mkjournal.c revision efc6f628e15de95bcd13e4f0ee223cb42115d520
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		goal;
202	blk_t		blk_to_zero;
203	int		zero_count;
204	char		*buf;
205	errcode_t	err;
206};
207
208static int mkjournal_proc(ext2_filsys	fs,
209			   blk_t	*blocknr,
210			   e2_blkcnt_t	blockcnt,
211			   blk_t	ref_block EXT2FS_ATTR((unused)),
212			   int		ref_offset EXT2FS_ATTR((unused)),
213			   void		*priv_data)
214{
215	struct mkjournal_struct *es = (struct mkjournal_struct *) priv_data;
216	blk_t	new_blk;
217	errcode_t	retval;
218
219	if (*blocknr) {
220		es->goal = *blocknr;
221		return 0;
222	}
223	retval = ext2fs_new_block(fs, es->goal, 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 = es->goal = new_blk;
262	ext2fs_block_alloc_stats(fs, new_blk, +1);
263
264	if (es->num_blocks == 0)
265		return (BLOCK_CHANGED | BLOCK_ABORT);
266	else
267		return BLOCK_CHANGED;
268
269}
270
271/*
272 * This function creates a journal using direct I/O routines.
273 */
274static errcode_t write_journal_inode(ext2_filsys fs, ext2_ino_t journal_ino,
275				     blk_t size, int flags)
276{
277	char			*buf;
278	dgrp_t			group, start, end, i;
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	if (fs->super->s_feature_incompat & EXT3_FEATURE_INCOMPAT_EXTENTS) {
302		inode.i_flags |= EXT4_EXTENTS_FL;
303		if ((retval = ext2fs_write_inode(fs, journal_ino, &inode)))
304			return retval;
305	}
306
307	/*
308	 * Set the initial goal block to be roughly at the middle of
309	 * the filesystem.  Pick a group that has the largest number
310	 * of free blocks.
311	 */
312	group = ext2fs_group_of_blk(fs, (fs->super->s_blocks_count -
313					 fs->super->s_first_data_block) / 2);
314	start = (group > 0) ? group-1 : group;
315	end = ((group+1) < fs->group_desc_count) ? group+1 : group;
316	group = start;
317	for (i=start+1; i <= end; i++)
318		if (fs->group_desc[i].bg_free_blocks_count >
319		    fs->group_desc[group].bg_free_blocks_count)
320			group = i;
321
322	es.goal = (fs->super->s_blocks_per_group * group) +
323		fs->super->s_first_data_block;
324
325	retval = ext2fs_block_iterate2(fs, journal_ino, BLOCK_FLAG_APPEND,
326				       0, mkjournal_proc, &es);
327	if (es.err) {
328		retval = es.err;
329		goto errout;
330	}
331	if (es.zero_count) {
332		retval = ext2fs_zero_blocks(fs, es.blk_to_zero,
333					    es.zero_count, 0, 0);
334		if (retval)
335			goto errout;
336	}
337
338	if ((retval = ext2fs_read_inode(fs, journal_ino, &inode)))
339		goto errout;
340
341 	inode.i_size += fs->blocksize * size;
342	ext2fs_iblk_add_blocks(fs, &inode, es.newblocks);
343	inode.i_mtime = inode.i_ctime = fs->now ? fs->now : time(0);
344	inode.i_links_count = 1;
345	inode.i_mode = LINUX_S_IFREG | 0600;
346
347	if ((retval = ext2fs_write_inode(fs, journal_ino, &inode)))
348		goto errout;
349	retval = 0;
350
351	memcpy(fs->super->s_jnl_blocks, inode.i_block, EXT2_N_BLOCKS*4);
352	fs->super->s_jnl_blocks[16] = inode.i_size;
353	fs->super->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
354	ext2fs_mark_super_dirty(fs);
355
356errout:
357	ext2fs_free_mem(&buf);
358	return retval;
359}
360
361/*
362 * Find a reasonable journal file size (in blocks) given the number of blocks
363 * in the filesystem.  For very small filesystems, it is not reasonable to
364 * have a journal that fills more than half of the filesystem.
365 */
366int ext2fs_default_journal_size(__u64 blocks)
367{
368	if (blocks < 2048)
369		return -1;
370	if (blocks < 32768)
371		return (1024);
372	if (blocks < 256*1024)
373		return (4096);
374	if (blocks < 512*1024)
375		return (8192);
376	if (blocks < 1024*1024)
377		return (16384);
378	return 32768;
379}
380
381/*
382 * This function adds a journal device to a filesystem
383 */
384errcode_t ext2fs_add_journal_device(ext2_filsys fs, ext2_filsys journal_dev)
385{
386	struct stat	st;
387	errcode_t	retval;
388	char		buf[1024];
389	journal_superblock_t	*jsb;
390	int		start;
391	__u32		i, nr_users;
392
393	/* Make sure the device exists and is a block device */
394	if (stat(journal_dev->device_name, &st) < 0)
395		return errno;
396
397	if (!S_ISBLK(st.st_mode))
398		return EXT2_ET_JOURNAL_NOT_BLOCK; /* Must be a block device */
399
400	/* Get the journal superblock */
401	start = 1;
402	if (journal_dev->blocksize == 1024)
403		start++;
404	if ((retval = io_channel_read_blk(journal_dev->io, start, -1024, buf)))
405		return retval;
406
407	jsb = (journal_superblock_t *) buf;
408	if ((jsb->s_header.h_magic != (unsigned) ntohl(JFS_MAGIC_NUMBER)) ||
409	    (jsb->s_header.h_blocktype != (unsigned) ntohl(JFS_SUPERBLOCK_V2)))
410		return EXT2_ET_NO_JOURNAL_SB;
411
412	if (ntohl(jsb->s_blocksize) != (unsigned long) fs->blocksize)
413		return EXT2_ET_UNEXPECTED_BLOCK_SIZE;
414
415	/* Check and see if this filesystem has already been added */
416	nr_users = ntohl(jsb->s_nr_users);
417	for (i=0; i < nr_users; i++) {
418		if (memcmp(fs->super->s_uuid,
419			   &jsb->s_users[i*16], 16) == 0)
420			break;
421	}
422	if (i >= nr_users) {
423		memcpy(&jsb->s_users[nr_users*16],
424		       fs->super->s_uuid, 16);
425		jsb->s_nr_users = htonl(nr_users+1);
426	}
427
428	/* Writeback the journal superblock */
429	if ((retval = io_channel_write_blk(journal_dev->io, start, -1024, buf)))
430		return retval;
431
432	fs->super->s_journal_inum = 0;
433	fs->super->s_journal_dev = st.st_rdev;
434	memcpy(fs->super->s_journal_uuid, jsb->s_uuid,
435	       sizeof(fs->super->s_journal_uuid));
436	fs->super->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
437	ext2fs_mark_super_dirty(fs);
438	return 0;
439}
440
441/*
442 * This function adds a journal inode to a filesystem, using either
443 * POSIX routines if the filesystem is mounted, or using direct I/O
444 * functions if it is not.
445 */
446errcode_t ext2fs_add_journal_inode(ext2_filsys fs, blk_t size, int flags)
447{
448	errcode_t		retval;
449	ext2_ino_t		journal_ino;
450	struct stat		st;
451	char			jfile[1024];
452	int			mount_flags, f;
453	int			fd = -1;
454
455	if ((retval = ext2fs_check_mount_point(fs->device_name, &mount_flags,
456					       jfile, sizeof(jfile)-10)))
457		return retval;
458
459	if (mount_flags & EXT2_MF_MOUNTED) {
460		strcat(jfile, "/.journal");
461
462		/*
463		 * If .../.journal already exists, make sure any
464		 * immutable or append-only flags are cleared.
465		 */
466#if defined(HAVE_CHFLAGS) && defined(UF_NODUMP)
467		(void) chflags (jfile, 0);
468#else
469#if HAVE_EXT2_IOCTLS
470		fd = open(jfile, O_RDONLY);
471		if (fd >= 0) {
472			f = 0;
473			ioctl(fd, EXT2_IOC_SETFLAGS, &f);
474			close(fd);
475		}
476#endif
477#endif
478
479		/* Create the journal file */
480		if ((fd = open(jfile, O_CREAT|O_WRONLY, 0600)) < 0)
481			return errno;
482
483		if ((retval = write_journal_file(fs, jfile, size, flags)))
484			goto errout;
485
486		/* Get inode number of the journal file */
487		if (fstat(fd, &st) < 0)
488			goto errout;
489
490#if defined(HAVE_CHFLAGS) && defined(UF_NODUMP)
491		retval = fchflags (fd, UF_NODUMP|UF_IMMUTABLE);
492#else
493#if HAVE_EXT2_IOCTLS
494		f = EXT2_NODUMP_FL | EXT2_IMMUTABLE_FL;
495		retval = ioctl(fd, EXT2_IOC_SETFLAGS, &f);
496#endif
497#endif
498		if (retval)
499			goto errout;
500
501		close(fd);
502		journal_ino = st.st_ino;
503	} else {
504		if ((mount_flags & EXT2_MF_BUSY) &&
505		    !(fs->flags & EXT2_FLAG_EXCLUSIVE)) {
506			retval = EBUSY;
507			goto errout;
508		}
509		journal_ino = EXT2_JOURNAL_INO;
510		if ((retval = write_journal_inode(fs, journal_ino,
511						  size, flags)))
512			return retval;
513	}
514
515	fs->super->s_journal_inum = journal_ino;
516	fs->super->s_journal_dev = 0;
517	memset(fs->super->s_journal_uuid, 0,
518	       sizeof(fs->super->s_journal_uuid));
519	fs->super->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
520
521	ext2fs_mark_super_dirty(fs);
522	return 0;
523errout:
524	if (fd > 0)
525		close(fd);
526	return retval;
527}
528
529#ifdef DEBUG
530main(int argc, char **argv)
531{
532	errcode_t	retval;
533	char		*device_name;
534	ext2_filsys 	fs;
535
536	if (argc < 2) {
537		fprintf(stderr, "Usage: %s filesystem\n", argv[0]);
538		exit(1);
539	}
540	device_name = argv[1];
541
542	retval = ext2fs_open (device_name, EXT2_FLAG_RW, 0, 0,
543			      unix_io_manager, &fs);
544	if (retval) {
545		com_err(argv[0], retval, "while opening %s", device_name);
546		exit(1);
547	}
548
549	retval = ext2fs_add_journal_inode(fs, 1024);
550	if (retval) {
551		com_err(argv[0], retval, "while adding journal to %s",
552			device_name);
553		exit(1);
554	}
555	retval = ext2fs_flush(fs);
556	if (retval) {
557		printf("Warning, had trouble writing out superblocks.\n");
558	}
559	ext2fs_close(fs);
560	exit(0);
561
562}
563#endif
564