[OpenAFS-devel] Re: Change in openafs[master]: cmdebug -dcache

[Jeffrey Hutzelman]

On Mon, 2011-01-24 at 13:34 -0500, Jeffrey Altman wrote:
> cmdebug is a very OpenAFS implementation specific tool and protocol.
> When I implemented the cache manager rpc support for it I found that
> there is simply not a good mapping for all of the data from the Unix CM
> to the Windows CM.  I suspect there would be a significant challenge
> supporting Arla and kafs as well.
> 
> The purpose of cmdebug (for the most part) is to gain a real time view
> of the internals of the cache manager.  As such I question whether such
> a tool should be standardized.

Mostly, I agree with you.  Certain operations supported by cmdebug, such
as fetching host addresses, cell configuration, and maybe even some
aspects of cache configuration, are abstract enough to be standardized.
Most of it, though, is clearly implementation specific.

However, cmdebug's interface to the cache manager is by making RXAFSCB
RPC's, and RXAFSCB is a standardized RPC service with a registry.  So,
while the details of CM debugging RPC's can and should be specific to
each implementation, the assignment of RPC numbers cannot be.

My preference here would be to assign distinct RPC numbers for each new
implementation-specific debugging RPC, just as we would for any other
RPC.  These numbers should not be reused from one implementation to the
next, since that could create problems when a debugging tool from one
implementation is inadvertently applied to a cache manager from another.


I do not think it would be inappropriate for the standardization group
to consider the security implications of some of the debugging
interfaces we've seen in the past, and perhaps publish an informational
document containing security considerations for such interfaces.
However, it would be just as appropriate for OpenAFS to develop such a
document on its own, and in either case, I don't think it's high on
anyone's priority list.

-- Jeff