mkjournal.c revision 6d8b37fa7661484ca50a4951cffbf531ab1bccbb 
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 Library
8 * General Public License, version 2.
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_blocks2(ext2_filsys fs, blk64_t blk, int num,
149			      blk64_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_blk64(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
195errcode_t ext2fs_zero_blocks(ext2_filsys fs, blk_t blk, int num,
196			     blk_t *ret_blk, int *ret_count)
197{
198	blk64_t ret_blk2;
199	errcode_t retval;
200
201	retval = ext2fs_zero_blocks2(fs, blk, num, &ret_blk2, ret_count);
202	if (retval)
203		*ret_blk = (blk_t) ret_blk2;
204	return retval;
205}
206
207/*
208 * Helper function for creating the journal using direct I/O routines
209 */
210struct mkjournal_struct {
211	int		num_blocks;
212	int		newblocks;
213	blk64_t		goal;
214	blk64_t		blk_to_zero;
215	int		zero_count;
216	char		*buf;
217	errcode_t	err;
218};
219
220static int mkjournal_proc(ext2_filsys	fs,
221			  blk64_t	*blocknr,
222			  e2_blkcnt_t	blockcnt,
223			  blk64_t	ref_block EXT2FS_ATTR((unused)),
224			  int		ref_offset EXT2FS_ATTR((unused)),
225			  void		*priv_data)
226{
227	struct mkjournal_struct *es = (struct mkjournal_struct *) priv_data;
228	blk64_t	new_blk;
229	errcode_t	retval;
230
231	if (*blocknr) {
232		es->goal = *blocknr;
233		return 0;
234	}
235	retval = ext2fs_new_block2(fs, es->goal, 0, &new_blk);
236	if (retval) {
237		es->err = retval;
238		return BLOCK_ABORT;
239	}
240	if (blockcnt >= 0)
241		es->num_blocks--;
242
243	es->newblocks++;
244	retval = 0;
245	if (blockcnt <= 0)
246		retval = io_channel_write_blk64(fs->io, new_blk, 1, es->buf);
247	else {
248		if (es->zero_count) {
249			if ((es->blk_to_zero + es->zero_count == new_blk) &&
250			    (es->zero_count < 1024))
251				es->zero_count++;
252			else {
253				retval = ext2fs_zero_blocks2(fs,
254							     es->blk_to_zero,
255							     es->zero_count,
256							     0, 0);
257				es->zero_count = 0;
258			}
259		}
260		if (es->zero_count == 0) {
261			es->blk_to_zero = new_blk;
262			es->zero_count = 1;
263		}
264	}
265
266	if (blockcnt == 0)
267		memset(es->buf, 0, fs->blocksize);
268
269	if (retval) {
270		es->err = retval;
271		return BLOCK_ABORT;
272	}
273	*blocknr = es->goal = new_blk;
274	ext2fs_block_alloc_stats2(fs, new_blk, +1);
275
276	if (es->num_blocks == 0)
277		return (BLOCK_CHANGED | BLOCK_ABORT);
278	else
279		return BLOCK_CHANGED;
280
281}
282
283/*
284 * This function creates a journal using direct I/O routines.
285 */
286static errcode_t write_journal_inode(ext2_filsys fs, ext2_ino_t journal_ino,
287				     blk64_t size, int flags)
288{
289	char			*buf;
290	dgrp_t			group, start, end, i, log_flex;
291	errcode_t		retval;
292	struct ext2_inode	inode;
293	struct mkjournal_struct	es;
294
295	if ((retval = ext2fs_create_journal_superblock(fs, size, flags, &buf)))
296		return retval;
297
298	if ((retval = ext2fs_read_bitmaps(fs)))
299		return retval;
300
301	if ((retval = ext2fs_read_inode(fs, journal_ino, &inode)))
302		return retval;
303
304	if (inode.i_blocks > 0)
305		return EEXIST;
306
307	es.num_blocks = size;
308	es.newblocks = 0;
309	es.buf = buf;
310	es.err = 0;
311	es.zero_count = 0;
312
313	if (fs->super->s_feature_incompat & EXT3_FEATURE_INCOMPAT_EXTENTS) {
314		inode.i_flags |= EXT4_EXTENTS_FL;
315		if ((retval = ext2fs_write_inode(fs, journal_ino, &inode)))
316			return retval;
317	}
318
319	/*
320	 * Set the initial goal block to be roughly at the middle of
321	 * the filesystem.  Pick a group that has the largest number
322	 * of free blocks.
323	 */
324	group = ext2fs_group_of_blk2(fs, (ext2fs_blocks_count(fs->super) -
325					 fs->super->s_first_data_block) / 2);
326	log_flex = 1 << fs->super->s_log_groups_per_flex;
327	if (fs->super->s_log_groups_per_flex && (group > log_flex)) {
328		group = group & ~(log_flex - 1);
329		while ((group < fs->group_desc_count) &&
330		       ext2fs_bg_free_blocks_count(fs, group) == 0)
331			group++;
332		if (group == fs->group_desc_count)
333			group = 0;
334		start = group;
335	} else
336		start = (group > 0) ? group-1 : group;
337	end = ((group+1) < fs->group_desc_count) ? group+1 : group;
338	group = start;
339	for (i=start+1; i <= end; i++)
340		if (ext2fs_bg_free_blocks_count(fs, i) >
341		    ext2fs_bg_free_blocks_count(fs, group))
342			group = i;
343
344	es.goal = (fs->super->s_blocks_per_group * group) +
345		fs->super->s_first_data_block;
346
347	retval = ext2fs_block_iterate3(fs, journal_ino, BLOCK_FLAG_APPEND,
348				       0, mkjournal_proc, &es);
349	if (es.err) {
350		retval = es.err;
351		goto errout;
352	}
353	if (es.zero_count) {
354		retval = ext2fs_zero_blocks2(fs, es.blk_to_zero,
355					    es.zero_count, 0, 0);
356		if (retval)
357			goto errout;
358	}
359
360	if ((retval = ext2fs_read_inode(fs, journal_ino, &inode)))
361		goto errout;
362
363 	inode.i_size += fs->blocksize * size;
364	ext2fs_iblk_add_blocks(fs, &inode, es.newblocks);
365	inode.i_mtime = inode.i_ctime = fs->now ? fs->now : time(0);
366	inode.i_links_count = 1;
367	inode.i_mode = LINUX_S_IFREG | 0600;
368
369	if ((retval = ext2fs_write_new_inode(fs, journal_ino, &inode)))
370		goto errout;
371	retval = 0;
372
373	memcpy(fs->super->s_jnl_blocks, inode.i_block, EXT2_N_BLOCKS*4);
374	fs->super->s_jnl_blocks[16] = inode.i_size;
375	fs->super->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
376	ext2fs_mark_super_dirty(fs);
377
378errout:
379	ext2fs_free_mem(&buf);
380	return retval;
381}
382
383/*
384 * Find a reasonable journal file size (in blocks) given the number of blocks
385 * in the filesystem.  For very small filesystems, it is not reasonable to
386 * have a journal that fills more than half of the filesystem.
387 */
388int ext2fs_default_journal_size(__u64 blocks)
389{
390	if (blocks < 2048)
391		return -1;
392	if (blocks < 32768)
393		return (1024);
394	if (blocks < 256*1024)
395		return (4096);
396	if (blocks < 512*1024)
397		return (8192);
398	if (blocks < 1024*1024)
399		return (16384);
400	return 32768;
401}
402
403/*
404 * This function adds a journal device to a filesystem
405 */
406errcode_t ext2fs_add_journal_device(ext2_filsys fs, ext2_filsys journal_dev)
407{
408	struct stat	st;
409	errcode_t	retval;
410	char		buf[1024];
411	journal_superblock_t	*jsb;
412	int		start;
413	__u32		i, nr_users;
414
415	/* Make sure the device exists and is a block device */
416	if (stat(journal_dev->device_name, &st) < 0)
417		return errno;
418
419	if (!S_ISBLK(st.st_mode))
420		return EXT2_ET_JOURNAL_NOT_BLOCK; /* Must be a block device */
421
422	/* Get the journal superblock */
423	start = 1;
424	if (journal_dev->blocksize == 1024)
425		start++;
426	if ((retval = io_channel_read_blk64(journal_dev->io, start, -1024,
427					    buf)))
428		return retval;
429
430	jsb = (journal_superblock_t *) buf;
431	if ((jsb->s_header.h_magic != (unsigned) ntohl(JFS_MAGIC_NUMBER)) ||
432	    (jsb->s_header.h_blocktype != (unsigned) ntohl(JFS_SUPERBLOCK_V2)))
433		return EXT2_ET_NO_JOURNAL_SB;
434
435	if (ntohl(jsb->s_blocksize) != (unsigned long) fs->blocksize)
436		return EXT2_ET_UNEXPECTED_BLOCK_SIZE;
437
438	/* Check and see if this filesystem has already been added */
439	nr_users = ntohl(jsb->s_nr_users);
440	for (i=0; i < nr_users; i++) {
441		if (memcmp(fs->super->s_uuid,
442			   &jsb->s_users[i*16], 16) == 0)
443			break;
444	}
445	if (i >= nr_users) {
446		memcpy(&jsb->s_users[nr_users*16],
447		       fs->super->s_uuid, 16);
448		jsb->s_nr_users = htonl(nr_users+1);
449	}
450
451	/* Writeback the journal superblock */
452	if ((retval = io_channel_write_blk64(journal_dev->io, start, -1024, buf)))
453		return retval;
454
455	fs->super->s_journal_inum = 0;
456	fs->super->s_journal_dev = st.st_rdev;
457	memcpy(fs->super->s_journal_uuid, jsb->s_uuid,
458	       sizeof(fs->super->s_journal_uuid));
459	fs->super->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
460	ext2fs_mark_super_dirty(fs);
461	return 0;
462}
463
464/*
465 * This function adds a journal inode to a filesystem, using either
466 * POSIX routines if the filesystem is mounted, or using direct I/O
467 * functions if it is not.
468 */
469errcode_t ext2fs_add_journal_inode(ext2_filsys fs, blk_t size, int flags)
470{
471	errcode_t		retval;
472	ext2_ino_t		journal_ino;
473	struct stat		st;
474	char			jfile[1024];
475	int			mount_flags, f;
476	int			fd = -1;
477
478	if ((retval = ext2fs_check_mount_point(fs->device_name, &mount_flags,
479					       jfile, sizeof(jfile)-10)))
480		return retval;
481
482	if (mount_flags & EXT2_MF_MOUNTED) {
483		strcat(jfile, "/.journal");
484
485		/*
486		 * If .../.journal already exists, make sure any
487		 * immutable or append-only flags are cleared.
488		 */
489#if defined(HAVE_CHFLAGS) && defined(UF_NODUMP)
490		(void) chflags (jfile, 0);
491#else
492#if HAVE_EXT2_IOCTLS
493		fd = open(jfile, O_RDONLY);
494		if (fd >= 0) {
495			f = 0;
496			ioctl(fd, EXT2_IOC_SETFLAGS, &f);
497			close(fd);
498		}
499#endif
500#endif
501
502		/* Create the journal file */
503		if ((fd = open(jfile, O_CREAT|O_WRONLY, 0600)) < 0)
504			return errno;
505
506		if ((retval = write_journal_file(fs, jfile, size, flags)))
507			goto errout;
508
509		/* Get inode number of the journal file */
510		if (fstat(fd, &st) < 0) {
511			retval = errno;
512			goto errout;
513		}
514
515#if defined(HAVE_CHFLAGS) && defined(UF_NODUMP)
516		retval = fchflags (fd, UF_NODUMP|UF_IMMUTABLE);
517#else
518#if HAVE_EXT2_IOCTLS
519		if (ioctl(fd, EXT2_IOC_GETFLAGS, &f) < 0) {
520			retval = errno;
521			goto errout;
522		}
523		f |= EXT2_NODUMP_FL | EXT2_IMMUTABLE_FL;
524		retval = ioctl(fd, EXT2_IOC_SETFLAGS, &f);
525#endif
526#endif
527		if (retval) {
528			retval = errno;
529			goto errout;
530		}
531
532		if (close(fd) < 0) {
533			retval = errno;
534			fd = -1;
535			goto errout;
536		}
537		journal_ino = st.st_ino;
538	} else {
539		if ((mount_flags & EXT2_MF_BUSY) &&
540		    !(fs->flags & EXT2_FLAG_EXCLUSIVE)) {
541			retval = EBUSY;
542			goto errout;
543		}
544		journal_ino = EXT2_JOURNAL_INO;
545		if ((retval = write_journal_inode(fs, journal_ino,
546						  size, flags)))
547			return retval;
548	}
549
550	fs->super->s_journal_inum = journal_ino;
551	fs->super->s_journal_dev = 0;
552	memset(fs->super->s_journal_uuid, 0,
553	       sizeof(fs->super->s_journal_uuid));
554	fs->super->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
555
556	ext2fs_mark_super_dirty(fs);
557	return 0;
558errout:
559	if (fd > 0)
560		close(fd);
561	return retval;
562}
563
564#ifdef DEBUG
565main(int argc, char **argv)
566{
567	errcode_t	retval;
568	char		*device_name;
569	ext2_filsys 	fs;
570
571	if (argc < 2) {
572		fprintf(stderr, "Usage: %s filesystem\n", argv[0]);
573		exit(1);
574	}
575	device_name = argv[1];
576
577	retval = ext2fs_open (device_name, EXT2_FLAG_RW, 0, 0,
578			      unix_io_manager, &fs);
579	if (retval) {
580		com_err(argv[0], retval, "while opening %s", device_name);
581		exit(1);
582	}
583
584	retval = ext2fs_add_journal_inode(fs, 1024);
585	if (retval) {
586		com_err(argv[0], retval, "while adding journal to %s",
587			device_name);
588		exit(1);
589	}
590	retval = ext2fs_flush(fs);
591	if (retval) {
592		printf("Warning, had trouble writing out superblocks.\n");
593	}
594	ext2fs_close(fs);
595	exit(0);
596
597}
598#endif
599