1/*
2 * journal.c --- code for handling the "ext3" journal
3 *
4 * Copyright (C) 2000 Andreas Dilger
5 * Copyright (C) 2000 Theodore Ts'o
6 *
7 * Parts of the code are based on fs/jfs/journal.c by Stephen C. Tweedie
8 * Copyright (C) 1999 Red Hat Software
9 *
10 * This file may be redistributed under the terms of the
11 * GNU General Public License version 2 or at your discretion
12 * any later version.
13 */
14
15#include "config.h"
16#ifdef HAVE_SYS_MOUNT_H
17#include <sys/param.h>
18#include <sys/mount.h>
19#define MNT_FL (MS_MGC_VAL | MS_RDONLY)
20#endif
21#ifdef HAVE_SYS_STAT_H
22#include <sys/stat.h>
23#endif
24
25#define E2FSCK_INCLUDE_INLINE_FUNCS
26#include "uuid/uuid.h"
27#include "journal.h"
28
29#ifdef CONFIG_JBD_DEBUG		/* Enabled by configure --enable-jfs-debug */
30static int bh_count = 0;
31#endif
32
33#if EXT2_FLAT_INCLUDES
34#include "blkid.h"
35#else
36#include "blkid/blkid.h"
37#endif
38
39/*
40 * Define USE_INODE_IO to use the inode_io.c / fileio.c codepaths.
41 * This creates a larger static binary, and a smaller binary using
42 * shared libraries.  It's also probably slightly less CPU-efficient,
43 * which is why it's not on by default.  But, it's a good way of
44 * testing the functions in inode_io.c and fileio.c.
45 */
46#undef USE_INODE_IO
47
48/* Checksumming functions */
49static int ext2fs_journal_verify_csum_type(journal_t *j,
50					   journal_superblock_t *jsb)
51{
52	if (!journal_has_csum_v2or3(j))
53		return 1;
54
55	return jsb->s_checksum_type == JBD2_CRC32C_CHKSUM;
56}
57
58static __u32 ext2fs_journal_sb_csum(journal_superblock_t *jsb)
59{
60	__u32 crc, old_crc;
61
62	old_crc = jsb->s_checksum;
63	jsb->s_checksum = 0;
64	crc = ext2fs_crc32c_le(~0, (unsigned char *)jsb,
65			       sizeof(journal_superblock_t));
66	jsb->s_checksum = old_crc;
67
68	return crc;
69}
70
71static int ext2fs_journal_sb_csum_verify(journal_t *j,
72					 journal_superblock_t *jsb)
73{
74	__u32 provided, calculated;
75
76	if (!journal_has_csum_v2or3(j))
77		return 1;
78
79	provided = ext2fs_be32_to_cpu(jsb->s_checksum);
80	calculated = ext2fs_journal_sb_csum(jsb);
81
82	return provided == calculated;
83}
84
85static errcode_t ext2fs_journal_sb_csum_set(journal_t *j,
86					    journal_superblock_t *jsb)
87{
88	__u32 crc;
89
90	if (!journal_has_csum_v2or3(j))
91		return 0;
92
93	crc = ext2fs_journal_sb_csum(jsb);
94	jsb->s_checksum = ext2fs_cpu_to_be32(crc);
95	return 0;
96}
97
98/* Kernel compatibility functions for handling the journal.  These allow us
99 * to use the recovery.c file virtually unchanged from the kernel, so we
100 * don't have to do much to keep kernel and user recovery in sync.
101 */
102int journal_bmap(journal_t *journal, blk64_t block, unsigned long long *phys)
103{
104#ifdef USE_INODE_IO
105	*phys = block;
106	return 0;
107#else
108	struct inode	*inode = journal->j_inode;
109	errcode_t	retval;
110	blk64_t		pblk;
111
112	if (!inode) {
113		*phys = block;
114		return 0;
115	}
116
117	retval = ext2fs_bmap2(inode->i_fs, inode->i_ino,
118			      &inode->i_ext2, NULL, 0, block, 0, &pblk);
119	*phys = pblk;
120	return (int) retval;
121#endif
122}
123
124struct buffer_head *getblk(kdev_t kdev, blk64_t blocknr, int blocksize)
125{
126	struct buffer_head *bh;
127	int bufsize = sizeof(*bh) + kdev->k_fs->blocksize -
128		sizeof(bh->b_data);
129	errcode_t retval;
130
131	retval = ext2fs_get_memzero(bufsize, &bh);
132	if (retval)
133		return NULL;
134
135#ifdef CONFIG_JBD_DEBUG
136	if (journal_enable_debug >= 3)
137		bh_count++;
138#endif
139	jfs_debug(4, "getblk for block %llu (%d bytes)(total %d)\n",
140		  (unsigned long long) blocknr, blocksize, bh_count);
141
142	bh->b_fs = kdev->k_fs;
143	if (kdev->k_dev == K_DEV_FS)
144		bh->b_io = kdev->k_fs->io;
145	else
146		bh->b_io = kdev->k_fs->journal_io;
147	bh->b_size = blocksize;
148	bh->b_blocknr = blocknr;
149
150	return bh;
151}
152
153int sync_blockdev(kdev_t kdev)
154{
155	io_channel	io;
156
157	if (kdev->k_dev == K_DEV_FS)
158		io = kdev->k_fs->io;
159	else
160		io = kdev->k_fs->journal_io;
161
162	return io_channel_flush(io) ? EIO : 0;
163}
164
165void ll_rw_block(int rw, int nr, struct buffer_head *bhp[])
166{
167	errcode_t retval;
168	struct buffer_head *bh;
169
170	for (; nr > 0; --nr) {
171		bh = *bhp++;
172		if (rw == READ && !bh->b_uptodate) {
173			jfs_debug(3, "reading block %llu/%p\n",
174				  bh->b_blocknr, (void *) bh);
175			retval = io_channel_read_blk64(bh->b_io,
176						     bh->b_blocknr,
177						     1, bh->b_data);
178			if (retval) {
179				com_err(bh->b_fs->device_name, retval,
180					"while reading block %llu\n",
181					bh->b_blocknr);
182				bh->b_err = (int) retval;
183				continue;
184			}
185			bh->b_uptodate = 1;
186		} else if (rw == WRITE && bh->b_dirty) {
187			jfs_debug(3, "writing block %llu/%p\n",
188				  bh->b_blocknr,
189				  (void *) bh);
190			retval = io_channel_write_blk64(bh->b_io,
191						      bh->b_blocknr,
192						      1, bh->b_data);
193			if (retval) {
194				com_err(bh->b_fs->device_name, retval,
195					"while writing block %llu\n",
196					bh->b_blocknr);
197				bh->b_err = (int) retval;
198				continue;
199			}
200			bh->b_dirty = 0;
201			bh->b_uptodate = 1;
202		} else {
203			jfs_debug(3, "no-op %s for block %llu\n",
204				  rw == READ ? "read" : "write",
205				  bh->b_blocknr);
206		}
207	}
208}
209
210void mark_buffer_dirty(struct buffer_head *bh)
211{
212	bh->b_dirty = 1;
213}
214
215static void mark_buffer_clean(struct buffer_head *bh)
216{
217	bh->b_dirty = 0;
218}
219
220void brelse(struct buffer_head *bh)
221{
222	if (bh->b_dirty)
223		ll_rw_block(WRITE, 1, &bh);
224	jfs_debug(3, "freeing block %llu/%p (total %d)\n",
225		  bh->b_blocknr, (void *) bh, --bh_count);
226	ext2fs_free_mem(&bh);
227}
228
229int buffer_uptodate(struct buffer_head *bh)
230{
231	return bh->b_uptodate;
232}
233
234void mark_buffer_uptodate(struct buffer_head *bh, int val)
235{
236	bh->b_uptodate = val;
237}
238
239void wait_on_buffer(struct buffer_head *bh)
240{
241	if (!bh->b_uptodate)
242		ll_rw_block(READ, 1, &bh);
243}
244
245
246static void ext2fs_clear_recover(ext2_filsys fs, int error)
247{
248	ext2fs_clear_feature_journal_needs_recovery(fs->super);
249
250	/* if we had an error doing journal recovery, we need a full fsck */
251	if (error)
252		fs->super->s_state &= ~EXT2_VALID_FS;
253	ext2fs_mark_super_dirty(fs);
254}
255
256/*
257 * This is a helper function to check the validity of the journal.
258 */
259struct process_block_struct {
260	e2_blkcnt_t	last_block;
261};
262
263static int process_journal_block(ext2_filsys fs,
264				 blk64_t	*block_nr,
265				 e2_blkcnt_t blockcnt,
266				 blk64_t ref_block EXT2FS_ATTR((unused)),
267				 int ref_offset EXT2FS_ATTR((unused)),
268				 void *priv_data)
269{
270	struct process_block_struct *p;
271	blk64_t	blk = *block_nr;
272
273	p = (struct process_block_struct *) priv_data;
274
275	if (!blk || blk < fs->super->s_first_data_block ||
276	    blk >= ext2fs_blocks_count(fs->super))
277		return BLOCK_ABORT;
278
279	if (blockcnt >= 0)
280		p->last_block = blockcnt;
281	return 0;
282}
283
284static errcode_t ext2fs_get_journal(ext2_filsys fs, journal_t **ret_journal)
285{
286	struct process_block_struct pb;
287	struct ext2_super_block *sb = fs->super;
288	struct ext2_super_block jsuper;
289	struct buffer_head	*bh;
290	struct inode		*j_inode = NULL;
291	struct kdev_s		*dev_fs = NULL, *dev_journal;
292	const char		*journal_name = 0;
293	journal_t		*journal = NULL;
294	errcode_t		retval = 0;
295	io_manager		io_ptr = 0;
296	unsigned long long	start = 0;
297	int			ext_journal = 0;
298	int			tried_backup_jnl = 0;
299
300	retval = ext2fs_get_memzero(sizeof(journal_t), &journal);
301	if (retval)
302		return retval;
303
304	retval = ext2fs_get_memzero(2 * sizeof(struct kdev_s), &dev_fs);
305	if (retval)
306		goto errout;
307	dev_journal = dev_fs+1;
308
309	dev_fs->k_fs = dev_journal->k_fs = fs;
310	dev_fs->k_dev = K_DEV_FS;
311	dev_journal->k_dev = K_DEV_JOURNAL;
312
313	journal->j_dev = dev_journal;
314	journal->j_fs_dev = dev_fs;
315	journal->j_inode = NULL;
316	journal->j_blocksize = fs->blocksize;
317
318	if (uuid_is_null(sb->s_journal_uuid)) {
319		if (!sb->s_journal_inum) {
320			retval = EXT2_ET_BAD_INODE_NUM;
321			goto errout;
322		}
323		retval = ext2fs_get_memzero(sizeof(*j_inode), &j_inode);
324		if (retval)
325			goto errout;
326
327		j_inode->i_fs = fs;
328		j_inode->i_ino = sb->s_journal_inum;
329
330		retval = ext2fs_read_inode(fs, sb->s_journal_inum,
331					   &j_inode->i_ext2);
332		if (retval) {
333try_backup_journal:
334			if (sb->s_jnl_backup_type != EXT3_JNL_BACKUP_BLOCKS ||
335			    tried_backup_jnl)
336				goto errout;
337			memset(&j_inode->i_ext2, 0, sizeof(struct ext2_inode));
338			memcpy(&j_inode->i_ext2.i_block[0], sb->s_jnl_blocks,
339			       EXT2_N_BLOCKS*4);
340			j_inode->i_ext2.i_size_high = sb->s_jnl_blocks[15];
341			j_inode->i_ext2.i_size = sb->s_jnl_blocks[16];
342			j_inode->i_ext2.i_links_count = 1;
343			j_inode->i_ext2.i_mode = LINUX_S_IFREG | 0600;
344			tried_backup_jnl++;
345		}
346		if (!j_inode->i_ext2.i_links_count ||
347		    !LINUX_S_ISREG(j_inode->i_ext2.i_mode)) {
348			retval = EXT2_ET_NO_JOURNAL;
349			goto try_backup_journal;
350		}
351		if (EXT2_I_SIZE(&j_inode->i_ext2) / journal->j_blocksize <
352		    JFS_MIN_JOURNAL_BLOCKS) {
353			retval = EXT2_ET_JOURNAL_TOO_SMALL;
354			goto try_backup_journal;
355		}
356		pb.last_block = -1;
357		retval = ext2fs_block_iterate3(fs, j_inode->i_ino,
358					       BLOCK_FLAG_HOLE, 0,
359					       process_journal_block, &pb);
360		if ((pb.last_block + 1) * fs->blocksize <
361		    (int) EXT2_I_SIZE(&j_inode->i_ext2)) {
362			retval = EXT2_ET_JOURNAL_TOO_SMALL;
363			goto try_backup_journal;
364		}
365		if (tried_backup_jnl && (fs->flags & EXT2_FLAG_RW)) {
366			retval = ext2fs_write_inode(fs, sb->s_journal_inum,
367						    &j_inode->i_ext2);
368			if (retval)
369				goto errout;
370		}
371
372		journal->j_maxlen = EXT2_I_SIZE(&j_inode->i_ext2) /
373			journal->j_blocksize;
374
375#ifdef USE_INODE_IO
376		retval = ext2fs_inode_io_intern2(fs, sb->s_journal_inum,
377						 &j_inode->i_ext2,
378						 &journal_name);
379		if (retval)
380			goto errout;
381
382		io_ptr = inode_io_manager;
383#else
384		journal->j_inode = j_inode;
385		fs->journal_io = fs->io;
386		retval = (errcode_t)journal_bmap(journal, 0, &start);
387		if (retval)
388			goto errout;
389#endif
390	} else {
391		ext_journal = 1;
392		if (!fs->journal_name) {
393			char uuid[37];
394			blkid_cache blkid;
395
396			blkid_get_cache(&blkid, NULL);
397			uuid_unparse(sb->s_journal_uuid, uuid);
398			fs->journal_name = blkid_get_devname(blkid,
399							      "UUID", uuid);
400			if (!fs->journal_name)
401				fs->journal_name = blkid_devno_to_devname(sb->s_journal_dev);
402			blkid_put_cache(blkid);
403		}
404		journal_name = fs->journal_name;
405
406		if (!journal_name) {
407			retval = EXT2_ET_LOAD_EXT_JOURNAL;
408			goto errout;
409		}
410
411		jfs_debug(1, "Using journal file %s\n", journal_name);
412		io_ptr = unix_io_manager;
413	}
414
415#if 0
416	test_io_backing_manager = io_ptr;
417	io_ptr = test_io_manager;
418#endif
419#ifndef USE_INODE_IO
420	if (ext_journal)
421#endif
422	{
423		retval = io_ptr->open(journal_name, fs->flags & EXT2_FLAG_RW,
424				      &fs->journal_io);
425	}
426	if (retval)
427		goto errout;
428
429	io_channel_set_blksize(fs->journal_io, fs->blocksize);
430
431	if (ext_journal) {
432		blk64_t maxlen;
433
434		start = ext2fs_journal_sb_start(fs->blocksize) - 1;
435		bh = getblk(dev_journal, start, fs->blocksize);
436		if (!bh) {
437			retval = EXT2_ET_NO_MEMORY;
438			goto errout;
439		}
440		ll_rw_block(READ, 1, &bh);
441		retval = bh->b_err;
442		if (retval) {
443			brelse(bh);
444			goto errout;
445		}
446		memcpy(&jsuper, start ? bh->b_data :
447				bh->b_data + SUPERBLOCK_OFFSET,
448		       sizeof(jsuper));
449#ifdef WORDS_BIGENDIAN
450		if (jsuper.s_magic == ext2fs_swab16(EXT2_SUPER_MAGIC))
451			ext2fs_swap_super(&jsuper);
452#endif
453		if (jsuper.s_magic != EXT2_SUPER_MAGIC ||
454		    !ext2fs_has_feature_journal_dev(&jsuper)) {
455			retval = EXT2_ET_LOAD_EXT_JOURNAL;
456			brelse(bh);
457			goto errout;
458		}
459		/* Make sure the journal UUID is correct */
460		if (memcmp(jsuper.s_uuid, fs->super->s_journal_uuid,
461			   sizeof(jsuper.s_uuid))) {
462			retval = EXT2_ET_LOAD_EXT_JOURNAL;
463			brelse(bh);
464			goto errout;
465		}
466
467		/* Check the superblock checksum */
468		if (ext2fs_has_feature_metadata_csum(&jsuper)) {
469			struct struct_ext2_filsys fsx;
470			struct ext2_super_block	superx;
471			void *p;
472
473			p = start ? bh->b_data : bh->b_data + SUPERBLOCK_OFFSET;
474			memcpy(&fsx, fs, sizeof(fsx));
475			memcpy(&superx, fs->super, sizeof(superx));
476			fsx.super = &superx;
477			ext2fs_set_feature_metadata_csum(fsx.super);
478			if (!ext2fs_superblock_csum_verify(&fsx, p)) {
479				retval = EXT2_ET_LOAD_EXT_JOURNAL;
480				brelse(bh);
481				goto errout;
482			}
483		}
484		brelse(bh);
485
486		maxlen = ext2fs_blocks_count(&jsuper);
487		journal->j_maxlen = (maxlen < 1ULL << 32) ? maxlen :
488				    (1ULL << 32) - 1;
489		start++;
490	}
491
492	bh = getblk(dev_journal, start, journal->j_blocksize);
493	if (!bh) {
494		retval = EXT2_ET_NO_MEMORY;
495		goto errout;
496	}
497
498	journal->j_sb_buffer = bh;
499	journal->j_superblock = (journal_superblock_t *)bh->b_data;
500
501#ifdef USE_INODE_IO
502	if (j_inode)
503		ext2fs_free_mem(&j_inode);
504#endif
505
506	*ret_journal = journal;
507	return 0;
508
509errout:
510	if (dev_fs)
511		ext2fs_free_mem(&dev_fs);
512	if (j_inode)
513		ext2fs_free_mem(&j_inode);
514	if (journal)
515		ext2fs_free_mem(&journal);
516	return retval;
517}
518
519static errcode_t ext2fs_journal_fix_bad_inode(ext2_filsys fs)
520{
521	struct ext2_super_block *sb = fs->super;
522	int recover = ext2fs_has_feature_journal_needs_recovery(fs->super);
523	int has_journal = ext2fs_has_feature_journal(fs->super);
524
525	if (has_journal || sb->s_journal_inum) {
526		/* The journal inode is bogus, remove and force full fsck */
527		return EXT2_ET_BAD_INODE_NUM;
528	} else if (recover) {
529		return EXT2_ET_UNSUPP_FEATURE;
530	}
531	return 0;
532}
533
534#define V1_SB_SIZE	0x0024
535static void clear_v2_journal_fields(journal_t *journal)
536{
537	ext2_filsys fs = journal->j_dev->k_fs;
538
539	memset(((char *) journal->j_superblock) + V1_SB_SIZE, 0,
540	       fs->blocksize-V1_SB_SIZE);
541	mark_buffer_dirty(journal->j_sb_buffer);
542}
543
544
545static errcode_t ext2fs_journal_load(journal_t *journal)
546{
547	ext2_filsys fs = journal->j_dev->k_fs;
548	journal_superblock_t *jsb;
549	struct buffer_head *jbh = journal->j_sb_buffer;
550
551	ll_rw_block(READ, 1, &jbh);
552	if (jbh->b_err)
553		return jbh->b_err;
554
555	jsb = journal->j_superblock;
556	/* If we don't even have JFS_MAGIC, we probably have a wrong inode */
557	if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER))
558		return ext2fs_journal_fix_bad_inode(fs);
559
560	switch (ntohl(jsb->s_header.h_blocktype)) {
561	case JFS_SUPERBLOCK_V1:
562		journal->j_format_version = 1;
563		if (jsb->s_feature_compat ||
564		    jsb->s_feature_incompat ||
565		    jsb->s_feature_ro_compat ||
566		    jsb->s_nr_users)
567			clear_v2_journal_fields(journal);
568		break;
569
570	case JFS_SUPERBLOCK_V2:
571		journal->j_format_version = 2;
572		if (ntohl(jsb->s_nr_users) > 1 &&
573		    uuid_is_null(fs->super->s_journal_uuid))
574			clear_v2_journal_fields(journal);
575		if (ntohl(jsb->s_nr_users) > 1)
576			return EXT2_ET_JOURNAL_UNSUPP_VERSION;
577		break;
578
579	/*
580	 * These should never appear in a journal super block, so if
581	 * they do, the journal is badly corrupted.
582	 */
583	case JFS_DESCRIPTOR_BLOCK:
584	case JFS_COMMIT_BLOCK:
585	case JFS_REVOKE_BLOCK:
586		return EXT2_ET_CORRUPT_JOURNAL_SB;
587
588	/* If we don't understand the superblock major type, but there
589	 * is a magic number, then it is likely to be a new format we
590	 * just don't understand, so leave it alone. */
591	default:
592		return EXT2_ET_JOURNAL_UNSUPP_VERSION;
593	}
594
595	if (JFS_HAS_INCOMPAT_FEATURE(journal, ~JFS_KNOWN_INCOMPAT_FEATURES))
596		return EXT2_ET_UNSUPP_FEATURE;
597
598	if (JFS_HAS_RO_COMPAT_FEATURE(journal, ~JFS_KNOWN_ROCOMPAT_FEATURES))
599		return EXT2_ET_RO_UNSUPP_FEATURE;
600
601	/* Checksum v1-3 are mutually exclusive features. */
602	if (jfs_has_feature_csum2(journal) && jfs_has_feature_csum3(journal))
603		return EXT2_ET_CORRUPT_JOURNAL_SB;
604
605	if (journal_has_csum_v2or3(journal) &&
606	    jfs_has_feature_checksum(journal))
607		return EXT2_ET_CORRUPT_JOURNAL_SB;
608
609	if (!ext2fs_journal_verify_csum_type(journal, jsb) ||
610	    !ext2fs_journal_sb_csum_verify(journal, jsb))
611		return EXT2_ET_CORRUPT_JOURNAL_SB;
612
613	if (journal_has_csum_v2or3(journal))
614		journal->j_csum_seed = jbd2_chksum(journal, ~0, jsb->s_uuid,
615						   sizeof(jsb->s_uuid));
616
617	/* We have now checked whether we know enough about the journal
618	 * format to be able to proceed safely, so any other checks that
619	 * fail we should attempt to recover from. */
620	if (jsb->s_blocksize != htonl(journal->j_blocksize))
621		return EXT2_ET_CORRUPT_JOURNAL_SB;
622
623	if (ntohl(jsb->s_maxlen) < journal->j_maxlen)
624		journal->j_maxlen = ntohl(jsb->s_maxlen);
625	else if (ntohl(jsb->s_maxlen) > journal->j_maxlen)
626		return EXT2_ET_CORRUPT_JOURNAL_SB;
627
628	journal->j_tail_sequence = ntohl(jsb->s_sequence);
629	journal->j_transaction_sequence = journal->j_tail_sequence;
630	journal->j_tail = ntohl(jsb->s_start);
631	journal->j_first = ntohl(jsb->s_first);
632	journal->j_last = ntohl(jsb->s_maxlen);
633
634	return 0;
635}
636
637static void ext2fs_journal_release(ext2_filsys fs, journal_t *journal,
638				   int reset, int drop)
639{
640	journal_superblock_t *jsb;
641
642	if (drop)
643		mark_buffer_clean(journal->j_sb_buffer);
644	else if (fs->flags & EXT2_FLAG_RW) {
645		jsb = journal->j_superblock;
646		jsb->s_sequence = htonl(journal->j_tail_sequence);
647		if (reset)
648			jsb->s_start = 0; /* this marks the journal as empty */
649		ext2fs_journal_sb_csum_set(journal, jsb);
650		mark_buffer_dirty(journal->j_sb_buffer);
651	}
652	brelse(journal->j_sb_buffer);
653
654	if (fs && fs->journal_io) {
655		if (fs->io != fs->journal_io)
656			io_channel_close(fs->journal_io);
657		fs->journal_io = NULL;
658	}
659
660#ifndef USE_INODE_IO
661	if (journal->j_inode)
662		ext2fs_free_mem(&journal->j_inode);
663#endif
664	if (journal->j_fs_dev)
665		ext2fs_free_mem(&journal->j_fs_dev);
666	ext2fs_free_mem(&journal);
667}
668
669/*
670 * This function makes sure that the superblock fields regarding the
671 * journal are consistent.
672 */
673static errcode_t ext2fs_check_ext3_journal(ext2_filsys fs)
674{
675	struct ext2_super_block *sb = fs->super;
676	journal_t *journal;
677	int recover = ext2fs_has_feature_journal_needs_recovery(fs->super);
678	errcode_t retval;
679
680	/* If we don't have any journal features, don't do anything more */
681	if (!ext2fs_has_feature_journal(sb) &&
682	    !recover && sb->s_journal_inum == 0 && sb->s_journal_dev == 0 &&
683	    uuid_is_null(sb->s_journal_uuid))
684		return 0;
685
686	retval = ext2fs_get_journal(fs, &journal);
687	if (retval)
688		return retval;
689
690	retval = ext2fs_journal_load(journal);
691	if (retval)
692		goto err;
693
694	/*
695	 * We want to make the flags consistent here.  We will not leave with
696	 * needs_recovery set but has_journal clear.  We can't get in a loop
697	 * with -y, -n, or -p, only if a user isn't making up their mind.
698	 */
699	if (!ext2fs_has_feature_journal(sb)) {
700		retval = EXT2_ET_JOURNAL_FLAGS_WRONG;
701		goto err;
702	}
703
704	if (ext2fs_has_feature_journal(sb) &&
705	    !ext2fs_has_feature_journal_needs_recovery(sb) &&
706	    journal->j_superblock->s_start != 0) {
707		retval = EXT2_ET_JOURNAL_FLAGS_WRONG;
708		goto err;
709	}
710
711	/*
712	 * If we don't need to do replay the journal, check to see if
713	 * the journal's errno is set; if so, we need to mark the file
714	 * system as being corrupt and clear the journal's s_errno.
715	 */
716	if (!ext2fs_has_feature_journal_needs_recovery(sb) &&
717	    journal->j_superblock->s_errno) {
718		fs->super->s_state |= EXT2_ERROR_FS;
719		ext2fs_mark_super_dirty(fs);
720		journal->j_superblock->s_errno = 0;
721		ext2fs_journal_sb_csum_set(journal, journal->j_superblock);
722		mark_buffer_dirty(journal->j_sb_buffer);
723	}
724
725err:
726	ext2fs_journal_release(fs, journal, 0, retval ? 1 : 0);
727	return retval;
728}
729
730static errcode_t recover_ext3_journal(ext2_filsys fs)
731{
732	journal_t *journal;
733	errcode_t retval;
734
735	journal_init_revoke_caches();
736	retval = ext2fs_get_journal(fs, &journal);
737	if (retval)
738		return retval;
739
740	retval = ext2fs_journal_load(journal);
741	if (retval)
742		goto errout;
743
744	retval = journal_init_revoke(journal, 1024);
745	if (retval)
746		goto errout;
747
748	retval = -journal_recover(journal);
749	if (retval)
750		goto errout;
751
752	if (journal->j_failed_commit) {
753		journal->j_superblock->s_errno = -EINVAL;
754		mark_buffer_dirty(journal->j_sb_buffer);
755	}
756
757errout:
758	journal_destroy_revoke(journal);
759	journal_destroy_revoke_caches();
760	ext2fs_journal_release(fs, journal, 1, 0);
761	return retval;
762}
763
764errcode_t ext2fs_run_ext3_journal(ext2_filsys *fsp)
765{
766	ext2_filsys fs = *fsp;
767	io_manager io_ptr = fs->io->manager;
768	errcode_t	retval, recover_retval;
769	io_stats	stats = 0;
770	unsigned long long kbytes_written = 0;
771	char *fsname;
772	int fsflags;
773	int fsblocksize;
774
775	if (!(fs->flags & EXT2_FLAG_RW))
776		return EXT2_ET_FILE_RO;
777
778	if (fs->flags & EXT2_FLAG_DIRTY)
779		ext2fs_flush(fs);	/* Force out any modifications */
780
781	recover_retval = recover_ext3_journal(fs);
782
783	/*
784	 * Reload the filesystem context to get up-to-date data from disk
785	 * because journal recovery will change the filesystem under us.
786	 */
787	if (fs->super->s_kbytes_written &&
788	    fs->io->manager->get_stats)
789		fs->io->manager->get_stats(fs->io, &stats);
790	if (stats && stats->bytes_written)
791		kbytes_written = stats->bytes_written >> 10;
792
793	ext2fs_mmp_stop(fs);
794	fsname = strdup(fs->device_name);
795	fsflags = fs->flags;
796	fsblocksize = fs->blocksize;
797	ext2fs_free(fs);
798	retval = ext2fs_open(fsname, fsflags,
799			     0, fsblocksize, io_ptr,
800			     fsp);
801	free(fsname);
802	if (retval)
803		return retval;
804
805	fs = *fsp;
806	fs->flags |= EXT2_FLAG_MASTER_SB_ONLY;
807	fs->super->s_kbytes_written += kbytes_written;
808
809	/* Set the superblock flags */
810	ext2fs_clear_recover(fs, recover_retval != 0);
811
812	/*
813	 * Do one last sanity check, and propagate journal->s_errno to
814	 * the EXT2_ERROR_FS flag in the fs superblock if needed.
815	 */
816	retval = ext2fs_check_ext3_journal(fs);
817	return retval ? retval : recover_retval;
818}
819
820errcode_t ext2fs_open_journal(ext2_filsys fs, journal_t **j)
821{
822	journal_t *journal;
823	errcode_t retval;
824
825	journal_init_revoke_caches();
826	retval = ext2fs_get_journal(fs, &journal);
827	if (retval)
828		return retval;
829
830	retval = ext2fs_journal_load(journal);
831	if (retval)
832		goto errout;
833
834	retval = journal_init_revoke(journal, 1024);
835	if (retval)
836		goto errout;
837
838	if (journal->j_failed_commit) {
839		journal->j_superblock->s_errno = -EINVAL;
840		mark_buffer_dirty(journal->j_sb_buffer);
841	}
842
843	*j = journal;
844	return 0;
845
846errout:
847	journal_destroy_revoke(journal);
848	journal_destroy_revoke_caches();
849	ext2fs_journal_release(fs, journal, 1, 0);
850	return retval;
851}
852
853errcode_t ext2fs_close_journal(ext2_filsys fs, journal_t **j)
854{
855	journal_t *journal = *j;
856
857	journal_destroy_revoke(journal);
858	journal_destroy_revoke_caches();
859	ext2fs_journal_release(fs, journal, 0, 0);
860	*j = NULL;
861
862	return 0;
863}
864
865void jbd2_commit_block_csum_set(journal_t *j, struct buffer_head *bh)
866{
867	struct commit_header *h;
868	__u32 csum;
869
870	if (!journal_has_csum_v2or3(j))
871		return;
872
873	h = (struct commit_header *)(bh->b_data);
874	h->h_chksum_type = 0;
875	h->h_chksum_size = 0;
876	h->h_chksum[0] = 0;
877	csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
878	h->h_chksum[0] = ext2fs_cpu_to_be32(csum);
879}
880
881void jbd2_revoke_csum_set(journal_t *j, struct buffer_head *bh)
882{
883	struct journal_revoke_tail *tail;
884	__u32 csum;
885
886	if (!journal_has_csum_v2or3(j))
887		return;
888
889	tail = (struct journal_revoke_tail *)(bh->b_data + j->j_blocksize -
890			sizeof(struct journal_revoke_tail));
891	tail->r_checksum = 0;
892	csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
893	tail->r_checksum = ext2fs_cpu_to_be32(csum);
894}
895
896void jbd2_descr_block_csum_set(journal_t *j, struct buffer_head *bh)
897{
898	struct journal_block_tail *tail;
899	__u32 csum;
900
901	if (!journal_has_csum_v2or3(j))
902		return;
903
904	tail = (struct journal_block_tail *)(bh->b_data + j->j_blocksize -
905			sizeof(struct journal_block_tail));
906	tail->t_checksum = 0;
907	csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
908	tail->t_checksum = ext2fs_cpu_to_be32(csum);
909}
910
911void jbd2_block_tag_csum_set(journal_t *j, journal_block_tag_t *tag,
912			     struct buffer_head *bh, __u32 sequence)
913{
914	journal_block_tag3_t *tag3 = (journal_block_tag3_t *)tag;
915	__u32 csum32;
916	__be32 seq;
917
918	if (!journal_has_csum_v2or3(j))
919		return;
920
921	seq = ext2fs_cpu_to_be32(sequence);
922	csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
923	csum32 = jbd2_chksum(j, csum32, bh->b_data, bh->b_size);
924
925	if (jfs_has_feature_csum3(j))
926		tag3->t_checksum = ext2fs_cpu_to_be32(csum32);
927	else
928		tag->t_checksum = ext2fs_cpu_to_be16(csum32);
929}
930
931