2 * journal.c --- code for handling the "ext3" journal
4 * Copyright (C) 2000 Andreas Dilger
5 * Copyright (C) 2000 Theodore Ts'o
7 * Parts of the code are based on fs/jfs/journal.c by Stephen C. Tweedie
8 * Copyright (C) 1999 Red Hat Software
10 * This file may be redistributed under the terms of the
11 * GNU General Public License version 2 or at your discretion
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)
21 #ifdef HAVE_SYS_STAT_H
25 #define E2FSCK_INCLUDE_INLINE_FUNCS
28 #include "uuid/uuid.h"
30 #ifdef CONFIG_JBD_DEBUG /* Enabled by configure --enable-jfs-debug */
31 static int bh_count = 0;
35 * Define USE_INODE_IO to use the inode_io.c / fileio.c codepaths.
36 * This creates a larger static binary, and a smaller binary using
37 * shared libraries. It's also probably slightly less CPU-efficient,
38 * which is why it's not on by default. But, it's a good way of
39 * testing the functions in inode_io.c and fileio.c.
43 /* Checksumming functions */
44 static int e2fsck_journal_verify_csum_type(journal_t *j,
45 journal_superblock_t *jsb)
47 if (!jbd2_journal_has_csum_v2or3(j))
50 return jsb->s_checksum_type == JBD2_CRC32C_CHKSUM;
53 static __u32 e2fsck_journal_sb_csum(journal_superblock_t *jsb)
57 old_crc = jsb->s_checksum;
59 crc = ext2fs_crc32c_le(~0, (unsigned char *)jsb,
60 sizeof(journal_superblock_t));
61 jsb->s_checksum = old_crc;
66 static int e2fsck_journal_sb_csum_verify(journal_t *j,
67 journal_superblock_t *jsb)
69 __u32 provided, calculated;
71 if (!jbd2_journal_has_csum_v2or3(j))
74 provided = ext2fs_be32_to_cpu(jsb->s_checksum);
75 calculated = e2fsck_journal_sb_csum(jsb);
77 return provided == calculated;
80 static errcode_t e2fsck_journal_sb_csum_set(journal_t *j,
81 journal_superblock_t *jsb)
85 if (!jbd2_journal_has_csum_v2or3(j))
88 crc = e2fsck_journal_sb_csum(jsb);
89 jsb->s_checksum = ext2fs_cpu_to_be32(crc);
93 /* Kernel compatibility functions for handling the journal. These allow us
94 * to use the recovery.c file virtually unchanged from the kernel, so we
95 * don't have to do much to keep kernel and user recovery in sync.
97 int jbd2_journal_bmap(journal_t *journal, unsigned long block,
98 unsigned long long *phys)
104 struct inode *inode = journal->j_inode;
113 retval= ext2fs_bmap2(inode->i_ctx->fs, inode->i_ino,
114 &inode->i_ext2, NULL, 0, (blk64_t) block,
117 return -1 * ((int) retval);
121 struct buffer_head *getblk(kdev_t kdev, unsigned long long blocknr,
124 struct buffer_head *bh;
125 int bufsize = sizeof(*bh) + kdev->k_ctx->fs->blocksize -
128 bh = e2fsck_allocate_memory(kdev->k_ctx, bufsize, "block buffer");
132 #ifdef CONFIG_JBD_DEBUG
133 if (journal_enable_debug >= 3)
136 jfs_debug(4, "getblk for block %llu (%d bytes)(total %d)\n",
137 blocknr, blocksize, bh_count);
139 bh->b_ctx = kdev->k_ctx;
140 if (kdev->k_dev == K_DEV_FS)
141 bh->b_io = kdev->k_ctx->fs->io;
143 bh->b_io = kdev->k_ctx->journal_io;
144 bh->b_size = blocksize;
145 bh->b_blocknr = blocknr;
150 int sync_blockdev(kdev_t kdev)
154 if (kdev->k_dev == K_DEV_FS)
155 io = kdev->k_ctx->fs->io;
157 io = kdev->k_ctx->journal_io;
159 return io_channel_flush(io) ? -EIO : 0;
162 void ll_rw_block(int rw, int op_flags, int nr, struct buffer_head *bhp[])
165 struct buffer_head *bh;
167 for (; nr > 0; --nr) {
169 if (rw == REQ_OP_READ && !bh->b_uptodate) {
170 jfs_debug(3, "reading block %llu/%p\n",
171 bh->b_blocknr, (void *) bh);
172 retval = io_channel_read_blk64(bh->b_io,
176 com_err(bh->b_ctx->device_name, retval,
177 "while reading block %llu\n",
179 bh->b_err = (int) retval;
183 } else if (rw == REQ_OP_WRITE && bh->b_dirty) {
184 jfs_debug(3, "writing block %llu/%p\n",
187 retval = io_channel_write_blk64(bh->b_io,
191 com_err(bh->b_ctx->device_name, retval,
192 "while writing block %llu\n",
194 bh->b_err = (int) retval;
200 jfs_debug(3, "no-op %s for block %llu\n",
201 rw == REQ_OP_READ ? "read" : "write",
207 void mark_buffer_dirty(struct buffer_head *bh)
212 static void mark_buffer_clean(struct buffer_head * bh)
217 void brelse(struct buffer_head *bh)
220 ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
221 jfs_debug(3, "freeing block %llu/%p (total %d)\n",
222 bh->b_blocknr, (void *) bh, --bh_count);
223 ext2fs_free_mem(&bh);
226 int buffer_uptodate(struct buffer_head *bh)
228 return bh->b_uptodate;
231 void mark_buffer_uptodate(struct buffer_head *bh, int val)
233 bh->b_uptodate = val;
236 void wait_on_buffer(struct buffer_head *bh)
239 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
243 static void e2fsck_clear_recover(e2fsck_t ctx, int error)
245 ext2fs_clear_feature_journal_needs_recovery(ctx->fs->super);
247 /* if we had an error doing journal recovery, we need a full fsck */
249 ctx->fs->super->s_state &= ~EXT2_VALID_FS;
250 ext2fs_mark_super_dirty(ctx->fs);
254 * This is a helper function to check the validity of the journal.
256 struct process_block_struct {
257 e2_blkcnt_t last_block;
260 static int process_journal_block(ext2_filsys fs,
262 e2_blkcnt_t blockcnt,
263 blk64_t ref_block EXT2FS_ATTR((unused)),
264 int ref_offset EXT2FS_ATTR((unused)),
267 struct process_block_struct *p;
268 blk64_t blk = *block_nr;
270 p = (struct process_block_struct *) priv_data;
272 if (!blk || blk < fs->super->s_first_data_block ||
273 blk >= ext2fs_blocks_count(fs->super))
277 p->last_block = blockcnt;
281 static errcode_t e2fsck_get_journal(e2fsck_t ctx, journal_t **ret_journal)
283 struct process_block_struct pb;
284 struct ext2_super_block *sb = ctx->fs->super;
285 struct ext2_super_block jsuper;
286 struct problem_context pctx;
287 struct buffer_head *bh;
288 struct inode *j_inode = NULL;
289 struct kdev_s *dev_fs = NULL, *dev_journal;
290 const char *journal_name = 0;
291 journal_t *journal = NULL;
292 errcode_t retval = 0;
293 io_manager io_ptr = 0;
294 unsigned long long start = 0;
297 int tried_backup_jnl = 0;
299 clear_problem_context(&pctx);
301 journal = e2fsck_allocate_memory(ctx, sizeof(journal_t), "journal");
303 return EXT2_ET_NO_MEMORY;
306 dev_fs = e2fsck_allocate_memory(ctx, 2*sizeof(struct kdev_s), "kdev");
308 retval = EXT2_ET_NO_MEMORY;
311 dev_journal = dev_fs+1;
313 dev_fs->k_ctx = dev_journal->k_ctx = ctx;
314 dev_fs->k_dev = K_DEV_FS;
315 dev_journal->k_dev = K_DEV_JOURNAL;
317 journal->j_dev = dev_journal;
318 journal->j_fs_dev = dev_fs;
319 journal->j_inode = NULL;
320 journal->j_blocksize = ctx->fs->blocksize;
322 if (uuid_is_null(sb->s_journal_uuid)) {
323 if (!sb->s_journal_inum) {
324 retval = EXT2_ET_BAD_INODE_NUM;
327 j_inode = e2fsck_allocate_memory(ctx, sizeof(*j_inode),
330 retval = EXT2_ET_NO_MEMORY;
334 j_inode->i_ctx = ctx;
335 j_inode->i_ino = sb->s_journal_inum;
337 if ((retval = ext2fs_read_inode(ctx->fs,
339 &j_inode->i_ext2))) {
341 if (sb->s_jnl_backup_type != EXT3_JNL_BACKUP_BLOCKS ||
344 memset(&j_inode->i_ext2, 0, sizeof(struct ext2_inode));
345 memcpy(&j_inode->i_ext2.i_block[0], sb->s_jnl_blocks,
347 j_inode->i_ext2.i_size_high = sb->s_jnl_blocks[15];
348 j_inode->i_ext2.i_size = sb->s_jnl_blocks[16];
349 j_inode->i_ext2.i_links_count = 1;
350 j_inode->i_ext2.i_mode = LINUX_S_IFREG | 0600;
351 e2fsck_use_inode_shortcuts(ctx, 1);
352 ctx->stashed_ino = j_inode->i_ino;
353 ctx->stashed_inode = &j_inode->i_ext2;
356 if (!j_inode->i_ext2.i_links_count ||
357 !LINUX_S_ISREG(j_inode->i_ext2.i_mode)) {
358 retval = EXT2_ET_NO_JOURNAL;
359 goto try_backup_journal;
361 if (EXT2_I_SIZE(&j_inode->i_ext2) / journal->j_blocksize <
362 JBD2_MIN_JOURNAL_BLOCKS) {
363 retval = EXT2_ET_JOURNAL_TOO_SMALL;
364 goto try_backup_journal;
367 retval = ext2fs_block_iterate3(ctx->fs, j_inode->i_ino,
369 process_journal_block, &pb);
370 if ((pb.last_block + 1) * ctx->fs->blocksize <
371 (int) EXT2_I_SIZE(&j_inode->i_ext2)) {
372 retval = EXT2_ET_JOURNAL_TOO_SMALL;
373 goto try_backup_journal;
375 if (tried_backup_jnl && !(ctx->options & E2F_OPT_READONLY)) {
376 retval = ext2fs_write_inode(ctx->fs, sb->s_journal_inum,
382 journal->j_total_len = EXT2_I_SIZE(&j_inode->i_ext2) /
383 journal->j_blocksize;
386 retval = ext2fs_inode_io_intern2(ctx->fs, sb->s_journal_inum,
392 io_ptr = inode_io_manager;
394 journal->j_inode = j_inode;
395 ctx->journal_io = ctx->fs->io;
396 if ((ret = jbd2_journal_bmap(journal, 0, &start)) != 0) {
397 retval = (errcode_t) (-1 * ret);
403 if (!ctx->journal_name) {
406 uuid_unparse(sb->s_journal_uuid, uuid);
407 ctx->journal_name = blkid_get_devname(ctx->blkid,
409 if (!ctx->journal_name)
410 ctx->journal_name = blkid_devno_to_devname(sb->s_journal_dev);
412 journal_name = ctx->journal_name;
415 fix_problem(ctx, PR_0_CANT_FIND_JOURNAL, &pctx);
416 retval = EXT2_ET_LOAD_EXT_JOURNAL;
420 jfs_debug(1, "Using journal file %s\n", journal_name);
421 io_ptr = unix_io_manager;
425 test_io_backing_manager = io_ptr;
426 io_ptr = test_io_manager;
432 int flags = IO_FLAG_RW;
433 if (!(ctx->mount_flags & EXT2_MF_ISROOT &&
434 ctx->mount_flags & EXT2_MF_READONLY))
435 flags |= IO_FLAG_EXCLUSIVE;
436 if ((ctx->mount_flags & EXT2_MF_READONLY) &&
437 (ctx->options & E2F_OPT_FORCE))
438 flags &= ~IO_FLAG_EXCLUSIVE;
441 retval = io_ptr->open(journal_name, flags,
447 io_channel_set_blksize(ctx->journal_io, ctx->fs->blocksize);
452 start = ext2fs_journal_sb_start(ctx->fs->blocksize) - 1;
453 bh = getblk(dev_journal, start, ctx->fs->blocksize);
455 retval = EXT2_ET_NO_MEMORY;
458 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
459 if ((retval = bh->b_err) != 0) {
463 memcpy(&jsuper, start ? bh->b_data : bh->b_data + SUPERBLOCK_OFFSET,
465 #ifdef WORDS_BIGENDIAN
466 if (jsuper.s_magic == ext2fs_swab16(EXT2_SUPER_MAGIC))
467 ext2fs_swap_super(&jsuper);
469 if (jsuper.s_magic != EXT2_SUPER_MAGIC ||
470 !ext2fs_has_feature_journal_dev(&jsuper)) {
471 fix_problem(ctx, PR_0_EXT_JOURNAL_BAD_SUPER, &pctx);
472 retval = EXT2_ET_LOAD_EXT_JOURNAL;
476 /* Make sure the journal UUID is correct */
477 if (memcmp(jsuper.s_uuid, ctx->fs->super->s_journal_uuid,
478 sizeof(jsuper.s_uuid))) {
479 fix_problem(ctx, PR_0_JOURNAL_BAD_UUID, &pctx);
480 retval = EXT2_ET_LOAD_EXT_JOURNAL;
485 /* Check the superblock checksum */
486 if (ext2fs_has_feature_metadata_csum(&jsuper)) {
487 struct struct_ext2_filsys fsx;
488 struct ext2_super_block superx;
491 p = start ? bh->b_data : bh->b_data + SUPERBLOCK_OFFSET;
492 memcpy(&fsx, ctx->fs, sizeof(fsx));
493 memcpy(&superx, ctx->fs->super, sizeof(superx));
495 ext2fs_set_feature_metadata_csum(fsx.super);
496 if (!ext2fs_superblock_csum_verify(&fsx, p) &&
497 fix_problem(ctx, PR_0_EXT_JOURNAL_SUPER_CSUM_INVALID,
499 ext2fs_superblock_csum_set(&fsx, p);
500 mark_buffer_dirty(bh);
505 maxlen = ext2fs_blocks_count(&jsuper);
506 journal->j_total_len = (maxlen < 1ULL << 32) ? maxlen : (1ULL << 32) - 1;
510 if (!(bh = getblk(dev_journal, start, journal->j_blocksize))) {
511 retval = EXT2_ET_NO_MEMORY;
515 journal->j_sb_buffer = bh;
516 journal->j_superblock = (journal_superblock_t *)bh->b_data;
520 ext2fs_free_mem(&j_inode);
523 *ret_journal = journal;
524 e2fsck_use_inode_shortcuts(ctx, 0);
528 e2fsck_use_inode_shortcuts(ctx, 0);
530 ext2fs_free_mem(&dev_fs);
532 ext2fs_free_mem(&j_inode);
534 ext2fs_free_mem(&journal);
538 static errcode_t e2fsck_journal_fix_bad_inode(e2fsck_t ctx,
539 struct problem_context *pctx)
541 struct ext2_super_block *sb = ctx->fs->super;
542 int recover = ext2fs_has_feature_journal_needs_recovery(ctx->fs->super);
543 int has_journal = ext2fs_has_feature_journal(ctx->fs->super);
545 if (has_journal || sb->s_journal_inum) {
546 /* The journal inode is bogus, remove and force full fsck */
547 pctx->ino = sb->s_journal_inum;
548 if (fix_problem(ctx, PR_0_JOURNAL_BAD_INODE, pctx)) {
549 if (has_journal && sb->s_journal_inum)
550 printf("*** journal has been deleted ***\n\n");
551 ext2fs_clear_feature_journal(sb);
552 sb->s_journal_inum = 0;
553 memset(sb->s_jnl_blocks, 0, sizeof(sb->s_jnl_blocks));
554 ctx->flags |= E2F_FLAG_JOURNAL_INODE;
555 ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
556 e2fsck_clear_recover(ctx, 1);
559 return EXT2_ET_CORRUPT_JOURNAL_SB;
560 } else if (recover) {
561 if (fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, pctx)) {
562 e2fsck_clear_recover(ctx, 1);
565 return EXT2_ET_UNSUPP_FEATURE;
570 #define V1_SB_SIZE 0x0024
571 static void clear_v2_journal_fields(journal_t *journal)
573 e2fsck_t ctx = journal->j_dev->k_ctx;
574 struct problem_context pctx;
576 clear_problem_context(&pctx);
578 if (!fix_problem(ctx, PR_0_CLEAR_V2_JOURNAL, &pctx))
581 ctx->flags |= E2F_FLAG_PROBLEMS_FIXED;
582 memset(((char *) journal->j_superblock) + V1_SB_SIZE, 0,
583 ctx->fs->blocksize-V1_SB_SIZE);
584 mark_buffer_dirty(journal->j_sb_buffer);
588 static errcode_t e2fsck_journal_load(journal_t *journal)
590 e2fsck_t ctx = journal->j_dev->k_ctx;
591 journal_superblock_t *jsb;
592 struct buffer_head *jbh = journal->j_sb_buffer;
593 struct problem_context pctx;
595 clear_problem_context(&pctx);
597 ll_rw_block(REQ_OP_READ, 0, 1, &jbh);
599 com_err(ctx->device_name, jbh->b_err, "%s",
600 _("reading journal superblock\n"));
604 jsb = journal->j_superblock;
605 /* If we don't even have JBD2_MAGIC, we probably have a wrong inode */
606 if (jsb->s_header.h_magic != htonl(JBD2_MAGIC_NUMBER))
607 return e2fsck_journal_fix_bad_inode(ctx, &pctx);
609 switch (ntohl(jsb->s_header.h_blocktype)) {
610 case JBD2_SUPERBLOCK_V1:
611 journal->j_format_version = 1;
612 if (jsb->s_feature_compat ||
613 jsb->s_feature_incompat ||
614 jsb->s_feature_ro_compat ||
616 clear_v2_journal_fields(journal);
619 case JBD2_SUPERBLOCK_V2:
620 journal->j_format_version = 2;
621 if (ntohl(jsb->s_nr_users) > 1 &&
622 uuid_is_null(ctx->fs->super->s_journal_uuid))
623 clear_v2_journal_fields(journal);
624 if (ntohl(jsb->s_nr_users) > 1) {
625 fix_problem(ctx, PR_0_JOURNAL_UNSUPP_MULTIFS, &pctx);
626 return EXT2_ET_JOURNAL_UNSUPP_VERSION;
631 * These should never appear in a journal super block, so if
632 * they do, the journal is badly corrupted.
634 case JBD2_DESCRIPTOR_BLOCK:
635 case JBD2_COMMIT_BLOCK:
636 case JBD2_REVOKE_BLOCK:
637 return EXT2_ET_CORRUPT_JOURNAL_SB;
639 /* If we don't understand the superblock major type, but there
640 * is a magic number, then it is likely to be a new format we
641 * just don't understand, so leave it alone. */
643 return EXT2_ET_JOURNAL_UNSUPP_VERSION;
646 if (JBD2_HAS_INCOMPAT_FEATURE(journal, ~JBD2_KNOWN_INCOMPAT_FEATURES))
647 return EXT2_ET_UNSUPP_FEATURE;
649 if (JBD2_HAS_RO_COMPAT_FEATURE(journal, ~JBD2_KNOWN_ROCOMPAT_FEATURES))
650 return EXT2_ET_RO_UNSUPP_FEATURE;
652 /* Checksum v1-3 are mutually exclusive features. */
653 if (jbd2_has_feature_csum2(journal) && jbd2_has_feature_csum3(journal))
654 return EXT2_ET_CORRUPT_JOURNAL_SB;
656 if (jbd2_journal_has_csum_v2or3(journal) &&
657 jbd2_has_feature_checksum(journal))
658 return EXT2_ET_CORRUPT_JOURNAL_SB;
660 if (!e2fsck_journal_verify_csum_type(journal, jsb) ||
661 !e2fsck_journal_sb_csum_verify(journal, jsb))
662 return EXT2_ET_CORRUPT_JOURNAL_SB;
664 if (jbd2_journal_has_csum_v2or3(journal))
665 journal->j_csum_seed = jbd2_chksum(journal, ~0, jsb->s_uuid,
666 sizeof(jsb->s_uuid));
668 /* We have now checked whether we know enough about the journal
669 * format to be able to proceed safely, so any other checks that
670 * fail we should attempt to recover from. */
671 if (jsb->s_blocksize != htonl(journal->j_blocksize)) {
672 com_err(ctx->program_name, EXT2_ET_CORRUPT_JOURNAL_SB,
673 _("%s: no valid journal superblock found\n"),
675 return EXT2_ET_CORRUPT_JOURNAL_SB;
678 if (ntohl(jsb->s_maxlen) < journal->j_total_len)
679 journal->j_total_len = ntohl(jsb->s_maxlen);
680 else if (ntohl(jsb->s_maxlen) > journal->j_total_len) {
681 com_err(ctx->program_name, EXT2_ET_CORRUPT_JOURNAL_SB,
682 _("%s: journal too short\n"),
684 return EXT2_ET_CORRUPT_JOURNAL_SB;
687 journal->j_tail_sequence = ntohl(jsb->s_sequence);
688 journal->j_transaction_sequence = journal->j_tail_sequence;
689 journal->j_tail = ntohl(jsb->s_start);
690 journal->j_first = ntohl(jsb->s_first);
691 if (jbd2_has_feature_fast_commit(journal)) {
692 if (ntohl(jsb->s_maxlen) - jbd2_journal_get_num_fc_blks(jsb)
693 < JBD2_MIN_JOURNAL_BLOCKS) {
694 com_err(ctx->program_name, EXT2_ET_CORRUPT_JOURNAL_SB,
695 _("%s: incorrect fast commit blocks\n"),
697 return EXT2_ET_CORRUPT_JOURNAL_SB;
699 journal->j_fc_last = ntohl(jsb->s_maxlen);
700 journal->j_last = journal->j_fc_last -
701 jbd2_journal_get_num_fc_blks(jsb);
702 journal->j_fc_first = journal->j_last + 1;
704 journal->j_last = ntohl(jsb->s_maxlen);
710 static void e2fsck_journal_reset_super(e2fsck_t ctx, journal_superblock_t *jsb,
721 /* Leave a valid existing V1 superblock signature alone.
722 * Anything unrecognisable we overwrite with a new V2
725 if (jsb->s_header.h_magic != htonl(JBD2_MAGIC_NUMBER) ||
726 jsb->s_header.h_blocktype != htonl(JBD2_SUPERBLOCK_V1)) {
727 jsb->s_header.h_magic = htonl(JBD2_MAGIC_NUMBER);
728 jsb->s_header.h_blocktype = htonl(JBD2_SUPERBLOCK_V2);
731 /* Zero out everything else beyond the superblock header */
733 p = ((char *) jsb) + sizeof(journal_header_t);
734 memset (p, 0, ctx->fs->blocksize-sizeof(journal_header_t));
736 jsb->s_blocksize = htonl(ctx->fs->blocksize);
737 jsb->s_maxlen = htonl(journal->j_total_len);
738 jsb->s_first = htonl(1);
740 /* Initialize the journal sequence number so that there is "no"
741 * chance we will find old "valid" transactions in the journal.
742 * This avoids the need to zero the whole journal (slow to do,
743 * and risky when we are just recovering the filesystem).
745 uuid_generate(u.uuid);
746 for (i = 0; i < 4; i ++)
748 jsb->s_sequence = htonl(new_seq);
749 e2fsck_journal_sb_csum_set(journal, jsb);
751 mark_buffer_dirty(journal->j_sb_buffer);
752 ll_rw_block(REQ_OP_WRITE, 0, 1, &journal->j_sb_buffer);
755 static errcode_t e2fsck_journal_fix_corrupt_super(e2fsck_t ctx,
757 struct problem_context *pctx)
759 struct ext2_super_block *sb = ctx->fs->super;
760 int recover = ext2fs_has_feature_journal_needs_recovery(ctx->fs->super);
762 if (ext2fs_has_feature_journal(sb)) {
763 if (fix_problem(ctx, PR_0_JOURNAL_BAD_SUPER, pctx)) {
764 e2fsck_journal_reset_super(ctx, journal->j_superblock,
766 journal->j_transaction_sequence = 1;
767 e2fsck_clear_recover(ctx, recover);
770 return EXT2_ET_CORRUPT_JOURNAL_SB;
771 } else if (e2fsck_journal_fix_bad_inode(ctx, pctx))
772 return EXT2_ET_CORRUPT_JOURNAL_SB;
777 static void e2fsck_journal_release(e2fsck_t ctx, journal_t *journal,
780 journal_superblock_t *jsb;
783 mark_buffer_clean(journal->j_sb_buffer);
784 else if (!(ctx->options & E2F_OPT_READONLY)) {
785 jsb = journal->j_superblock;
786 jsb->s_sequence = htonl(journal->j_tail_sequence);
788 jsb->s_start = 0; /* this marks the journal as empty */
789 e2fsck_journal_sb_csum_set(journal, jsb);
790 mark_buffer_dirty(journal->j_sb_buffer);
792 brelse(journal->j_sb_buffer);
794 if (ctx->journal_io) {
795 if (ctx->fs && ctx->fs->io != ctx->journal_io)
796 io_channel_close(ctx->journal_io);
801 if (journal->j_inode)
802 ext2fs_free_mem(&journal->j_inode);
804 if (journal->j_fs_dev)
805 ext2fs_free_mem(&journal->j_fs_dev);
806 ext2fs_free_mem(&journal);
810 * This function makes sure that the superblock fields regarding the
811 * journal are consistent.
813 errcode_t e2fsck_check_ext3_journal(e2fsck_t ctx)
815 struct ext2_super_block *sb = ctx->fs->super;
817 int recover = ext2fs_has_feature_journal_needs_recovery(ctx->fs->super);
818 struct problem_context pctx;
820 int reset = 0, force_fsck = 0;
823 /* If we don't have any journal features, don't do anything more */
824 if (!ext2fs_has_feature_journal(sb) &&
825 !recover && sb->s_journal_inum == 0 && sb->s_journal_dev == 0 &&
826 uuid_is_null(sb->s_journal_uuid))
829 clear_problem_context(&pctx);
830 pctx.num = sb->s_journal_inum;
832 retval = e2fsck_get_journal(ctx, &journal);
834 if ((retval == EXT2_ET_BAD_INODE_NUM) ||
835 (retval == EXT2_ET_BAD_BLOCK_NUM) ||
836 (retval == EXT2_ET_JOURNAL_TOO_SMALL) ||
837 (retval == EXT2_ET_NO_JOURNAL))
838 return e2fsck_journal_fix_bad_inode(ctx, &pctx);
842 retval = e2fsck_journal_load(journal);
844 if ((retval == EXT2_ET_CORRUPT_JOURNAL_SB) ||
845 ((retval == EXT2_ET_UNSUPP_FEATURE) &&
846 (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_INCOMPAT,
848 ((retval == EXT2_ET_RO_UNSUPP_FEATURE) &&
849 (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_ROCOMPAT,
851 ((retval == EXT2_ET_JOURNAL_UNSUPP_VERSION) &&
852 (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_VERSION, &pctx))))
853 retval = e2fsck_journal_fix_corrupt_super(ctx, journal,
855 e2fsck_journal_release(ctx, journal, 0, 1);
860 * We want to make the flags consistent here. We will not leave with
861 * needs_recovery set but has_journal clear. We can't get in a loop
862 * with -y, -n, or -p, only if a user isn't making up their mind.
865 if (!ext2fs_has_feature_journal(sb)) {
866 recover = ext2fs_has_feature_journal_needs_recovery(sb);
867 if (fix_problem(ctx, PR_0_JOURNAL_HAS_JOURNAL, &pctx)) {
869 !fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, &pctx))
872 * Need a full fsck if we are releasing a
873 * journal stored on a reserved inode.
875 force_fsck = recover ||
876 (sb->s_journal_inum < EXT2_FIRST_INODE(sb));
877 /* Clear all of the journal fields */
878 sb->s_journal_inum = 0;
879 sb->s_journal_dev = 0;
880 memset(sb->s_journal_uuid, 0,
881 sizeof(sb->s_journal_uuid));
882 e2fsck_clear_recover(ctx, force_fsck);
883 } else if (!(ctx->options & E2F_OPT_READONLY)) {
884 ext2fs_set_feature_journal(sb);
885 ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
886 ext2fs_mark_super_dirty(ctx->fs);
890 if (ext2fs_has_feature_journal(sb) &&
891 !ext2fs_has_feature_journal_needs_recovery(sb) &&
892 journal->j_superblock->s_start != 0) {
893 /* Print status information */
894 fix_problem(ctx, PR_0_JOURNAL_RECOVERY_CLEAR, &pctx);
896 problem = PR_0_JOURNAL_RUN_DEFAULT;
898 problem = PR_0_JOURNAL_RUN;
899 if (fix_problem(ctx, problem, &pctx)) {
900 ctx->options |= E2F_OPT_FORCE;
901 ext2fs_set_feature_journal_needs_recovery(sb);
902 ext2fs_mark_super_dirty(ctx->fs);
903 } else if (fix_problem(ctx,
904 PR_0_JOURNAL_RESET_JOURNAL, &pctx)) {
906 sb->s_state &= ~EXT2_VALID_FS;
907 ext2fs_mark_super_dirty(ctx->fs);
910 * If the user answers no to the above question, we
911 * ignore the fact that journal apparently has data;
912 * accidentally replaying over valid data would be far
913 * worse than skipping a questionable recovery.
915 * XXX should we abort with a fatal error here? What
916 * will the ext3 kernel code do if a filesystem with
917 * !NEEDS_RECOVERY but with a non-zero
918 * journal->j_superblock->s_start is mounted?
923 * If we don't need to do replay the journal, check to see if
924 * the journal's errno is set; if so, we need to mark the file
925 * system as being corrupt and clear the journal's s_errno.
927 if (!ext2fs_has_feature_journal_needs_recovery(sb) &&
928 journal->j_superblock->s_errno) {
929 ctx->fs->super->s_state |= EXT2_ERROR_FS;
930 ext2fs_mark_super_dirty(ctx->fs);
931 journal->j_superblock->s_errno = 0;
932 e2fsck_journal_sb_csum_set(journal, journal->j_superblock);
933 mark_buffer_dirty(journal->j_sb_buffer);
936 e2fsck_journal_release(ctx, journal, reset, 0);
940 static errcode_t recover_ext3_journal(e2fsck_t ctx)
942 struct problem_context pctx;
946 clear_problem_context(&pctx);
948 retval = jbd2_journal_init_revoke_record_cache();
952 retval = jbd2_journal_init_revoke_table_cache();
956 retval = e2fsck_get_journal(ctx, &journal);
960 retval = e2fsck_journal_load(journal);
964 retval = jbd2_journal_init_revoke(journal, 1024);
968 retval = -jbd2_journal_recover(journal);
972 if (journal->j_failed_commit) {
973 pctx.ino = journal->j_failed_commit;
974 fix_problem(ctx, PR_0_JNL_TXN_CORRUPT, &pctx);
975 journal->j_superblock->s_errno = -EINVAL;
976 mark_buffer_dirty(journal->j_sb_buffer);
979 journal->j_tail_sequence = journal->j_transaction_sequence;
982 jbd2_journal_destroy_revoke(journal);
983 jbd2_journal_destroy_revoke_record_cache();
984 jbd2_journal_destroy_revoke_table_cache();
985 e2fsck_journal_release(ctx, journal, 1, 0);
989 errcode_t e2fsck_run_ext3_journal(e2fsck_t ctx)
991 io_manager io_ptr = ctx->fs->io->manager;
992 int blocksize = ctx->fs->blocksize;
993 errcode_t retval, recover_retval;
995 unsigned long long kbytes_written = 0;
997 printf(_("%s: recovering journal\n"), ctx->device_name);
998 if (ctx->options & E2F_OPT_READONLY) {
999 printf(_("%s: won't do journal recovery while read-only\n"),
1001 return EXT2_ET_FILE_RO;
1004 if (ctx->fs->flags & EXT2_FLAG_DIRTY)
1005 ext2fs_flush(ctx->fs); /* Force out any modifications */
1007 recover_retval = recover_ext3_journal(ctx);
1010 * Reload the filesystem context to get up-to-date data from disk
1011 * because journal recovery will change the filesystem under us.
1013 if (ctx->fs->super->s_kbytes_written &&
1014 ctx->fs->io->manager->get_stats)
1015 ctx->fs->io->manager->get_stats(ctx->fs->io, &stats);
1016 if (stats && stats->bytes_written)
1017 kbytes_written = stats->bytes_written >> 10;
1019 ext2fs_mmp_stop(ctx->fs);
1020 ext2fs_free(ctx->fs);
1021 retval = ext2fs_open(ctx->filesystem_name, ctx->openfs_flags,
1022 ctx->superblock, blocksize, io_ptr,
1025 com_err(ctx->program_name, retval,
1026 _("while trying to re-open %s"),
1028 fatal_error(ctx, 0);
1030 ctx->fs->priv_data = ctx;
1031 ctx->fs->now = ctx->now;
1032 ctx->fs->flags |= EXT2_FLAG_MASTER_SB_ONLY;
1033 ctx->fs->super->s_kbytes_written += kbytes_written;
1035 /* Set the superblock flags */
1036 e2fsck_clear_recover(ctx, recover_retval != 0);
1039 * Do one last sanity check, and propagate journal->s_errno to
1040 * the EXT2_ERROR_FS flag in the fs superblock if needed.
1042 retval = e2fsck_check_ext3_journal(ctx);
1043 return retval ? retval : recover_retval;
1047 * This function will move the journal inode from a visible file in
1048 * the filesystem directory hierarchy to the reserved inode if necessary.
1050 static const char * const journal_names[] = {
1051 ".journal", "journal", ".journal.dat", "journal.dat", 0 };
1053 void e2fsck_move_ext3_journal(e2fsck_t ctx)
1055 struct ext2_super_block *sb = ctx->fs->super;
1056 struct problem_context pctx;
1057 struct ext2_inode inode;
1058 ext2_filsys fs = ctx->fs;
1061 const char * const * cpp;
1065 clear_problem_context(&pctx);
1068 * If the filesystem is opened read-only, or there is no
1069 * journal, then do nothing.
1071 if ((ctx->options & E2F_OPT_READONLY) ||
1072 (sb->s_journal_inum == 0) ||
1073 !ext2fs_has_feature_journal(sb))
1077 * Read in the journal inode
1079 if (ext2fs_read_inode(fs, sb->s_journal_inum, &inode) != 0)
1083 * If it's necessary to backup the journal inode, do so.
1085 if ((sb->s_jnl_backup_type == 0) ||
1086 ((sb->s_jnl_backup_type == EXT3_JNL_BACKUP_BLOCKS) &&
1087 memcmp(inode.i_block, sb->s_jnl_blocks, EXT2_N_BLOCKS*4))) {
1088 if (fix_problem(ctx, PR_0_BACKUP_JNL, &pctx)) {
1089 memcpy(sb->s_jnl_blocks, inode.i_block,
1091 sb->s_jnl_blocks[15] = inode.i_size_high;
1092 sb->s_jnl_blocks[16] = inode.i_size;
1093 sb->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
1094 ext2fs_mark_super_dirty(fs);
1095 fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1100 * If the journal is already the hidden inode, then do nothing
1102 if (sb->s_journal_inum == EXT2_JOURNAL_INO)
1106 * The journal inode had better have only one link and not be readable.
1108 if (inode.i_links_count != 1)
1112 * If the filesystem is mounted, or we can't tell whether
1113 * or not it's mounted, do nothing.
1115 retval = ext2fs_check_if_mounted(ctx->filesystem_name, &mount_flags);
1116 if (retval || (mount_flags & EXT2_MF_MOUNTED))
1120 * If we can't find the name of the journal inode, then do
1123 for (cpp = journal_names; *cpp; cpp++) {
1124 retval = ext2fs_lookup(fs, EXT2_ROOT_INO, *cpp,
1125 strlen(*cpp), 0, &ino);
1126 if ((retval == 0) && (ino == sb->s_journal_inum))
1132 /* We need the inode bitmap to be loaded */
1133 retval = ext2fs_read_bitmaps(fs);
1138 if (!fix_problem(ctx, PR_0_MOVE_JOURNAL, &pctx))
1142 * OK, we've done all the checks, let's actually move the
1143 * journal inode. Errors at this point mean we need to force
1144 * an ext2 filesystem check.
1146 if ((retval = ext2fs_unlink(fs, EXT2_ROOT_INO, *cpp, ino, 0)) != 0)
1148 if ((retval = ext2fs_write_inode(fs, EXT2_JOURNAL_INO, &inode)) != 0)
1150 sb->s_journal_inum = EXT2_JOURNAL_INO;
1151 ext2fs_mark_super_dirty(fs);
1152 fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1153 inode.i_links_count = 0;
1154 inode.i_dtime = ctx->now;
1155 if ((retval = ext2fs_write_inode(fs, ino, &inode)) != 0)
1158 group = ext2fs_group_of_ino(fs, ino);
1159 ext2fs_unmark_inode_bitmap2(fs->inode_map, ino);
1160 ext2fs_mark_ib_dirty(fs);
1161 ext2fs_bg_free_inodes_count_set(fs, group, ext2fs_bg_free_inodes_count(fs, group) + 1);
1162 ext2fs_group_desc_csum_set(fs, group);
1163 fs->super->s_free_inodes_count++;
1167 pctx.errcode = retval;
1168 fix_problem(ctx, PR_0_ERR_MOVE_JOURNAL, &pctx);
1169 fs->super->s_state &= ~EXT2_VALID_FS;
1170 ext2fs_mark_super_dirty(fs);
1175 * This function makes sure the superblock hint for the external
1176 * journal is correct.
1178 int e2fsck_fix_ext3_journal_hint(e2fsck_t ctx)
1180 struct ext2_super_block *sb = ctx->fs->super;
1181 struct problem_context pctx;
1182 char uuid[37], *journal_name;
1185 if (!ext2fs_has_feature_journal(sb) ||
1186 uuid_is_null(sb->s_journal_uuid))
1189 uuid_unparse(sb->s_journal_uuid, uuid);
1190 journal_name = blkid_get_devname(ctx->blkid, "UUID", uuid);
1194 if (stat(journal_name, &st) < 0) {
1199 if (st.st_rdev != sb->s_journal_dev) {
1200 clear_problem_context(&pctx);
1201 pctx.num = st.st_rdev;
1202 if (fix_problem(ctx, PR_0_EXTERNAL_JOURNAL_HINT, &pctx)) {
1203 sb->s_journal_dev = st.st_rdev;
1204 ext2fs_mark_super_dirty(ctx->fs);