Hi Wes,

> I note that this is still listing the requested IANA extended community as
> non-transitive, but has removed the explicit prohibition to sending this
> community to an eBGP peer based on my prior feedback. Is that going to work,
> or does the community need to be redefined as transitive as well?
> 
> Also, if we're going to allow it to be used in eBGP in even a limited
> fashion, I think that we do need some text that covers this transitive vs
> intransitive distinction in a bit more detail. For example, when I
> recommended that we not explicitly prohibit sending this community on an
> eBGP session, I also said that it should default to non-transitive for eBGP
> in order to prevent misconfigurations from causing validation info of
> questionable origin from moving between ASNs. In other words, it should be a
> configurable option, but we want to express that the preference is to have
> people do their own RPKI validation at their ASBR rather than relying on an
> upstream and to limit propagation of this information since it's of
> questionable accuracy and trust. We may also want to say that the receiving
> peer must be explicitly configured to accept the extended community, and by
> default it should drop any instance of the validation community on eBGP. Not
> sure if we need to explicitly state that peers that don't understand/support
> the community should silently ignore it, since I'm pretty sure that's
> default behavior.

Yes, making the extended community non-transitive, but allowing it to be
exchanged across ebgp sessions helps this.

> 
> Updating the deleted text from the original document might give us something
> like this:
> Since the extended community is primarily intended to provide a method to
> carry prefix validation information within an ASN, it SHOULD be a
> non-transitive attribute, but there are some applications where it may be
> necessary for it to transit between ASNs. In order to limit the propagation
> of this information outside of the originating ASN, it MUST NOT be attached
> by default while advertising routes to EBGP peers, even if the session is
> configured to send other communities. It MAY be explicitly configured to be
> sent between eBGP peers. Additionally, if it is present in an UPDATE message
> received from an EBGP peer, by default the router MUST remove the origin
> validation state extended community before parsing rest of the message. The
> router MAY be explicitly configured to allow the origin validation state
> from an eBGP peer to be accepted, but this is not recommended unless there
> is an existing trust relationship between the ASNs (such as upstream ISP and
> customer). Any ASN that learns validation information from an eBGP peer via
> this community MUST NOT pass that community beyond routers in its own ASN.

Thanks! I am fine with adding this text. Will wait for a few if others want to
chime in...

- Pradosh
_______________________________________________
sidr mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/sidr

Reply via email to