ocfs2: correctly set i_blocks after inline dir gets expanded

We were setting i_blocks based on allocation before the extent insert, which
is wrong as the value is a calculation based on ip_clusters which gets
updated as a result of the insert. This patch moves the line in question
to just after the call to ocfs2_insert_extent().

Without this fix, inline directories were temporarily having an i_blocks
value of zero immediately after expansion to extents.

Reported-and-tested-by: Tristan Ye <tristan.ye@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>

diff --git a/fs/ocfs2/dir.c b/fs/ocfs2/dir.c
index 8e9c4a4..9cce563 100644 (file)
--- a/fs/ocfs2/dir.c
+++ b/fs/ocfs2/dir.c
@@ -1300,7 +1300,6 @@ static int ocfs2_expand_inline_dir(struct inode *dir, struct buffer_head *di_bh,
        di->i_size = cpu_to_le64(sb->s_blocksize);
        di->i_ctime = di->i_mtime = cpu_to_le64(dir->i_ctime.tv_sec);
        di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(dir->i_ctime.tv_nsec);
-       dir->i_blocks = ocfs2_inode_sector_count(dir);
 
        /*
         * This should never fail as our extent list is empty and all
@@ -1313,6 +1312,12 @@ static int ocfs2_expand_inline_dir(struct inode *dir, struct buffer_head *di_bh,
                goto out_commit;
        }
 
+       /*
+        * Set i_blocks after the extent insert for the most up to
+        * date ip_clusters value.
+        */
+       dir->i_blocks = ocfs2_inode_sector_count(dir);
+
        ret = ocfs2_journal_dirty(handle, di_bh);
        if (ret) {
                mlog_errno(ret);