mkjournal.c revision bc507e31ad8a890232dc2b3aec9ee9d68e95622a 
1/*
2 * mkjournal.c --- make a journal for a filesystem
3 *
4 * Copyright (C) 2000 Theodore Ts'o.
5 *
6 * %Begin-Header%
7 * This file may be redistributed under the terms of the GNU Public
8 * License.
9 * %End-Header%
10 */
11
12#include <stdio.h>
13#include <string.h>
14#if HAVE_UNISTD_H
15#include <unistd.h>
16#endif
17#if HAVE_ERRNO_H
18#include <errno.h>
19#endif
20#include <fcntl.h>
21#include <time.h>
22#if HAVE_SYS_STAT_H
23#include <sys/stat.h>
24#endif
25#if HAVE_SYS_TYPES_H
26#include <sys/types.h>
27#endif
28#if HAVE_SYS_IOCTL_H
29#include <sys/ioctl.h>
30#endif
31#if HAVE_NETINET_IN_H
32#include <netinet/in.h>
33#endif
34
35#include "ext2_fs.h"
36#include "e2p/e2p.h"
37#include "ext2fs.h"
38#include "jfs_user.h"
39
40/*
41 * This function automatically sets up the journal superblock and
42 * returns it as an allocated block.
43 */
44errcode_t ext2fs_create_journal_superblock(ext2_filsys fs,
45					   __u32 size, int flags,
46					   char  **ret_jsb)
47{
48	errcode_t		retval;
49	journal_superblock_t	*jsb;
50
51	if (size < 1024)
52		return EXT2_ET_JOURNAL_TOO_SMALL;
53
54	if ((retval = ext2fs_get_mem(fs->blocksize, &jsb)))
55		return retval;
56
57	memset (jsb, 0, fs->blocksize);
58
59	jsb->s_header.h_magic = htonl(JFS_MAGIC_NUMBER);
60	if (flags & EXT2_MKJOURNAL_V1_SUPER)
61		jsb->s_header.h_blocktype = htonl(JFS_SUPERBLOCK_V1);
62	else
63		jsb->s_header.h_blocktype = htonl(JFS_SUPERBLOCK_V2);
64	jsb->s_blocksize = htonl(fs->blocksize);
65	jsb->s_maxlen = htonl(size);
66	jsb->s_nr_users = htonl(1);
67	jsb->s_first = htonl(1);
68	jsb->s_sequence = htonl(1);
69	memcpy(jsb->s_uuid, fs->super->s_uuid, sizeof(fs->super->s_uuid));
70	/*
71	 * If we're creating an external journal device, we need to
72	 * adjust these fields.
73	 */
74	if (fs->super->s_feature_incompat &
75	    EXT3_FEATURE_INCOMPAT_JOURNAL_DEV) {
76		jsb->s_nr_users = 0;
77		if (fs->blocksize == 1024)
78			jsb->s_first = htonl(3);
79		else
80			jsb->s_first = htonl(2);
81	}
82
83	*ret_jsb = (char *) jsb;
84	return 0;
85}
86
87/*
88 * This function writes a journal using POSIX routines.  It is used
89 * for creating external journals and creating journals on live
90 * filesystems.
91 */
92static errcode_t write_journal_file(ext2_filsys fs, char *filename,
93				    blk_t size, int flags)
94{
95	errcode_t	retval;
96	char		*buf = 0;
97	int		fd, ret_size;
98	blk_t		i;
99
100	if ((retval = ext2fs_create_journal_superblock(fs, size, flags, &buf)))
101		return retval;
102
103	/* Open the device or journal file */
104	if ((fd = open(filename, O_WRONLY)) < 0) {
105		retval = errno;
106		goto errout;
107	}
108
109	/* Write the superblock out */
110	retval = EXT2_ET_SHORT_WRITE;
111	ret_size = write(fd, buf, fs->blocksize);
112	if (ret_size < 0) {
113		retval = errno;
114		goto errout;
115	}
116	if (ret_size != (int) fs->blocksize)
117		goto errout;
118	memset(buf, 0, fs->blocksize);
119
120	for (i = 1; i < size; i++) {
121		ret_size = write(fd, buf, fs->blocksize);
122		if (ret_size < 0) {
123			retval = errno;
124			goto errout;
125		}
126		if (ret_size != (int) fs->blocksize)
127			goto errout;
128	}
129	close(fd);
130
131	retval = 0;
132errout:
133	ext2fs_free_mem(&buf);
134	return retval;
135}
136
137/*
138 * Convenience function which zeros out _num_ blocks starting at
139 * _blk_.  In case of an error, the details of the error is returned
140 * via _ret_blk_ and _ret_count_ if they are non-NULL pointers.
141 * Returns 0 on success, and an error code on an error.
142 *
143 * As a special case, if the first argument is NULL, then it will
144 * attempt to free the static zeroizing buffer.  (This is to keep
145 * programs that check for memory leaks happy.)
146 */
147#define STRIDE_LENGTH 8
148errcode_t ext2fs_zero_blocks(ext2_filsys fs, blk_t blk, int num,
149			     blk_t *ret_blk, int *ret_count)
150{
151	int		j, count;
152	static char	*buf;
153	errcode_t	retval;
154
155	/* If fs is null, clean up the static buffer and return */
156	if (!fs) {
157		if (buf) {
158			free(buf);
159			buf = 0;
160		}
161		return 0;
162	}
163	/* Allocate the zeroizing buffer if necessary */
164	if (!buf) {
165		buf = malloc(fs->blocksize * STRIDE_LENGTH);
166		if (!buf)
167			return ENOMEM;
168		memset(buf, 0, fs->blocksize * STRIDE_LENGTH);
169	}
170	/* OK, do the write loop */
171	j=0;
172	while (j < num) {
173		if (blk % STRIDE_LENGTH) {
174			count = STRIDE_LENGTH - (blk % STRIDE_LENGTH);
175			if (count > (num - j))
176				count = num - j;
177		} else {
178			count = num - j;
179			if (count > STRIDE_LENGTH)
180				count = STRIDE_LENGTH;
181		}
182		retval = io_channel_write_blk(fs->io, blk, count, buf);
183		if (retval) {
184			if (ret_count)
185				*ret_count = count;
186			if (ret_blk)
187				*ret_blk = blk;
188			return retval;
189		}
190		j += count; blk += count;
191	}
192	return 0;
193}
194
195/*
196 * Helper function for creating the journal using direct I/O routines
197 */
198struct mkjournal_struct {
199	int		num_blocks;
200	int		newblocks;
201	blk_t		blk_to_zero;
202	int		zero_count;
203	char		*buf;
204	errcode_t	err;
205};
206
207static int mkjournal_proc(ext2_filsys	fs,
208			   blk_t	*blocknr,
209			   e2_blkcnt_t	blockcnt,
210			   blk_t	ref_block EXT2FS_ATTR((unused)),
211			   int		ref_offset EXT2FS_ATTR((unused)),
212			   void		*priv_data)
213{
214	struct mkjournal_struct *es = (struct mkjournal_struct *) priv_data;
215	blk_t	new_blk;
216	static blk_t	last_blk = 0;
217	errcode_t	retval;
218
219	if (*blocknr) {
220		last_blk = *blocknr;
221		return 0;
222	}
223	retval = ext2fs_new_block(fs, last_blk, 0, &new_blk);
224	if (retval) {
225		es->err = retval;
226		return BLOCK_ABORT;
227	}
228	if (blockcnt > 0)
229		es->num_blocks--;
230
231	es->newblocks++;
232	retval = 0;
233	if (blockcnt <= 0)
234		retval = io_channel_write_blk(fs->io, new_blk, 1, es->buf);
235	else {
236		if (es->zero_count) {
237			if ((es->blk_to_zero + es->zero_count == new_blk) &&
238			    (es->zero_count < 1024))
239				es->zero_count++;
240			else {
241				retval = ext2fs_zero_blocks(fs,
242							    es->blk_to_zero,
243							    es->zero_count,
244							    0, 0);
245				es->zero_count = 0;
246			}
247		}
248		if (es->zero_count == 0) {
249			es->blk_to_zero = new_blk;
250			es->zero_count = 1;
251		}
252	}
253
254	if (blockcnt == 0)
255		memset(es->buf, 0, fs->blocksize);
256
257	if (retval) {
258		es->err = retval;
259		return BLOCK_ABORT;
260	}
261	*blocknr = new_blk;
262	last_blk = new_blk;
263	ext2fs_block_alloc_stats(fs, new_blk, +1);
264
265	if (es->num_blocks == 0)
266		return (BLOCK_CHANGED | BLOCK_ABORT);
267	else
268		return BLOCK_CHANGED;
269
270}
271
272/*
273 * This function creates a journal using direct I/O routines.
274 */
275static errcode_t write_journal_inode(ext2_filsys fs, ext2_ino_t journal_ino,
276				     blk_t size, int flags)
277{
278	char			*buf;
279	errcode_t		retval;
280	struct ext2_inode	inode;
281	struct mkjournal_struct	es;
282
283	if ((retval = ext2fs_create_journal_superblock(fs, size, flags, &buf)))
284		return retval;
285
286	if ((retval = ext2fs_read_bitmaps(fs)))
287		return retval;
288
289	if ((retval = ext2fs_read_inode(fs, journal_ino, &inode)))
290		return retval;
291
292	if (inode.i_blocks > 0)
293		return EEXIST;
294
295	es.num_blocks = size;
296	es.newblocks = 0;
297	es.buf = buf;
298	es.err = 0;
299	es.zero_count = 0;
300
301	retval = ext2fs_block_iterate2(fs, journal_ino, BLOCK_FLAG_APPEND,
302				       0, mkjournal_proc, &es);
303	if (es.err) {
304		retval = es.err;
305		goto errout;
306	}
307	if (es.zero_count) {
308		retval = ext2fs_zero_blocks(fs, es.blk_to_zero,
309					    es.zero_count, 0, 0);
310		if (retval)
311			goto errout;
312	}
313
314	if ((retval = ext2fs_read_inode(fs, journal_ino, &inode)))
315		goto errout;
316
317 	inode.i_size += fs->blocksize * size;
318	ext2fs_iblk_add_blocks(fs, &inode, es.newblocks);
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
323	if ((retval = ext2fs_write_inode(fs, journal_ino, &inode)))
324		goto errout;
325	retval = 0;
326
327	memcpy(fs->super->s_jnl_blocks, inode.i_block, EXT2_N_BLOCKS*4);
328	fs->super->s_jnl_blocks[16] = inode.i_size;
329	fs->super->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
330	ext2fs_mark_super_dirty(fs);
331
332errout:
333	ext2fs_free_mem(&buf);
334	return retval;
335}
336
337/*
338 * Find a reasonable journal file size (in blocks) given the number of blocks
339 * in the filesystem.  For very small filesystems, it is not reasonable to
340 * have a journal that fills more than half of the filesystem.
341 */
342int ext2fs_default_journal_size(__u64 blocks)
343{
344	if (blocks < 2048)
345		return -1;
346	if (blocks < 32768)
347		return (1024);
348	if (blocks < 256*1024)
349		return (4096);
350	if (blocks < 512*1024)
351		return (8192);
352	if (blocks < 1024*1024)
353		return (16384);
354	return 32768;
355}
356
357/*
358 * This function adds a journal device to a filesystem
359 */
360errcode_t ext2fs_add_journal_device(ext2_filsys fs, ext2_filsys journal_dev)
361{
362	struct stat	st;
363	errcode_t	retval;
364	char		buf[1024];
365	journal_superblock_t	*jsb;
366	int		start;
367	__u32		i, nr_users;
368
369	/* Make sure the device exists and is a block device */
370	if (stat(journal_dev->device_name, &st) < 0)
371		return errno;
372
373	if (!S_ISBLK(st.st_mode))
374		return EXT2_ET_JOURNAL_NOT_BLOCK; /* Must be a block device */
375
376	/* Get the journal superblock */
377	start = 1;
378	if (journal_dev->blocksize == 1024)
379		start++;
380	if ((retval = io_channel_read_blk(journal_dev->io, start, -1024, buf)))
381		return retval;
382
383	jsb = (journal_superblock_t *) buf;
384	if ((jsb->s_header.h_magic != (unsigned) ntohl(JFS_MAGIC_NUMBER)) ||
385	    (jsb->s_header.h_blocktype != (unsigned) ntohl(JFS_SUPERBLOCK_V2)))
386		return EXT2_ET_NO_JOURNAL_SB;
387
388	if (ntohl(jsb->s_blocksize) != (unsigned long) fs->blocksize)
389		return EXT2_ET_UNEXPECTED_BLOCK_SIZE;
390
391	/* Check and see if this filesystem has already been added */
392	nr_users = ntohl(jsb->s_nr_users);
393	for (i=0; i < nr_users; i++) {
394		if (memcmp(fs->super->s_uuid,
395			   &jsb->s_users[i*16], 16) == 0)
396			break;
397	}
398	if (i >= nr_users) {
399		memcpy(&jsb->s_users[nr_users*16],
400		       fs->super->s_uuid, 16);
401		jsb->s_nr_users = htonl(nr_users+1);
402	}
403
404	/* Writeback the journal superblock */
405	if ((retval = io_channel_write_blk(journal_dev->io, start, -1024, buf)))
406		return retval;
407
408	fs->super->s_journal_inum = 0;
409	fs->super->s_journal_dev = st.st_rdev;
410	memcpy(fs->super->s_journal_uuid, jsb->s_uuid,
411	       sizeof(fs->super->s_journal_uuid));
412	fs->super->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
413	ext2fs_mark_super_dirty(fs);
414	return 0;
415}
416
417/*
418 * This function adds a journal inode to a filesystem, using either
419 * POSIX routines if the filesystem is mounted, or using direct I/O
420 * functions if it is not.
421 */
422errcode_t ext2fs_add_journal_inode(ext2_filsys fs, blk_t size, int flags)
423{
424	errcode_t		retval;
425	ext2_ino_t		journal_ino;
426	struct stat		st;
427	char			jfile[1024];
428	int			mount_flags, f;
429	int			fd = -1;
430
431	if ((retval = ext2fs_check_mount_point(fs->device_name, &mount_flags,
432					       jfile, sizeof(jfile)-10)))
433		return retval;
434
435	if (mount_flags & EXT2_MF_MOUNTED) {
436		strcat(jfile, "/.journal");
437
438		/*
439		 * If .../.journal already exists, make sure any
440		 * immutable or append-only flags are cleared.
441		 */
442#if defined(HAVE_CHFLAGS) && defined(UF_NODUMP)
443		(void) chflags (jfile, 0);
444#else
445#if HAVE_EXT2_IOCTLS
446		fd = open(jfile, O_RDONLY);
447		if (fd >= 0) {
448			f = 0;
449			ioctl(fd, EXT2_IOC_SETFLAGS, &f);
450			close(fd);
451		}
452#endif
453#endif
454
455		/* Create the journal file */
456		if ((fd = open(jfile, O_CREAT|O_WRONLY, 0600)) < 0)
457			return errno;
458
459		if ((retval = write_journal_file(fs, jfile, size, flags)))
460			goto errout;
461
462		/* Get inode number of the journal file */
463		if (fstat(fd, &st) < 0)
464			goto errout;
465
466#if defined(HAVE_CHFLAGS) && defined(UF_NODUMP)
467		retval = fchflags (fd, UF_NODUMP|UF_IMMUTABLE);
468#else
469#if HAVE_EXT2_IOCTLS
470		f = EXT2_NODUMP_FL | EXT2_IMMUTABLE_FL;
471		retval = ioctl(fd, EXT2_IOC_SETFLAGS, &f);
472#endif
473#endif
474		if (retval)
475			goto errout;
476
477		close(fd);
478		journal_ino = st.st_ino;
479	} else {
480		if ((mount_flags & EXT2_MF_BUSY) &&
481		    !(fs->flags & EXT2_FLAG_EXCLUSIVE)) {
482			retval = EBUSY;
483			goto errout;
484		}
485		journal_ino = EXT2_JOURNAL_INO;
486		if ((retval = write_journal_inode(fs, journal_ino,
487						  size, flags)))
488			return retval;
489	}
490
491	fs->super->s_journal_inum = journal_ino;
492	fs->super->s_journal_dev = 0;
493	memset(fs->super->s_journal_uuid, 0,
494	       sizeof(fs->super->s_journal_uuid));
495	fs->super->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
496
497	ext2fs_mark_super_dirty(fs);
498	return 0;
499errout:
500	if (fd > 0)
501		close(fd);
502	return retval;
503}
504
505#ifdef DEBUG
506main(int argc, char **argv)
507{
508	errcode_t	retval;
509	char		*device_name;
510	ext2_filsys 	fs;
511
512	if (argc < 2) {
513		fprintf(stderr, "Usage: %s filesystem\n", argv[0]);
514		exit(1);
515	}
516	device_name = argv[1];
517
518	retval = ext2fs_open (device_name, EXT2_FLAG_RW, 0, 0,
519			      unix_io_manager, &fs);
520	if (retval) {
521		com_err(argv[0], retval, "while opening %s", device_name);
522		exit(1);
523	}
524
525	retval = ext2fs_add_journal_inode(fs, 1024);
526	if (retval) {
527		com_err(argv[0], retval, "while adding journal to %s",
528			device_name);
529		exit(1);
530	}
531	retval = ext2fs_flush(fs);
532	if (retval) {
533		printf("Warning, had trouble writing out superblocks.\n");
534	}
535	ext2fs_close(fs);
536	exit(0);
537
538}
539#endif
540