Hi authors, I welcome the work in this document to update the BCPs and improve the status quo.
May I suggest a few additions to the document (or ask if you have considered these already)? * Don't send or accept internal or deprecated attributes unless explicitly agreed - An example of a fully internal attribute is D-PATH, this should never be seen in the DFZ (https://www.rfc-editor.org/info/rfc10039/). One can see all IANA deprecated attributes here: https://www.iana.org/assignments/bgp-parameters. I don't think blocking all attributes except for an approved list, or blocking unknown attributes is helpful, that makes addition of new attributes a pain. Instead, just block what should _never_ be in the DFZ. We see that some of our downstream customers receive these attributes and their BGP daemon dies because they are using the same internal attribute and it somehow doesn't make sense in the context it was received from their transit provider. Or they try to process deprecated attributes. Having reached out to some originating networks in the past to try and get them removed, they don't, so I think that unless we block them they will continue to linger. * Don't send or accept extended communities unless explicitly agreed - some networks with older equipment use extended communities because they didn't support large communities around the time standard communities no longer became suitable (i.e. for signalling 32bit AS numbers). However, enabling the exchange of extended communities then leads to all manner of extended communities being leaked on outbound and not being filtered on ingress (route targets for example). It could be a security risk and is not needed now that all major BGP daemons support large communities. * Don't accept paths with an AS_SET in them - they are deprecated but people are still originating such routes. A bit like people leaking extended communities and deprecated or internal attributes, they will continue to linger and the cause problems like not knowing which ASN to contact when there is a problem, problems with ROA validation, and problems with building AS path filters. * Maybe add something about enforcing first hop AS (when not peering with an IXP route server)? If you've had any thoughts about these things already I'm interested to hear them. Thank you, James.
signature.asc
Description: OpenPGP digital signature
_______________________________________________ GROW mailing list -- [email protected] To unsubscribe send an email to [email protected]
