Release Meeting Minutes/2011-03-01
Greg Hudson, Tom Yu, Zhanna Tsitskova, Sam Hartman, Will Fiveash, Simo Sorce
[some crypto backend discussion]
- Spec won't change. (probably) Should be easy to add KDC support (for accepting both wrong and right forms). Client side is not so clear.
- Don't like the retry approach (latency etc.). Look for announcement.
- At some point drop old behavior. It's a CMS thing (not part of our ASN.1 code). Code would actually have been cleaner if I'd dont it correctly. Probably not going to be that gross. On the client, might cause problems with robustness... signing path vs anonymous path are undifferentiated for longer. Will try to get to it after IETF.
- KDC fix is more valuable.
[debate about adding knobs for testing]
- Config option for turning on (hierarchical) walk_rtree might be wrong. KDC has implemented capaths for a long time. Heimdal has no realm walk (on the client). Their KDC does (or maybe it forces explicit capaths). Will probably just disable hierarchical walk_rtree.
Trust KDC-local name resolution
- [referrals draft stuff] Would be easy to put in Love's flag "trust me for local aliasing".
- Ticket flag, maybe per-realm (client config) option for "trust KDC for name resolution"
Sam mentions enc-padata stuff (from referrals draft)
Greg agrees with Sam. Also, changing rdns default might be very painful?
- athena.dialup might break (DNS round robin type situation).
- It's a general issue with distributed services. Shared principals mean you have to share replay caches, etc. Does GS2 have issues with replay resistance?
- It's important if you use CFX. Channel bindings. If using endpoint channel bindings, you really want to use DCE-style (3-legged). Otherwose, unique channel bindings might be safe.
- Heimdal does no rdns; they do AI_CANONNAME.
- So they have the forward DNS problem too...
- gethostbyname or getaddrinfo on Solaris or Ultrix wouldn't give you the canonical name if you gave it an alias.
[Tom to propose "trust KDC for aliases" flag type thing]