retval = ext2fs_zero_blocks2(fs,
290 if (es->zero_count == 0) {
291 es->blk_to_zero = new_blk;
297 memset(es->buf, 0, fs->blocksize);
303 *blocknr = es->goal = new_blk;
305 if (es->num_blocks == 0)
306 return (BLOCK_CHANGED | BLOCK_ABORT);
308 return BLOCK_CHANGED;
313 * Calculate the initial goal block to be roughly at the middle of the
314 * filesystem. Pick a group that has the largest number of free
317 static blk64_t get_midpoint_journal_block(ext2_filsys fs)
319 dgrp_t group, start, end, i, log_flex;
321 group = ext2fs_group_of_blk2(fs, (ext2fs_blocks_count(fs->super) -
322 fs->super->s_first_data_block) / 2);
323 log_flex = 1 << fs->super->s_log_groups_per_flex;
324 if (fs->super->s_log_groups_per_flex && (group > log_flex)) {
325 group = group & ~(log_flex - 1);
326 while ((group < fs->group_desc_count) &&
327 ext2fs_bg_free_blocks_count(fs, group) == 0)
329 if (group == fs->group_desc_count)
333 start = (group > 0) ? group-1 : group;
334 end = ((group+1) < fs->group_desc_count) ? group+1 : group;
336 for (i = start + 1; i <= end; i++)
337 if (ext2fs_bg_free_blocks_count(fs, i) >
338 ext2fs_bg_free_blocks_count(fs, group))
340 return ext2fs_group_first_block2(fs, group);
344 * This function creates a journal using direct I/O routines.
346 static errcode_t write_journal_inode(ext2_filsys fs, ext2_ino_t journal_ino,
347 blk_t num_blocks, blk64_t goal, int flags)
351 struct ext2_inode inode;
352 unsigned long long inode_size;
353 struct mkjournal_struct es;
355 if ((retval = ext2fs_create_journal_superblock(fs, num_blocks, flags,
359 if ((retval = ext2fs_read_bitmaps(fs)))
362 if ((retval = ext2fs_read_inode(fs, journal_ino, &inode)))
365 if (inode.i_blocks > 0) {
370 es.num_blocks = num_blocks;
376 es.goal = (goal != ~0ULL) ? goal : get_midpoint_journal_block(fs);
378 if (fs->super->s_feature_incompat & EXT3_FEATURE_INCOMPAT_EXTENTS) {
379 inode.i_flags |= EXT4_EXTENTS_FL;
380 if ((retval = ext2fs_write_inode(fs, journal_ino, &inode)))
384 retval = ext2fs_block_iterate3(fs, journal_ino, BLOCK_FLAG_APPEND,
385 0, mkjournal_proc, &es);
393 retval = ext2fs_zero_blocks2(fs, es.blk_to_zero,
394 es.zero_count, 0, 0);
399 if ((retval = ext2fs_read_inode(fs, journal_ino, &inode)))
402 inode_size = (unsigned long long)fs->blocksize * num_blocks;
403 ext2fs_iblk_add_blocks(fs, &inode, es.newblocks);
404 inode.i_mtime = inode.i_ctime = fs->now ? fs->now : time(0);
405 inode.i_links_count = 1;
406 inode.i_mode = LINUX_S_IFREG | 0600;
407 retval = ext2fs_inode_size_set(fs, &inode, inode_size);
411 if ((retval = ext2fs_write_new_inode(fs, journal_ino, &inode)))
415 memcpy(fs->super->s_jnl_blocks, inode.i_block, EXT2_N_BLOCKS*4);
416 fs->super->s_jnl_blocks[15] = inode.i_size_high;
417 fs->super->s_jnl_blocks[16] = inode.i_size;
418 fs->super->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
419 ext2fs_mark_super_dirty(fs);
422 ext2fs_free_mem(&buf);
427 * Find a reasonable journal file size (in blocks) given the number of blocks
428 * in the filesystem. For very small filesystems, it is not reasonable to
429 * have a journal that fills more than half of the filesystem.
431 int ext2fs_default_journal_size(__u64 num_blocks)
433 if (num_blocks < 2048)
435 if (num_blocks < 32768)
437 if (num_blocks < 256*1024)
439 if (num_blocks < 512*1024)
441 if (num_blocks < 1024*1024)
446 int ext2fs_journal_sb_start(int blocksize)
448 if (blocksize == EXT2_MIN_BLOCK_SIZE)
454 * This function adds a journal device to a filesystem
456 errcode_t ext2fs_add_journal_device(ext2_filsys fs, ext2_filsys journal_dev)
460 char buf[SUPERBLOCK_SIZE];
461 journal_superblock_t *jsb;
465 /* Make sure the device exists and is a block device */
466 if (stat(journal_dev->device_name, &st) < 0)
469 if (!S_ISBLK(st.st_mode))
470 return EXT2_ET_JOURNAL_NOT_BLOCK; /* Must be a block device */
472 /* Get the journal superblock */
473 start = ext2fs_journal_sb_start(journal_dev->blocksize);
474 if ((retval = io_channel_read_blk64(journal_dev->io, start,
479 jsb = (journal_superblock_t *) buf;
480 if ((jsb->s_header.h_magic != (unsigned) ntohl(JFS_MAGIC_NUMBER)) ||
481 (jsb->s_header.h_blocktype != (unsigned) ntohl(JFS_SUPERBLOCK_V2)))
482 return EXT2_ET_NO_JOURNAL_SB;
484 if (ntohl(jsb->s_blocksize) != (unsigned long) fs->blocksize)
485 return EXT2_ET_UNEXPECTED_BLOCK_SIZE;
487 /* Check and see if this filesystem has already been added */
488 nr_users = ntohl(jsb->s_nr_users);
489 for (i=0; i < nr_users; i++) {
490 if (memcmp(fs->super->s_uuid,
491 &jsb->s_users[i*16], 16) == 0)
495 memcpy(&jsb->s_users[nr_users*16],
496 fs->super->s_uuid, 16);
497 jsb->s_nr_users = htonl(nr_users+1);
500 /* Writeback the journal superblock */
501 if ((retval = io_channel_write_blk64(journal_dev->io, start,
502 -SUPERBLOCK_SIZE, buf)))
505 fs->super->s_journal_inum = 0;
506 fs->super->s_journal_dev = st.st_rdev;
507 memcpy(fs->super->s_journal_uuid, jsb->s_uuid,
508 sizeof(fs->super->s_journal_uuid));
509 memset(fs->super->s_jnl_blocks, 0, sizeof(fs->super->s_jnl_blocks));
510 fs->super->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
511 ext2fs_mark_super_dirty(fs);
516 * This function adds a journal inode to a filesystem, using either
517 * POSIX routines if the filesystem is mounted, or using direct I/O
518 * functions if it is not.
520 errcode_t ext2fs_add_journal_inode2(ext2_filsys fs, blk_t num_blocks,
521 blk64_t goal, int flags)
524 ext2_ino_t journal_ino;
530 if (flags & EXT2_MKJOURNAL_NO_MNT_CHECK)
532 else if ((retval = ext2fs_check_mount_point(fs->device_name,
534 jfile, sizeof(jfile)-10)))
537 if (mount_flags & EXT2_MF_MOUNTED) {
541 strcat(jfile, "/.journal");
544 * If .../.journal already exists, make sure any
545 * immutable or append-only flags are cleared.
547 #if defined(HAVE_CHFLAGS) && defined(UF_NODUMP)
548 (void) chflags (jfile, 0);
551 fd = open(jfile, O_RDONLY);
553 retval = ioctl(fd, EXT2_IOC_SETFLAGS, &f);
561 /* Create the journal file */
562 if ((fd = open(jfile, O_CREAT|O_WRONLY, 0600)) < 0)
565 /* Note that we can't do lazy journal initialization for mounted
566 * filesystems, since the zero writing is also allocating the
567 * journal blocks. We could use fallocate, but not all kernels
568 * support that, and creating a journal on a mounted ext2
569 * filesystems is extremely rare these days... Ignore it. */
570 flags &= ~EXT2_MKJOURNAL_LAZYINIT;
572 if ((retval = write_journal_file(fs, jfile, num_blocks, flags)))
575 /* Get inode number of the journal file */
576 if (fstat(fd, &st) < 0) {
581 #if defined(HAVE_CHFLAGS) && defined(UF_NODUMP)
582 retval = fchflags (fd, UF_NODUMP|UF_IMMUTABLE);
585 if (ioctl(fd, EXT2_IOC_GETFLAGS, &f) < 0) {
589 f |= EXT2_NODUMP_FL | EXT2_IMMUTABLE_FL;
590 retval = ioctl(fd, EXT2_IOC_SETFLAGS, &f);
603 journal_ino = st.st_ino;
604 memset(fs->super->s_jnl_blocks, 0,
605 sizeof(fs->super->s_jnl_blocks));
607 if ((mount_flags & EXT2_MF_BUSY) &&
608 !(fs->flags & EXT2_FLAG_EXCLUSIVE)) {
612 journal_ino = EXT2_JOURNAL_INO;
613 if ((retval = write_journal_inode(fs, journal_ino,
614 num_blocks, goal, flags)))
618 fs->super->s_journal_inum = journal_ino;
619 fs->super->s_journal_dev = 0;
620 memset(fs->super->s_journal_uuid, 0,
621 sizeof(fs->super->s_journal_uuid));
622 fs->super->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
624 ext2fs_mark_super_dirty(fs);
632 errcode_t ext2fs_add_journal_inode(ext2_filsys fs, blk_t num_blocks, int flags)
634 return ext2fs_add_journal_inode2(fs, num_blocks, ~0ULL, flags);
639 main(int argc, char **argv)
646 fprintf(stderr, "Usage: %s filesystem\n", argv[0]);
649 device_name = argv[1];
651 retval = ext2fs_open (device_name, EXT2_FLAG_RW, 0, 0,
652 unix_io_manager, &fs);
654 com_err(argv[0], retval, "while opening %s", device_name);
658 retval = ext2fs_add_journal_inode(fs, JFS_MIN_JOURNAL_BLOCKS, 0);
660 com_err(argv[0], retval, "while adding journal to %s",
664 retval = ext2fs_flush(fs);
666 printf("Warning, had trouble writing out superblocks.\n");
668 ext2fs_close_free(&fs);