ocfs2: Fix rename/extend race
authorSunil Mushran <sunil.musran@oracle.com>
Mon, 6 Aug 2007 22:11:56 +0000 (15:11 -0700)
committerMark Fasheh <mark.fasheh@oracle.com>
Fri, 10 Aug 2007 00:27:10 +0000 (17:27 -0700)
If one process is extending a file while another is renaming it, there
exists a window when rename could flush the old inode's stale i_size to
disk. This patch recognizes the fact that rename is only updating the old
inode's ctime, so it ensures only that value is flushed to disk.

Signed-off-by: Sunil Mushran <sunil.musran@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
fs/ocfs2/namei.c

index d430fdab16e9ea2e9bdad95bfc8346f87b26bb24..701e6d04ed5d2a7ed45fa9bb6270fa6f7fa91c50 100644 (file)
@@ -1080,6 +1080,7 @@ static int ocfs2_rename(struct inode *old_dir,
        struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
                                                    // this is the 1st dirent bh
        nlink_t old_dir_nlink = old_dir->i_nlink;
+       struct ocfs2_dinode *old_di;
 
        /* At some point it might be nice to break this function up a
         * bit. */
@@ -1354,7 +1355,20 @@ static int ocfs2_rename(struct inode *old_dir,
 
        old_inode->i_ctime = CURRENT_TIME;
        mark_inode_dirty(old_inode);
-       ocfs2_mark_inode_dirty(handle, old_inode, old_inode_bh);
+
+       status = ocfs2_journal_access(handle, old_inode, old_inode_bh,
+                                     OCFS2_JOURNAL_ACCESS_WRITE);
+       if (status >= 0) {
+               old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
+
+               old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
+               old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
+
+               status = ocfs2_journal_dirty(handle, old_inode_bh);
+               if (status < 0)
+                       mlog_errno(status);
+       } else
+               mlog_errno(status);
 
        /* now that the name has been added to new_dir, remove the old name */
        status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);