cluster: STABLE3 - GFS2: gfs2_fsck should fix journal sequence numberproblems

[Bob Peterson <rpeterso at fedoraproject dot org>]

Gitweb:        http://git.fedorahosted.org/git/cluster.git?p=cluster.git;a=commitdiff;h=9f93e46ce2307418d3e0a73a6b78ca51ece479d1
Commit:        9f93e46ce2307418d3e0a73a6b78ca51ece479d1
Parent:        e793a85a83fb9e6a0ecfce33a1a9883abfe6c177
Author:        Bob Peterson <rpeterso@redhat.com>
AuthorDate:    Fri May 1 16:34:05 2009 -0500
Committer:     Bob Peterson <rpeterso@redhat.com>
CommitterDate: Fri May 1 16:34:05 2009 -0500

GFS2: gfs2_fsck should fix journal sequence number problems

bz 498646

Due to bug #457557 or if a customer accidentally mounts their
file system as "lock_nolock" from two nodes, the journals
can have sequence number problems.  These sequence number
problems prevent gfs2 from ever mounting the file system again.
Therefore, this enables gfs2_fsck to detect and repair most
sequence number problems.
---
 gfs2/fsck/fs_recovery.c |   76 +++++++++++++++++++++++++++++++++++++++++++++--
 1 files changed, 73 insertions(+), 3 deletions(-)

diff --git a/gfs2/fsck/fs_recovery.c b/gfs2/fsck/fs_recovery.c
index fb830fe..637e8c1 100644
--- a/gfs2/fsck/fs_recovery.c
+++ b/gfs2/fsck/fs_recovery.c
@@ -318,6 +318,59 @@ int foreach_descriptor(struct gfs2_inode *ip, unsigned int start,
 }
 
 /**
+ * fix_journal_seq_no - Fix log header sequencing problems
+ * @ip: the journal incore inode
+ */
+int fix_journal_seq_no(struct gfs2_inode *ip)
+{
+	int error = 0, wrapped = 0;
+	uint32_t jd_blocks = ip->i_di.di_size / ip->i_sbd->sd_sb.sb_bsize;
+	uint32_t blk;
+	struct gfs2_log_header lh;
+	uint64_t highest_seq = 0, lowest_seq = 0, prev_seq = 0;
+	int new = 0;
+	uint64_t dblock;
+	uint32_t extlen;
+	struct gfs2_buffer_head *bh;
+
+	for (blk = 0; blk < jd_blocks; blk++) {
+		error = get_log_header(ip, blk, &lh);
+		if (error == 1) /* if not a log header */
+			continue; /* just journal data--ignore it */
+		if (!lowest_seq || lh.lh_sequence < lowest_seq)
+			lowest_seq = lh.lh_sequence;
+		if (!highest_seq || lh.lh_sequence > highest_seq)
+			highest_seq = lh.lh_sequence;
+		if (lh.lh_sequence > prev_seq) {
+			prev_seq = lh.lh_sequence;
+			continue;
+		}
+		/* The sequence number is not higher than the previous one,
+		   so it's either wrap-around or a sequencing problem. */
+		if (!wrapped && lh.lh_sequence == lowest_seq) {
+			wrapped = 1;
+			prev_seq = lh.lh_sequence;
+			continue;
+		}
+		log_err("Journal block %u (0x%x): sequence no. 0x%llx out of "
+			"order.\n", blk, blk, lh.lh_sequence);
+		log_info("Low: 0x%llx, High: 0x%llx, Prev: 0x%llx\n",
+			 (unsigned long long)lowest_seq,
+			 (unsigned long long)highest_seq,
+			 (unsigned long long)prev_seq);
+		highest_seq++;
+		lh.lh_sequence = highest_seq;
+		prev_seq = lh.lh_sequence;
+		log_warn("Renumbering it as 0x%llx\n", lh.lh_sequence);
+		block_map(ip, blk, &new, &dblock, &extlen, FALSE, not_updated);
+		bh = bread(&ip->i_sbd->buf_list, dblock);
+		gfs2_log_header_out(&lh, bh->b_data);
+		brelse(bh, updated);
+	}
+	return 0;
+}
+
+/**
  * gfs2_recover_journal - recovery a given journal
  * @ip: the journal incore inode
  *
@@ -338,9 +391,26 @@ int gfs2_recover_journal(struct gfs2_inode *ip, int j)
 
 	osi_list_init(&sd_revoke_list);
 	error = gfs2_find_jhead(ip, &head);
-	if (error)
-		goto out;
-
+	if (error) {
+		if (!query(&opts, "\nJournal #%d (\"journal%d\") is corrupt.  "
+			   "Okay to repair it? (y/n)", j+1, j)) {
+			log_err("jid=%u: The journal was not repaired.\n", j);
+			goto out;
+		}
+		log_info("jid=%u: Repairing journal...\n", j);
+		error = fix_journal_seq_no(ip);
+		if (error) {
+			log_err("jid=%u: Unable to repair the bad journal.\n",
+				j);
+			goto out;
+		}
+		error = gfs2_find_jhead(ip, &head);
+		if (error) {
+			log_err("jid=%u: Unable to fix the bad journal.\n", j);
+			goto out;
+		}
+		log_err("jid=%u: The journal was successfully fixed.\n", j);
+	}
 	if (head.lh_flags & GFS2_LOG_HEAD_UNMOUNT) {
 		log_info("jid=%u: Journal is clean.\n", j);
 		return 0;