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>
40 #include "kernel-jbd.h"
43 * This function automatically sets up the journal superblock and
44 * returns it as an allocated block.
46 errcode_t ext2fs_create_journal_superblock2(ext2_filsys fs,
47 struct ext2fs_journal_params *jparams,
48 int flags, char **ret_jsb)
51 journal_superblock_t *jsb;
53 if (jparams->num_journal_blocks < JBD2_MIN_JOURNAL_BLOCKS)
54 return EXT2_ET_JOURNAL_TOO_SMALL;
56 if ((retval = ext2fs_get_mem(fs->blocksize, &jsb)))
59 memset (jsb, 0, fs->blocksize);
61 jsb->s_header.h_magic = htonl(JBD2_MAGIC_NUMBER);
62 if (flags & EXT2_MKJOURNAL_V1_SUPER)
63 jsb->s_header.h_blocktype = htonl(JBD2_SUPERBLOCK_V1);
65 jsb->s_header.h_blocktype = htonl(JBD2_SUPERBLOCK_V2);
66 jsb->s_blocksize = htonl(fs->blocksize);
67 jsb->s_maxlen = htonl(jparams->num_journal_blocks + jparams->num_fc_blocks);
68 jsb->s_nr_users = htonl(1);
69 jsb->s_first = htonl(1);
70 jsb->s_sequence = htonl(1);
71 jsb->s_num_fc_blks = htonl(jparams->num_fc_blocks);
72 memcpy(jsb->s_uuid, fs->super->s_uuid, sizeof(fs->super->s_uuid));
74 * If we're creating an external journal device, we need to
75 * adjust these fields.
77 if (ext2fs_has_feature_journal_dev(fs->super)) {
79 jsb->s_first = htonl(ext2fs_journal_sb_start(fs->blocksize) + 1);
82 *ret_jsb = (char *) jsb;
86 errcode_t ext2fs_create_journal_superblock(ext2_filsys fs, __u32 num_blocks,
87 int flags, char **ret_sb)
89 struct ext2fs_journal_params jparams;
91 jparams.num_journal_blocks = num_blocks;
92 jparams.num_fc_blocks = 0;
94 return ext2fs_create_journal_superblock2(fs, &jparams, flags, ret_sb);
98 * This function writes a journal using POSIX routines. It is used
99 * for creating external journals and creating journals on live
102 static errcode_t write_journal_file(ext2_filsys fs, char *filename,
103 struct ext2fs_journal_params *jparams,
111 if ((retval = ext2fs_create_journal_superblock2(fs, jparams, flags,
115 /* Open the device or journal file */
116 if ((fd = open(filename, O_WRONLY)) < 0) {
121 /* Write the superblock out */
122 retval = EXT2_ET_SHORT_WRITE;
123 ret_size = write(fd, buf, fs->blocksize);
128 if (ret_size != (int) fs->blocksize)
130 memset(buf, 0, fs->blocksize);
132 if (flags & EXT2_MKJOURNAL_LAZYINIT)
135 for (i = 1; i < jparams->num_journal_blocks + jparams->num_fc_blocks; i++) {
136 ret_size = write(fd, buf, fs->blocksize);
141 if (ret_size != (int) fs->blocksize)
150 ext2fs_free_mem(&buf);
155 * Convenience function which zeros out _num_ blocks starting at
156 * _blk_. In case of an error, the details of the error is returned
157 * via _ret_blk_ and _ret_count_ if they are non-NULL pointers.
158 * Returns 0 on success, and an error code on an error.
160 * As a special case, if the first argument is NULL, then it will
161 * attempt to free the static zeroizing buffer. (This is to keep
162 * programs that check for memory leaks happy.)
164 #define MAX_STRIDE_LENGTH (4194304 / (int) fs->blocksize)
165 errcode_t ext2fs_zero_blocks2(ext2_filsys fs, blk64_t blk, int num,
166 blk64_t *ret_blk, int *ret_count)
170 static int stride_length;
173 /* If fs is null, clean up the static buffer and return */
183 /* Deal with zeroing less than 1 block */
187 /* Try a zero out command, if supported */
188 retval = io_channel_zeroout(fs->io, blk, num);
192 /* Allocate the zeroizing buffer if necessary */
193 if (num > stride_length && stride_length < MAX_STRIDE_LENGTH) {
195 int new_stride = num;
197 if (new_stride > MAX_STRIDE_LENGTH)
198 new_stride = MAX_STRIDE_LENGTH;
199 p = realloc(buf, fs->blocksize * new_stride);
201 return EXT2_ET_NO_MEMORY;
203 stride_length = new_stride;
204 memset(buf, 0, fs->blocksize * stride_length);
206 /* OK, do the write loop */
209 if (blk % stride_length) {
210 count = stride_length - (blk % stride_length);
211 if (count > (num - j))
215 if (count > stride_length)
216 count = stride_length;
218 retval = io_channel_write_blk64(fs->io, blk, count, buf);
226 j += count; blk += count;
231 errcode_t ext2fs_zero_blocks(ext2_filsys fs, blk_t blk, int num,
232 blk_t *ret_blk, int *ret_count)
237 retval = ext2fs_zero_blocks2(fs, blk, num, &ret_blk2, ret_count);
239 *ret_blk = (blk_t) ret_blk2;
244 * Calculate the initial goal block to be roughly at the middle of the
245 * filesystem. Pick a group that has the largest number of free
248 static blk64_t get_midpoint_journal_block(ext2_filsys fs)
250 dgrp_t group, start, end, i, log_flex;
252 group = ext2fs_group_of_blk2(fs, (ext2fs_blocks_count(fs->super) -
253 fs->super->s_first_data_block) / 2);
254 log_flex = 1U << fs->super->s_log_groups_per_flex;
255 if (fs->super->s_log_groups_per_flex && (group > log_flex)) {
256 group = group & ~(log_flex - 1);
257 while ((group < fs->group_desc_count) &&
258 ext2fs_bg_free_blocks_count(fs, group) == 0)
260 if (group == fs->group_desc_count)
264 start = (group > 0) ? group-1 : group;
265 end = ((group+1) < fs->group_desc_count) ? group+1 : group;
267 for (i = start + 1; i <= end; i++)
268 if (ext2fs_bg_free_blocks_count(fs, i) >
269 ext2fs_bg_free_blocks_count(fs, group))
271 return ext2fs_group_first_block2(fs, group);
275 * This function creates a journal using direct I/O routines.
277 static errcode_t write_journal_inode(ext2_filsys fs, ext2_ino_t journal_ino,
278 struct ext2fs_journal_params *jparams,
279 blk64_t goal, int flags)
283 struct ext2_inode inode;
284 unsigned long long inode_size;
285 int falloc_flags = EXT2_FALLOCATE_FORCE_INIT;
289 if ((retval = ext2fs_create_journal_superblock2(fs, jparams, flags,
293 if ((retval = ext2fs_read_bitmaps(fs)))
296 if ((retval = ext2fs_read_inode(fs, journal_ino, &inode)))
299 if (inode.i_blocks > 0) {
305 goal = get_midpoint_journal_block(fs);
307 if (ext2fs_has_feature_extents(fs->super))
308 inode.i_flags |= EXT4_EXTENTS_FL;
310 if (!(flags & EXT2_MKJOURNAL_LAZYINIT))
311 falloc_flags |= EXT2_FALLOCATE_ZERO_BLOCKS;
313 inode_size = (unsigned long long)fs->blocksize *
314 (jparams->num_journal_blocks + jparams->num_fc_blocks);
315 now = ext2fsP_get_time(fs);
316 ext2fs_inode_xtime_set(&inode, i_mtime, now);
317 ext2fs_inode_xtime_set(&inode, i_ctime, now);
318 inode.i_links_count = 1;
319 inode.i_mode = LINUX_S_IFREG | 0600;
320 retval = ext2fs_inode_size_set(fs, &inode, inode_size);
324 retval = ext2fs_fallocate(fs, falloc_flags, journal_ino,
326 jparams->num_journal_blocks + jparams->num_fc_blocks);
330 if ((retval = ext2fs_write_new_inode(fs, journal_ino, &inode)))
333 retval = ext2fs_bmap2(fs, journal_ino, &inode, NULL, 0, 0, NULL, &zblk);
337 retval = io_channel_write_blk64(fs->io, zblk, 1, buf);
341 memcpy(fs->super->s_jnl_blocks, inode.i_block, EXT2_N_BLOCKS*4);
342 fs->super->s_jnl_blocks[15] = inode.i_size_high;
343 fs->super->s_jnl_blocks[16] = inode.i_size;
344 fs->super->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
345 ext2fs_mark_super_dirty(fs);
348 ext2fs_free_mem(&buf);
353 * Find a reasonable journal file size (in blocks) given the number of blocks
354 * in the filesystem. For very small filesystems, it is not reasonable to
355 * have a journal that fills more than half of the filesystem.
357 * n.b. comments assume 4k blocks
359 int ext2fs_default_journal_size(__u64 num_blocks)
361 if (num_blocks < 2048)
363 if (num_blocks < 32768) /* 128 MB */
364 return (1024); /* 4 MB */
365 if (num_blocks < 256*1024) /* 1 GB */
366 return (4096); /* 16 MB */
367 if (num_blocks < 512*1024) /* 2 GB */
368 return (8192); /* 32 MB */
369 if (num_blocks < 4096*1024) /* 16 GB */
370 return (16384); /* 64 MB */
371 if (num_blocks < 8192*1024) /* 32 GB */
372 return (32768); /* 128 MB */
373 if (num_blocks < 16384*1024) /* 64 GB */
374 return (65536); /* 256 MB */
375 if (num_blocks < 32768*1024) /* 128 GB */
376 return (131072); /* 512 MB */
377 return 262144; /* 1 GB */
380 errcode_t ext2fs_get_journal_params(struct ext2fs_journal_params *params,
386 memset(params, 0, sizeof(*params));
387 if (ext2fs_has_feature_journal_dev(fs->super)) {
388 total_blks = ext2fs_blocks_count(fs->super);
389 if (total_blks < JBD2_MIN_JOURNAL_BLOCKS)
390 return EXT2_ET_JOURNAL_TOO_SMALL;
392 if (!ext2fs_has_feature_fast_commit(fs->super)) {
393 params->num_journal_blocks = total_blks;
394 params->num_fc_blocks = 0;
397 params->num_journal_blocks = ext2fs_blocks_count(fs->super) *
398 EXT2_JOURNAL_TO_FC_BLKS_RATIO /
399 (EXT2_JOURNAL_TO_FC_BLKS_RATIO + 1);
400 if (JBD2_MIN_JOURNAL_BLOCKS > params->num_journal_blocks)
401 params->num_journal_blocks = JBD2_MIN_JOURNAL_BLOCKS;
402 params->num_fc_blocks = total_blks - params->num_journal_blocks;
406 ret = ext2fs_default_journal_size(ext2fs_blocks_count(fs->super));
408 return EXT2_ET_JOURNAL_TOO_SMALL;
410 params->num_journal_blocks = ret;
411 if (ext2fs_has_feature_fast_commit(fs->super))
412 params->num_fc_blocks = params->num_journal_blocks /
413 EXT2_JOURNAL_TO_FC_BLKS_RATIO;
417 int ext2fs_journal_sb_start(int blocksize)
419 if (blocksize == EXT2_MIN_BLOCK_SIZE)
425 * This function adds a journal device to a filesystem
427 errcode_t ext2fs_add_journal_device(ext2_filsys fs, ext2_filsys journal_dev)
431 char buf[SUPERBLOCK_SIZE];
432 journal_superblock_t *jsb;
436 /* Make sure the device exists and is a block device */
437 if (stat(journal_dev->device_name, &st) < 0)
440 if (!S_ISBLK(st.st_mode))
441 return EXT2_ET_JOURNAL_NOT_BLOCK; /* Must be a block device */
443 /* Get the journal superblock */
444 start = ext2fs_journal_sb_start(journal_dev->blocksize);
445 if ((retval = io_channel_read_blk64(journal_dev->io, start,
450 jsb = (journal_superblock_t *) buf;
451 if ((jsb->s_header.h_magic != (unsigned) ntohl(JBD2_MAGIC_NUMBER)) ||
452 (jsb->s_header.h_blocktype != (unsigned) ntohl(JBD2_SUPERBLOCK_V2)))
453 return EXT2_ET_NO_JOURNAL_SB;
455 if (ntohl(jsb->s_blocksize) != (unsigned long) fs->blocksize)
456 return EXT2_ET_UNEXPECTED_BLOCK_SIZE;
458 /* Check and see if this filesystem has already been added */
459 nr_users = ntohl(jsb->s_nr_users);
460 if (nr_users > JBD2_USERS_MAX)
461 return EXT2_ET_CORRUPT_JOURNAL_SB;
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,
475 -SUPERBLOCK_SIZE, 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 memset(fs->super->s_jnl_blocks, 0, sizeof(fs->super->s_jnl_blocks));
483 ext2fs_set_feature_journal(fs->super);
484 ext2fs_mark_super_dirty(fs);
489 * This function adds a journal inode to a filesystem, using either
490 * POSIX routines if the filesystem is mounted, or using direct I/O
491 * functions if it is not.
493 errcode_t ext2fs_add_journal_inode3(ext2_filsys fs, struct ext2fs_journal_params *jparams,
494 blk64_t goal, int flags)
497 ext2_ino_t journal_ino;
503 if (flags & EXT2_MKJOURNAL_NO_MNT_CHECK)
505 else if ((retval = ext2fs_check_mount_point(fs->device_name,
507 jfile, sizeof(jfile)-10)))
510 if (mount_flags & EXT2_MF_MOUNTED) {
514 strcat(jfile, "/.journal");
517 * If .../.journal already exists, make sure any
518 * immutable or append-only flags are cleared.
520 #if defined(HAVE_CHFLAGS) && defined(UF_NODUMP)
521 (void) chflags (jfile, 0);
524 fd = open(jfile, O_RDONLY);
526 retval = ioctl(fd, EXT2_IOC_SETFLAGS, &f);
534 /* Create the journal file */
535 if ((fd = open(jfile, O_CREAT|O_WRONLY, 0600)) < 0)
538 /* Note that we can't do lazy journal initialization for mounted
539 * filesystems, since the zero writing is also allocating the
540 * journal blocks. We could use fallocate, but not all kernels
541 * support that, and creating a journal on a mounted ext2
542 * filesystems is extremely rare these days... Ignore it. */
543 flags &= ~EXT2_MKJOURNAL_LAZYINIT;
545 if ((retval = write_journal_file(fs, jfile, jparams, flags)))
548 /* Get inode number of the journal file */
549 if (fstat(fd, &st) < 0) {
554 #if defined(HAVE_CHFLAGS) && defined(UF_NODUMP)
555 retval = fchflags (fd, UF_NODUMP|UF_IMMUTABLE);
558 if (ioctl(fd, EXT2_IOC_GETFLAGS, &f) < 0) {
562 f |= EXT2_NODUMP_FL | EXT2_IMMUTABLE_FL;
563 retval = ioctl(fd, EXT2_IOC_SETFLAGS, &f);
576 journal_ino = st.st_ino;
577 memset(fs->super->s_jnl_blocks, 0,
578 sizeof(fs->super->s_jnl_blocks));
580 if ((mount_flags & EXT2_MF_BUSY) &&
581 !(fs->flags & EXT2_FLAG_EXCLUSIVE)) {
585 journal_ino = EXT2_JOURNAL_INO;
586 if ((retval = write_journal_inode(fs, journal_ino,
587 jparams, goal, flags)))
591 fs->super->s_journal_inum = journal_ino;
592 fs->super->s_journal_dev = 0;
593 memset(fs->super->s_journal_uuid, 0,
594 sizeof(fs->super->s_journal_uuid));
595 ext2fs_set_feature_journal(fs->super);
597 ext2fs_mark_super_dirty(fs);
605 errcode_t ext2fs_add_journal_inode2(ext2_filsys fs, blk_t num_blocks,
606 blk64_t goal, int flags)
608 struct ext2fs_journal_params jparams;
610 jparams.num_journal_blocks = num_blocks;
611 jparams.num_fc_blocks = 0;
613 return ext2fs_add_journal_inode3(fs, &jparams, goal, flags);
616 errcode_t ext2fs_add_journal_inode(ext2_filsys fs, blk_t num_blocks, int flags)
618 return ext2fs_add_journal_inode2(fs, num_blocks, ~0ULL, flags);
623 main(int argc, char **argv)
630 fprintf(stderr, "Usage: %s filesystem\n", argv[0]);
633 device_name = argv[1];
635 retval = ext2fs_open (device_name, EXT2_FLAG_RW, 0, 0,
636 unix_io_manager, &fs);
638 com_err(argv[0], retval, "while opening %s", device_name);
642 retval = ext2fs_add_journal_inode(fs, JBD2_MIN_JOURNAL_BLOCKS, 0);
644 com_err(argv[0], retval, "while adding journal to %s",
648 retval = ext2fs_flush(fs);
650 printf("Warning, had trouble writing out superblocks.\n");
652 ext2fs_close_free(&fs);