Did you guys review the functional spec that has been floating around on cwiki? On May 23, 2015 8:27 AM, "Wido den Hollander" <w...@widodh.nl> wrote:
> -----BEGIN PGP SIGNED MESSAGE----- > Hash: SHA1 > > > > On 05/22/2015 11:05 PM, server24 Cloudstack wrote: > > Hi Wido, > > > > was nice talking to you about this. > > > > On 5/21/2015 8:59 PM, Wido den Hollander wrote: > > > >> (IPv6) routers should send out RAs (Router Advertisements) with > >> the managed-other-flag [0][1], telling Instances to ONLY use that > >> routers as their default gateways and NOT to use SLAAC to > >> autoconfigure their IP-Address. > > > > OK, so no autonomous flag > > > > No, the "managed other flag" as described in RFC 4862. Meaning that > the Routers should only be used as a default gateway and DHCPv6 should > be used for obtaining a address. > > >> The (ip6tables) Security Groups should allow ICMPv6 by default. > >> IPv6 traffic breaks really hard without ICMPv6 traffic, for > >> example PMTU doesn't work properly and breaks IPv6 connections. > > yes, and default ip(6)tables should be in place to block > > VNC-related traffic except to the Virtual Console (as currently VNC > > ports on IPv6 are world-wide-open in BASIC network)! > > > > Yes, but in that case you are talking about the Console Proxy which > should be firewalled properly. > > >> In CloudStack we might configure a /48, but tell it to hand out > >> addresses for each instance from a /64 out of that /48. That > >> means we can have 65k Instances in that pod. Some firewall > >> policies block a complete /64 when they see malicious traffic > >> coming from that subnet, so if the subnet is big enough we should > >> try to keep all the IPv6 addresses from one Instance in the same > >> /64 subnet. This could also simplify the iptable rules. > > so one /48 per pod? RIRs provide either /48 or /32 (the latter to > > the providers) IPv6 blocks. So this should then be configurable, > > both per instance and per pod. One /48 per pod still looks large to > > me.. > > > > A /48 should be a possibility. If you only have a /64 available that > should be no problem either. > > > On the other hand any prefix more specific than /64 could break > > IPv6 features, so that there are at least some practical values to > > rely on. > >> Security grouping has to be extended to also support IPv6, but > >> should allow ICMPv6 by default. > > yes, ICMPv6 should be on by default (maybe it should be forced to > > be always on for IPv6?). > > > >> At the end of June 2015 we want to keep a one-day meetup in > >> Amsterdam with various developers to discuss some more details. > > > > great work and very good meeting, was a pleasure to be there. > > > > Thomas Moroder > > > > -- Incubatec GmbH - Srl Via Scurcia'str. 36, 39046 Ortisei(BZ), > > ITALY Registered with the chamber of commerce of Bolzano the 8th of > > November 2001 with REA-No. 168204 (s.c. of EUR 10.000 f.p.u.) > > President: Thomas Moroder, VAT-No. IT 02283140214 Tel: > > +39.0471796829 - Fax: +39.0471797949 > > > > IMPRINT: http://www.incubatec.com/imprint.html PRIVACY: > > http://www.server24.it/informativa_completa.html > > > -----BEGIN PGP SIGNATURE----- > Version: GnuPG v1 > > iQIcBAEBAgAGBQJVYJwhAAoJEAGbWC3bPspCWXEQAISPV1PdGWa6KOck9IsTVXBt > jUTpyFyg+qnmlG+QQ3LWOFjXRVUvQroryBbxkBnEbNm5d5qOsKptwwOaXMOut4A2 > Nv4WCcFlAjnj78c9C/mpJvqu+Bh/WLKy4mBaMEqLiSAzqoz+CPlaiubJ/TDXR+jp > XSY3XNk/jhdI02QTPNHvYc1ZbWNjuZrb+YVqEzFLra25I0bfXuq2tcBVDEMr1zmA > qQVfCabkAx/a8wW2wGnz2GSg1UFDJUHOb7c6bae9nE5wo1MYpXyjpO53IBpRQuPt > +VMzkjyf75yS1LFel9zS/BzV97mgEBxux9Nb3M9f/ZJW0QvS9onZdIhYgCeDyxJe > M/XTD6M0O+ha4mFaYTeSudWoQnv/ZZ1P5RuPTQyQRD4P7nSkorz6QOR/SVb+OhXG > 7tqd0GaUS/OFTC69wBTN/t9m98mYRZ5s4XtuocE5DadHqIv5JslrKLzC2YJEWonm > RqL2Gow0/h+k99esJatmCZq+6jXwsy9pIVsLspfDt+GKBOVw8sHNTDUPTvdVzffv > Qa+gVl5gg9RvSonJ8xS7sHI/p/gDFJrN1lWzxl9YWyurjDKFz2zI0OWfOKGZdhrE > Ywgzb+2ExzGSgLSE6AL8awLbl1N57TOlQI4SlfN7Ph4kaS2T9eCleAXP3BxPSXqK > Hji1OI5/luKcQVyYqwaT > =acrP > -----END PGP SIGNATURE----- >