On 16-Mar-06, at 3:20 PM, Manuel Bouyer wrote:

> Hi,
> can you try the attached patch ? obviously m_copyback() doesn't work
> the way I though it was, and as it's not even capable of allocating
> a cluster when needeed it's probably less efficient than plain  
> MCLGET/memcpy.

I ran two cycles of cvs co and rdiff-backup and no panic. In the  
past, that was enough to trigger the problem.

-- 
     aew