Re: [PATCH] update sb->s_frozen when freezing read-only mounted device, too

[Akinobu Mita <akinobu.mita@gmail.com>]

freeze_bdev() with read-only mounted device(*) does not change sb->s_frozen
from SB_UNFROZEN to SB_FREEZE_TRANS.

Because of this behavior, xfs_freeze can break read-only filesystem.

Because xfs_thaw does nothing for the filesystem whose sb->s_frozen is
SB_UNFROZEN. So frozen read-only XFS filesystem will never be unfrozen.
Thus we cannot do any unmount/remount operations for that filesystem.

This patch updates sb->s_frozen when freeze_bdev() is called for read-only
mounted device, too.

(*) freezing read-only filesystem is not so pointless.  Because it can
prevent from someone trying to remount read/write while freezing.

Cc: David Chinner <dgc@sgi.com>
Cc:  Tim Shimmin <xfs-masters@oss.sgi.com>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>

---
 fs/buffer.c |   25 ++++++++++++++++---------
 1 file changed, 16 insertions(+), 9 deletions(-)

Index: 2.6-git/fs/buffer.c
===================================================================
--- 2.6-git.orig/fs/buffer.c
+++ 2.6-git/fs/buffer.c
@@ -190,21 +190,28 @@ struct super_block *freeze_bdev(struct b
 
 	down(&bdev->bd_mount_sem);
 	sb = get_super(bdev);
-	if (sb && !(sb->s_flags & MS_RDONLY)) {
-		sb->s_frozen = SB_FREEZE_WRITE;
-		smp_wmb();
-
-		__fsync_super(sb);
+	if (!sb)
+		goto out;
 
+	if (sb->s_flags & MS_RDONLY) {
 		sb->s_frozen = SB_FREEZE_TRANS;
 		smp_wmb();
+		goto out;
+	}
 
-		sync_blockdev(sb->s_bdev);
+	sb->s_frozen = SB_FREEZE_WRITE;
+	smp_wmb();
 
-		if (sb->s_op->write_super_lockfs)
-			sb->s_op->write_super_lockfs(sb);
-	}
+	__fsync_super(sb);
+
+	sb->s_frozen = SB_FREEZE_TRANS;
+	smp_wmb();
 
+	sync_blockdev(sb->s_bdev);
+
+	if (sb->s_op->write_super_lockfs)
+		sb->s_op->write_super_lockfs(sb);
+out:
 	sync_blockdev(bdev);
 	return sb;	/* thaw_bdev releases s->s_umount and bd_mount_sem */
 }