Moin,

Some things have changed since BCP194 came to be, and I figured it
might make sense to take shot at updating it. I presented the idea at
OPSEC at 117, and after reading the draft Job suggested to rather
discuss it in GROW instead of OPSEC, so here it is to pick apart:

https://datatracker.ietf.org/doc/draft-fiebig-grow-bgpopsecupd/

The draft is an import and change of BCP194, currently focusing on a
'maximum recommendation' option as a starting point for discussion;
Specifically, the following things have been changed/added:

- Unify terminology, import from RFC9234 and the ASPA draft;
  Expand definitions.
- Use peer only in the AS relationship context; Add mutual-upstream 
  relation.
- Add note on the (hypothetical) PMTUD attack.
- Clarify that IXPs decide whether their IXP LAN(s) should be 
  advertised; Removed IXP example in the Appendix.
- Noted additional IXP Lan cases (upstream via etc.) and next-hop 
  filtering
- Split between dynamic (IRR and down) and (semi) static filter 
  (bogons etc.) sources
- Add ASPA and BGP roles/OTC
- Note use of community-based filtering, also discussing leak-potential
  of static filters being used.
- Add large-communities/scrubbing
- Describe Rulesets more explicitly and in terms of ideal order for 
  more relationships
- Emphasize outbound filtering
- Suggest iBGP filtering (this may be a bridge too far)
- Add MAY for scrubbing of attributes
- Add note on ASPATH max-length filtering
- Reference RFC9319
- Add max-prefix limits and global limits
- Add considerations for filter implementation and generation/runtime
- Removed the definition of what a Tier-1 is.

Current Todo/Discussion points:
- Unify terminology around route/prefix/NLRI, which is currently a bit
  mixed up.
- Consider adding a point on not using LPREF on RS sessions/honoring  
  GSHUT
- Use of MUST vs. SHOULD; Other contemporary drafts use stronger 
  language; From a general standpoint I'd argue 'MUST' might be better
  under the premise of 'to follow best practices this MUST be done'
  with the caveat of 'best practices SHOULD' be followed; Still,
  currently it still uses BCP14 SHOULD for all points.
- Putting in a point on having preparations in place to filter specific
  BGP options at the border in case one's infrastructure topples when a
  specific option is sent.
- Figure out if this should be update or obsolete.
- Fix nits in the abstract.

Looking forward to hear opinions, especially on the somewhat more
stretchy points (recommending filtering iBGP for example ;-)).

Also, if anyone wants to join in on writing, let me know.

With best regards,
Tobias

-- 
Dr.-Ing. Tobias Fiebig
T +31 616 80 98 99
M [email protected]

_______________________________________________
GROW mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/grow

Reply via email to