On 04/24/2012 03:04 PM, Kaleb Keithley wrote:
From: "Niels de Vos"<ndevos@xxxxxxxxxx>
Sent: Tuesday, April 24, 2012 7:11:02 AM>
At the moment Gluster uses quite some RPC program numbers which are not
officially registered for Gluster usage.
RFC 5531 (http://tools.ietf.org/html/rfc5531) contains "Appendix B:
Requesting RPC-Related Numbers from IANA". I think it will be good to
request program numbers for the different RPC protocols. It should even be
possible to serve two program numbers at the same time, keeping compatibility
between minor releases, but at the same time prepare for the switch.
Thoughts?
I'm not sure we would solve anything by registering them.
This is not about solving a problem, rather about preventing one and keeping
aligned with standards.
Unlike NFS, where we run on non-standard ports, using registered prognums
and progvers lets us advertise via portmapper to clients what port(s) to use
> for mountd, nfs, and nlm services.
Registering also prevents that others will use the same prognums. It is maybe
not very likely that other protocols will collide with the Gluster prognums, but
there is no guarantee.
If we ever intend to run glusterd on a port other than 24007, and advertise
through portmapper, then I could imagine there would be some value in
registering our prognums and progvers.
At the moment, there is a conflict in the authentication that is is included in
the RPC-header for some calls. The AUTH-flavor is set to '5', which is assigned
to AUTH_RSA.
Luckily AUTH_RSA is not implemented in any RPC-protocols I could find. It may be
acceptable for tools like Wireshark to unofficially assign this '5' to the
Gluster protocol and decode the contents.
But we don't advertise in portmapper, and AFAIK, we're not planning to use
anyport other than 24007 (and 24009+)
I presume we've registered port 24007 at least with IANA. If not, we should.
Thanks,
Niels
