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.
28 #include <sys/types.h>
31 #include <sys/ioctl.h>
34 #include <netinet/in.h>
37 #include <sys/utsname.h>
38 #define KERNEL_VERSION(a,b,c) (((a) << 16) + ((b) << 8) + (c))
45 #include "kernel-jbd.h"
48 * This function automatically sets up the journal superblock and
49 * returns it as an allocated block.
51 errcode_t ext2fs_create_journal_superblock2(ext2_filsys fs,
52 struct ext2fs_journal_params *jparams,
53 int flags, char **ret_jsb)
56 journal_superblock_t *jsb;
58 if (jparams->num_journal_blocks < JBD2_MIN_JOURNAL_BLOCKS)
59 return EXT2_ET_JOURNAL_TOO_SMALL;
61 if ((retval = ext2fs_get_mem(fs->blocksize, &jsb)))
64 memset (jsb, 0, fs->blocksize);
66 jsb->s_header.h_magic = htonl(JBD2_MAGIC_NUMBER);
67 if (flags & EXT2_MKJOURNAL_V1_SUPER)
68 jsb->s_header.h_blocktype = htonl(JBD2_SUPERBLOCK_V1);
70 jsb->s_header.h_blocktype = htonl(JBD2_SUPERBLOCK_V2);
71 jsb->s_blocksize = htonl(fs->blocksize);
72 jsb->s_maxlen = htonl(jparams->num_journal_blocks + jparams->num_fc_blocks);
73 jsb->s_nr_users = htonl(1);
74 jsb->s_first = htonl(1);
75 jsb->s_sequence = htonl(1);
76 jsb->s_num_fc_blks = htonl(jparams->num_fc_blocks);
77 memcpy(jsb->s_uuid, fs->super->s_uuid, sizeof(fs->super->s_uuid));
79 * If we're creating an external journal device, we need to
80 * adjust these fields.
82 if (ext2fs_has_feature_journal_dev(fs->super)) {
84 jsb->s_first = htonl(ext2fs_journal_sb_start(fs->blocksize) + 1);
87 *ret_jsb = (char *) jsb;
91 errcode_t ext2fs_create_journal_superblock(ext2_filsys fs, __u32 num_blocks,
92 int flags, char **ret_sb)
94 struct ext2fs_journal_params jparams;
96 jparams.num_journal_blocks = num_blocks;
97 jparams.num_fc_blocks = 0;
99 return ext2fs_create_journal_superblock2(fs, &jparams, flags, ret_sb);
103 * This function writes a journal using POSIX routines. It is used
104 * for creating external journals and creating journals on live
107 static errcode_t write_journal_file(ext2_filsys fs, char *filename,
108 struct ext2fs_journal_params *jparams,
116 if ((retval = ext2fs_create_journal_superblock2(fs, jparams, flags,
120 /* Open the device or journal file */
121 if ((fd = open(filename, O_WRONLY)) < 0) {
126 /* Write the superblock out */
127 retval = EXT2_ET_SHORT_WRITE;
128 ret_size = write(fd, buf, fs->blocksize);
133 if (ret_size != (int) fs->blocksize)
135 memset(buf, 0, fs->blocksize);
137 if (flags & EXT2_MKJOURNAL_LAZYINIT)
140 for (i = 1; i < jparams->num_journal_blocks + jparams->num_fc_blocks; i++) {
141 ret_size = write(fd, buf, fs->blocksize);
146 if (ret_size != (int) fs->blocksize)
155 ext2fs_free_mem(&buf);
160 * Convenience function which zeros out _num_ blocks starting at
161 * _blk_. In case of an error, the details of the error is returned
162 * via _ret_blk_ and _ret_count_ if they are non-NULL pointers.
163 * Returns 0 on success, and an error code on an error.
165 * As a special case, if the first argument is NULL, then it will
166 * attempt to free the static zeroizing buffer. (This is to keep
167 * programs that check for memory leaks happy.)
169 #define MAX_STRIDE_LENGTH (4194304 / (int) fs->blocksize)
170 errcode_t ext2fs_zero_blocks2(ext2_filsys fs, blk64_t blk, int num,
171 blk64_t *ret_blk, int *ret_count)
175 static int stride_length;
178 /* If fs is null, clean up the static buffer and return */
188 /* Deal with zeroing less than 1 block */
192 /* Try a zero out command, if supported */
193 retval = io_channel_zeroout(fs->io, blk, num);
197 /* Allocate the zeroizing buffer if necessary */
198 if (num > stride_length && stride_length < MAX_STRIDE_LENGTH) {
200 int new_stride = num;
202 if (new_stride > MAX_STRIDE_LENGTH)
203 new_stride = MAX_STRIDE_LENGTH;
204 p = realloc(buf, fs->blocksize * new_stride);
206 return EXT2_ET_NO_MEMORY;
208 stride_length = new_stride;
209 memset(buf, 0, fs->blocksize * stride_length);
211 /* OK, do the write loop */
214 if (blk % stride_length) {
215 count = stride_length - (blk % stride_length);
216 if (count > (num - j))
220 if (count > stride_length)
221 count = stride_length;
223 retval = io_channel_write_blk64(fs->io, blk, count, buf);
231 j += count; blk += count;
236 errcode_t ext2fs_zero_blocks(ext2_filsys fs, blk_t blk, int num,
237 blk_t *ret_blk, int *ret_count)
242 retval = ext2fs_zero_blocks2(fs, blk, num, &ret_blk2, ret_count);
244 *ret_blk = (blk_t) ret_blk2;
249 * Calculate the initial goal block to be roughly at the middle of the
250 * filesystem. Pick a group that has the largest number of free
253 static blk64_t get_midpoint_journal_block(ext2_filsys fs)
255 dgrp_t group, start, end, i, log_flex;
257 group = ext2fs_group_of_blk2(fs, (ext2fs_blocks_count(fs->super) -
258 fs->super->s_first_data_block) / 2);
259 log_flex = 1 << fs->super->s_log_groups_per_flex;
260 if (fs->super->s_log_groups_per_flex && (group > log_flex)) {
261 group = group & ~(log_flex - 1);
262 while ((group < fs->group_desc_count) &&
263 ext2fs_bg_free_blocks_count(fs, group) == 0)
265 if (group == fs->group_desc_count)
269 start = (group > 0) ? group-1 : group;
270 end = ((group+1) < fs->group_desc_count) ? group+1 : group;
272 for (i = start + 1; i <= end; i++)
273 if (ext2fs_bg_free_blocks_count(fs, i) >
274 ext2fs_bg_free_blocks_count(fs, group))
276 return ext2fs_group_first_block2(fs, group);
280 * This function creates a journal using direct I/O routines.
282 static errcode_t write_journal_inode(ext2_filsys fs, ext2_ino_t journal_ino,
283 struct ext2fs_journal_params *jparams,
284 blk64_t goal, int flags)
288 struct ext2_inode inode;
289 unsigned long long inode_size;
290 int falloc_flags = EXT2_FALLOCATE_FORCE_INIT;
293 if ((retval = ext2fs_create_journal_superblock2(fs, jparams, flags,
297 if ((retval = ext2fs_read_bitmaps(fs)))
300 if ((retval = ext2fs_read_inode(fs, journal_ino, &inode)))
303 if (inode.i_blocks > 0) {
309 goal = get_midpoint_journal_block(fs);
311 if (ext2fs_has_feature_extents(fs->super))
312 inode.i_flags |= EXT4_EXTENTS_FL;
314 if (!(flags & EXT2_MKJOURNAL_LAZYINIT))
315 falloc_flags |= EXT2_FALLOCATE_ZERO_BLOCKS;
317 inode_size = (unsigned long long)fs->blocksize *
318 (jparams->num_journal_blocks + jparams->num_fc_blocks);
319 inode.i_mtime = inode.i_ctime = fs->now ? fs->now : time(0);
320 inode.i_links_count = 1;
321 inode.i_mode = LINUX_S_IFREG | 0600;
322 retval = ext2fs_inode_size_set(fs, &inode, inode_size);
326 retval = ext2fs_fallocate(fs, falloc_flags, journal_ino,
328 jparams->num_journal_blocks + jparams->num_fc_blocks);
332 if ((retval = ext2fs_write_new_inode(fs, journal_ino, &inode)))
335 retval = ext2fs_bmap2(fs, journal_ino, &inode, NULL, 0, 0, NULL, &zblk);
339 retval = io_channel_write_blk64(fs->io, zblk, 1, buf);
343 memcpy(fs->super->s_jnl_blocks, inode.i_block, EXT2_N_BLOCKS*4);
344 fs->super->s_jnl_blocks[15] = inode.i_size_high;
345 fs->super->s_jnl_blocks[16] = inode.i_size;
346 fs->super->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
347 ext2fs_mark_super_dirty(fs);
350 ext2fs_free_mem(&buf);
355 static int parse_version_number(const char *s)
357 int major, minor, rev;
363 major = strtol(cp, &endptr, 10);
364 if (cp == endptr || *endptr != '.')
367 minor = strtol(cp, &endptr, 10);
368 if (cp == endptr || *endptr != '.')
371 rev = strtol(cp, &endptr, 10);
374 return KERNEL_VERSION(major, minor, rev);
377 int ext2fs_is_before_linux_ver(unsigned int major, unsigned int minor,
381 static int linux_version_code = -1;
387 if (linux_version_code < 0)
388 linux_version_code = parse_version_number(ut.release);
389 if (linux_version_code == 0)
392 return linux_version_code < KERNEL_VERSION(major, minor, rev);
395 int ext2fs_is_before_linux_ver(unsigned int major, unsigned int minor,
403 * Find a reasonable journal file size (in blocks) given the number of blocks
404 * in the filesystem. For very small filesystems, it is not reasonable to
405 * have a journal that fills more than half of the filesystem.
407 * n.b. comments assume 4k blocks
409 int ext2fs_default_journal_size(__u64 num_blocks)
411 if (num_blocks < 2048)
413 if (num_blocks <= 8192) /* 32 MB */
414 return (1024); /* 4 MB */
415 if (num_blocks < 32768) /* 128 MB */
416 return (2048); /* 8 MB */
417 if (num_blocks < 256*1024) /* 1 GB */
418 return (4096); /* 16 MB */
419 if (num_blocks < 512*1024) /* 2 GB */
420 return (8192); /* 32 MB */
421 if (num_blocks < 4096*1024) /* 16 GB */
422 return (16384); /* 64 MB */
423 if (num_blocks < 8192*1024) /* 32 GB */
424 return (32768); /* 128 MB */
425 if (num_blocks < 16384*1024) /* 64 GB */
426 return (65536); /* 256 MB */
427 if (num_blocks < 32768*1024) /* 128 GB */
428 return (131072); /* 512 MB */
429 return 262144; /* 1 GB */
432 errcode_t ext2fs_get_journal_params(struct ext2fs_journal_params *params,
438 memset(params, 0, sizeof(*params));
439 if (ext2fs_has_feature_journal_dev(fs->super)) {
440 total_blks = ext2fs_blocks_count(fs->super);
441 if (total_blks < JBD2_MIN_JOURNAL_BLOCKS)
442 return EXT2_ET_JOURNAL_TOO_SMALL;
444 if (!ext2fs_has_feature_fast_commit(fs->super)) {
445 params->num_journal_blocks = total_blks;
446 params->num_fc_blocks = 0;
449 params->num_journal_blocks = ext2fs_blocks_count(fs->super) *
450 EXT2_JOURNAL_TO_FC_BLKS_RATIO /
451 (EXT2_JOURNAL_TO_FC_BLKS_RATIO + 1);
452 if (JBD2_MIN_JOURNAL_BLOCKS > params->num_journal_blocks)
453 params->num_journal_blocks = JBD2_MIN_JOURNAL_BLOCKS;
454 params->num_fc_blocks = total_blks - params->num_journal_blocks;
458 ret = ext2fs_default_journal_size(ext2fs_blocks_count(fs->super));
460 return EXT2_ET_JOURNAL_TOO_SMALL;
462 params->num_journal_blocks = ret;
463 if (ext2fs_has_feature_fast_commit(fs->super))
464 params->num_fc_blocks = params->num_journal_blocks /
465 EXT2_JOURNAL_TO_FC_BLKS_RATIO;
469 int ext2fs_journal_sb_start(int blocksize)
471 if (blocksize == EXT2_MIN_BLOCK_SIZE)
477 * This function adds a journal device to a filesystem
479 errcode_t ext2fs_add_journal_device(ext2_filsys fs, ext2_filsys journal_dev)
483 char buf[SUPERBLOCK_SIZE];
484 journal_superblock_t *jsb;
488 /* Make sure the device exists and is a block device */
489 if (stat(journal_dev->device_name, &st) < 0)
492 if (!S_ISBLK(st.st_mode))
493 return EXT2_ET_JOURNAL_NOT_BLOCK; /* Must be a block device */
495 /* Get the journal superblock */
496 start = ext2fs_journal_sb_start(journal_dev->blocksize);
497 if ((retval = io_channel_read_blk64(journal_dev->io, start,
502 jsb = (journal_superblock_t *) buf;
503 if ((jsb->s_header.h_magic != (unsigned) ntohl(JBD2_MAGIC_NUMBER)) ||
504 (jsb->s_header.h_blocktype != (unsigned) ntohl(JBD2_SUPERBLOCK_V2)))
505 return EXT2_ET_NO_JOURNAL_SB;
507 if (ntohl(jsb->s_blocksize) != (unsigned long) fs->blocksize)
508 return EXT2_ET_UNEXPECTED_BLOCK_SIZE;
510 /* Check and see if this filesystem has already been added */
511 nr_users = ntohl(jsb->s_nr_users);
512 if (nr_users > JBD2_USERS_MAX)
513 return EXT2_ET_CORRUPT_JOURNAL_SB;
514 for (i=0; i < nr_users; i++) {
515 if (memcmp(fs->super->s_uuid,
516 &jsb->s_users[i*16], 16) == 0)
520 memcpy(&jsb->s_users[nr_users*16],
521 fs->super->s_uuid, 16);
522 jsb->s_nr_users = htonl(nr_users+1);
525 /* Writeback the journal superblock */
526 if ((retval = io_channel_write_blk64(journal_dev->io, start,
527 -SUPERBLOCK_SIZE, buf)))
530 fs->super->s_journal_inum = 0;
531 fs->super->s_journal_dev = st.st_rdev;
532 memcpy(fs->super->s_journal_uuid, jsb->s_uuid,
533 sizeof(fs->super->s_journal_uuid));
534 memset(fs->super->s_jnl_blocks, 0, sizeof(fs->super->s_jnl_blocks));
535 ext2fs_set_feature_journal(fs->super);
536 ext2fs_mark_super_dirty(fs);
541 * This function adds a journal inode to a filesystem, using either
542 * POSIX routines if the filesystem is mounted, or using direct I/O
543 * functions if it is not.
545 errcode_t ext2fs_add_journal_inode3(ext2_filsys fs, struct ext2fs_journal_params *jparams,
546 blk64_t goal, int flags)
549 ext2_ino_t journal_ino;
555 if (flags & EXT2_MKJOURNAL_NO_MNT_CHECK)
557 else if ((retval = ext2fs_check_mount_point(fs->device_name,
559 jfile, sizeof(jfile)-10)))
562 if (mount_flags & EXT2_MF_MOUNTED) {
566 strcat(jfile, "/.journal");
569 * If .../.journal already exists, make sure any
570 * immutable or append-only flags are cleared.
572 #if defined(HAVE_CHFLAGS) && defined(UF_NODUMP)
573 (void) chflags (jfile, 0);
576 fd = open(jfile, O_RDONLY);
578 retval = ioctl(fd, EXT2_IOC_SETFLAGS, &f);
586 /* Create the journal file */
587 if ((fd = open(jfile, O_CREAT|O_WRONLY, 0600)) < 0)
590 /* Note that we can't do lazy journal initialization for mounted
591 * filesystems, since the zero writing is also allocating the
592 * journal blocks. We could use fallocate, but not all kernels
593 * support that, and creating a journal on a mounted ext2
594 * filesystems is extremely rare these days... Ignore it. */
595 flags &= ~EXT2_MKJOURNAL_LAZYINIT;
597 if ((retval = write_journal_file(fs, jfile, jparams, flags)))
600 /* Get inode number of the journal file */
601 if (fstat(fd, &st) < 0) {
606 #if defined(HAVE_CHFLAGS) && defined(UF_NODUMP)
607 retval = fchflags (fd, UF_NODUMP|UF_IMMUTABLE);
610 if (ioctl(fd, EXT2_IOC_GETFLAGS, &f) < 0) {
614 f |= EXT2_NODUMP_FL | EXT2_IMMUTABLE_FL;
615 retval = ioctl(fd, EXT2_IOC_SETFLAGS, &f);
628 journal_ino = st.st_ino;
629 memset(fs->super->s_jnl_blocks, 0,
630 sizeof(fs->super->s_jnl_blocks));
632 if ((mount_flags & EXT2_MF_BUSY) &&
633 !(fs->flags & EXT2_FLAG_EXCLUSIVE)) {
637 journal_ino = EXT2_JOURNAL_INO;
638 if ((retval = write_journal_inode(fs, journal_ino,
639 jparams, goal, flags)))
643 fs->super->s_journal_inum = journal_ino;
644 fs->super->s_journal_dev = 0;
645 memset(fs->super->s_journal_uuid, 0,
646 sizeof(fs->super->s_journal_uuid));
647 ext2fs_set_feature_journal(fs->super);
649 ext2fs_mark_super_dirty(fs);
657 errcode_t ext2fs_add_journal_inode2(ext2_filsys fs, blk_t num_blocks,
658 blk64_t goal, int flags)
660 struct ext2fs_journal_params jparams;
662 jparams.num_journal_blocks = num_blocks;
663 jparams.num_fc_blocks = 0;
665 return ext2fs_add_journal_inode3(fs, &jparams, goal, flags);
668 errcode_t ext2fs_add_journal_inode(ext2_filsys fs, blk_t num_blocks, int flags)
670 return ext2fs_add_journal_inode2(fs, num_blocks, ~0ULL, flags);
675 main(int argc, char **argv)
682 fprintf(stderr, "Usage: %s filesystem\n", argv[0]);
685 device_name = argv[1];
687 retval = ext2fs_open (device_name, EXT2_FLAG_RW, 0, 0,
688 unix_io_manager, &fs);
690 com_err(argv[0], retval, "while opening %s", device_name);
694 retval = ext2fs_add_journal_inode(fs, JBD2_MIN_JOURNAL_BLOCKS, 0);
696 com_err(argv[0], retval, "while adding journal to %s",
700 retval = ext2fs_flush(fs);
702 printf("Warning, had trouble writing out superblocks.\n");
704 ext2fs_close_free(&fs);