For the record, this bug is also a  problem when working with distribute in that if a distribute child is not up at the time of initing libglusterfsclient, then the layout would not have been constructed completely for the root directory. Since we never send revalidates in libglusterfsclient for root, the root layout, which failed to get constructed during init phase, never gets constructed as the lookup onn the root is never sent.

This is one reason why we face the problem on holes and anomalies in distribute when a distribute child was not up during glusterfs_init. Although, the fix to this will not be the complete solution to the problem of "first access after libglusterfsclietn init", it will reduce the window within which that problem can occur.

PATCH: http://patches.gluster.com/patch/1482 in master (libglusterfsclient: Re-validate root inode on every path resolution)

PATCH: http://patches.gluster.com/patch/1481 in release-2.0 (libglusterfsclient: Re-validate root inode on every path resolution)