| Network Working Group | M. Stenberg |
| Internet-Draft | June 26, 2013 |
| Intended status: Standards Track | |
| Expires: December 28, 2013 |
Hybrid Unicast/Multicast DNS-Based Service Discovery Auto-Configuration Using OSPFv3
draft-stenberg-homenet-dnssdext-hybrid-proxy-ospf-00
This document describes how a proxy functioning between Unicast DNS-Based Service Discovery and Multicast DNS can be automatically configured using automatically configured routing protocol or some other network-level state sharing mechanism. Zero-configuration OSPFv3 is used to describe one concrete way to implement this scheme.
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at http://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."
This Internet-Draft will expire on December 28, 2013.
Copyright (c) 2013 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
Section 3 ("Hybrid Proxy Operation") of [I-D.cheshire-mdnsext-hybrid] describes how to translate queries from Unicast DNS-Based Service Discovery described in [RFC6763] to Multicast DNS described in [RFC6762], and how to filter the responses and translate them back to unicast DNS.
This document describes what sort of configuration the participating DNS servers require, as well as how it can be provided using auto-configured OSPFv3 described in [I-D.ietf-ospf-ospfv3-autoconfig] and a naming scheme which does not even need to be same across the whole covered network. The scheme can be used to provision both forward and reverse DNS zones which employ hybrid proxy for heavy lifting.
While this document describes the data to be transferred in auto-configured OSPFv3 TLVs, in principle same scheme could work across other routing protocols, or even some non-routing protocol, as long as the consistent state for it is available across the whole covered network (by for example site-scoped multicast, or some other flooding scheme).
We go through the mandatory specification of the language used in Section 2, then describe what needs to be configured in hybrid proxies and participating DNS servers across the network in Section 3. How the data is exchanged in OSPFv3 is described in Section 4. Finally, some overall notes on desired behavior of different router components is mentioned in Section 5.
In this document, the key words "MAY", "MUST, "MUST NOT", "OPTIONAL", "RECOMMENDED", "SHOULD", and "SHOULD NOT", are to be interpreted as described in [RFC2119].
Beyond the low-level translation mechanism between unicast and multicast service discovery, the hybrid proxy draft [I-D.cheshire-mdnsext-hybrid] describes just that there have to be NS records pointing to hybrid proxy responsible for each link within the covered network.
The links to be covered is also non-trivial choice; we can use the border discovery functionality (if available) to determine internal and external links. Or we can use OSPFv3 presence (or lack of it) on a link to determine internal links within the covered network, and some other signs (depending on the deployment) such as DHCPv6 Prefix Delegation (as described in [RFC3633] to determine external links that should not be covered.
For each covered link we want forward DNS zone delegation to an appropriate router which is connected to a link, and running hybrid proxy. We also want to populate reverse DNS zone similarly per each prefix in use. Links' forward DNS zone names should be unique.
There should be DNS-SD domain search list provided for the network's domain which contains domain for each physical link only once, regardless of how many routers and hybrid proxy implementations are connected to it.
Yet another case to consider is the list of DNS-SD domains that we want hosts to enumerate for domain lists. Typically, it contains only that the network's domain, but there may be also other networks we may want to pretend to be local but are in different scope, or controlled by different organization. For example, a home user might see both home domain's services (TBD-TLD), as well as ISP's services under isp.example.com.
Any naming-related choice on a router may have conflicts in the network.
We use similar conflict resolution scheme as described in the prefix assignment draft[I-D.arkko-homenet-prefix-assignment]. That is, if a conflict is encountered, the router with highest router ID MUST keep the name they have chosen. The one(s) with lower router ID MUST either try different one (that is not in use at all according to the current link state information), or choose not to publish the name altogether.
If router needs to pick a different name, any algorithm works, although simple algorithm choice is just like the one described in Multicast DNS[RFC6762]: append -2, -3, and so forth, until there are no conflicts in the network for the given name.
How to name the links of a whole network in automated fashion? Two different approaches seem obvious:
The first choice is appealing as it can be much more friendly (especially given manual configuration). For example, it could mean just lan.example.com and wlan.example.com for a simple home network. The second choice, on the other hand, has a nice property of being local choice as long as router name can be made unique.
The type of naming scheme to use can be left as implementation option. And the actual names themselves SHOULD be also overridable, if the end-user wants to customize them in some way.
Note that any manual configuration, which SHOULD be possible, MUST override the defaults provided here or chosen by the creator of the implementation.
It is not obvious how to produce network-wide unique link names for the (unique-link).(domain) scheme. One option would be to base it on type of physical network layer, and then hope that the number of the networks won't be significant enough to confuse (e.g. "lan", or "wlan").
In general network-wide unique link names should be only used in small networks. Given larger network, after conflict resolution, finding which network is 'lan-42.example.com' may be challenging.
Recommendation is to use some short form which indicates the type of router it is, for example, "openwrt.example.com". As the name is visible to users, it should be kept as short as possible. If theory even more exact model could be helpful, for example, "openwrt-buffalo-wzr-600-dhr.example.com". In practise, though, providing some other records indicating exact router information (and access to management UI) might be more sensible.
If scheme 2 is used, and there is no desire to implement conflict resolution related TLV described in Section 4.3, a safe default might be to default to router ID; that is, use as router name value such as r-(router ID as single 32-bit number). It is guaranteed to be unique across the network, but not as user-friendly as the descriptive router name promoted here.
Recommendation for (link) portion of (link).(router).(domain) is to use either physical network layer type as base, possibly even just interface name on the router, if it's descriptive enough, for example, eth0.router1.example.com and wlan0.router1.example.com may be good enough.
To implement this specification fully, support for following three different new OSPFv3 auto-configuration TLVs is needed. However, only the DNS Delegated Zone TLVs MUST be supported, and the other two SHOULD be supported.
This TLV is effectively a combined NS and A/AAAA record for a zone. It MUST be supported by implementations conforming to this specification. Implementations SHOULD provide forward zone per link (or optimizing a bit, zone per link with Multicast DNS traffic). Implementations MAY provide reverse zone per prefix using this same mechanism. If multiple routers advertise same reverse zone, it should be assumed that they all have access to the link with that prefix. However, as noted in Section 5.3, mainly only the router with highest router ID on the link should publish this TLV.
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| TBD-BY-IANA-1 | Length |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| |
| Address |
| (16 bytes) |
| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|Reserved |S|B| Zone (DNS label sequence - variable length) |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
DNS Delegated Zone TLV
This TLV is used to indicate the base (domain) to be used for the network. If multiple routers advertise different ones, the conflict resolution rules in Section 3.1 should result in only the one with highest router ID advertising one, eventually. In case of such conflict, user SHOULD be notified somehow about this, if possible, using the configuration interface or some other notification mechanism for the routers.
This TLV SHOULD be supported if at all possible. It may be derived using some future DHCPv6 option, or be set by manual configuration. Even on routers without manual configuration options, being able to read the domain name provided by a different router could make the user experience better due to consistent naming of zones across the network.
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| TBD-BY-IANA-2 | Length |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|Domain (DNS label sequence - variable length) |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Domain Name TLV
Like the Zone field in Section 4.1, the Domain Name TLV's contents are encoded as label sequence.
By default, if no router advertises domain name TLV, hard-coded default (TBD) should be used.
This TLV is used to advertise a router's name. After the conflict resolution procedure described in Section 3.1 finishes, there should be exactly zero to one routers publishing each router name.
This TLV SHOULD be supported if at all possible. If not supported, and another router chooses to use the (link).(router) naming scheme with this router's name, the contents of the network's domain may look misleading (but due to conflict resolution of per-link zones, still functional).
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| TBD-BY-IANA-3 | Length |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|Name (not even null terminated - variable length) |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Router Name TLV
If the router name has been configured manually, and there is a conflict, user SHOULD be notified somehow about this, if possible, using the configuration interface or some other notification mechanism for the routers.
This TLV is used to announce address of a fallback recursive DNS server (provided by e.g. ISP). If the DNS server implementations used in the network are not full recursive resolver implementations, they may respond to network-specific queries within network, and forward the rest to the provided DNS servers. Even recursive resolver implementations may want to use these servers, if available, for better caching and therefore more responsive user experience.
Typically, these addresses are gleaned from (for example) a DHCPv4/DHCPv6 exchange, or from Router Advertisements.
Any router on the home network can publish 0-N of these TLVs, and the order in which they are used is not defined (we assume that the DNS view of the world is consistent; this may not be true in all cases).
This TLV SHOULD be supported by routers, but the routers (and DNS servers in the network) MUST be able to cope even in the absence of the TLV. This can be handled by (for example) DNS servers providing recursive resolving fallback functionality, or defaulting to some known global recursive resolver.
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| TBD-BY-IANA-4 | Length |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| |
| Address |
| (16 bytes) |
| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
DNS Server TLV
The address may be again either IPv4 or IPv6 address, with the IPv4 address encoded under the ::FFFF:/96 prefix.
It is important to note that if the network's domain forward or reverse resolution will not work globally, using network-external DNS server directly is not good. Therefore the network's local DNS servers should be announced to hosts in e.g. DHCPv4/DHCPv6/RA, and then only those DNS servers can use the contents of this TLV as fall-back for non-local resolution if so desired. How these local DNS server addresses are propagated within home network is outside the scope of this document
The routers following this specification SHOULD provide the used (domain) as one item in the search path to it's hosts, so that DNS-SD browsing will work correctly. They also SHOULD include any DNS Delegated Zone TLVs' zones, that have S bit set.
The hybrid proxy implementation SHOULD support both forward zones, and IPv4 and IPv6 reverse zones. It SHOULD also detect whether or not there are any Multicast DNS entities on a link, and make that information available to OSPFv3 daemon. This can be done by (for example) passively monitoring traffic on all covered links, and doing infrequent service enumerations on links that seem to be up, but without any Multicast DNS traffic (if so desired).
Hybrid proxy SHOULD also publish it's own name via Multicast DNS (both forward A/AAAA records, as well as reverse PTR records) to facilitate applications that trace network topology.
OSPFv3 daemon should avoid publishing TLVs about links that have no Multicast DNS traffic to keep the DNS-SD browse domain list as concise as possible. It also SHOULD NOT publish delegated zones for links that it does not have highest router ID that supports this specification. (Support for this specification can be deduced by e.g. presence of any TLVs from this draft advertised by a router.)
OSPFv3 daemon (or other entity with access to the TLVs) SHOULD generate zone information for DNS implementation that will be used to serve the (domain) zone to hosts. Domain Name TLV described in Section 4.2 should be used as base for the zone, and then all DNS Delegated Zones described in Section 4.1 should be used to produce the rest of the entries in zone (see Appendix A.4 for example interpretation of the TLVs in Appendix A.3.
There is a trade-off between security and zero-configuration in general; if used routing protocol is not authenticated (and in zero-configuration case, it most likely is not), it is vulnerable to local spoofing attacks. We assume that this scheme is used either within (lower layer) secured networks, or with not-quite-zero-configuration routing protocol set-up which has authentication.
If some sort of dynamic inclusion of links to be covered using border discovery or such is used, then effectively service discovery will share fate with border discovery (and also security issues if any).
This document makes two allocations out of the OSPFv3 Auto- Configuration (AC) LSA TLV namespace [I-D.ietf-ospf-ospfv3-autoconfig]:
| [I-D.cheshire-mdnsext-hybrid] | Cheshire, S., "Hybrid Unicast/Multicast DNS-Based Service Discovery", Internet-Draft draft-cheshire-mdnsext-hybrid-01, January 2013. |
| [I-D.ietf-ospf-ospfv3-autoconfig] | Lindem, A. and J. Arkko, "OSPFv3 Auto-Configuration", Internet-Draft draft-ietf-ospf-ospfv3-autoconfig-00, October 2012. |
| [RFC1035] | Mockapetris, P., "Domain names - implementation and specification", STD 13, RFC 1035, November 1987. |
| [RFC2119] | Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997. |
| [RFC6762] | Cheshire, S. and M. Krochmal, "Multicast DNS", RFC 6762, February 2013. |
| [RFC6763] | Cheshire, S. and M. Krochmal, "DNS-Based Service Discovery", RFC 6763, February 2013. |
| [I-D.arkko-homenet-prefix-assignment] | Arkko, J. and A. Lindem, "Prefix Assignment in a Home Network", Internet-Draft draft-arkko-homenet-prefix-assignment-01, October 2011. |
| [RFC3633] | Troan, O. and R. Droms, "IPv6 Prefix Options for Dynamic Host Configuration Protocol (DHCP) version 6", RFC 3633, December 2003. |
| [RFC3646] | Droms, R., "DNS Configuration options for Dynamic Host Configuration Protocol for IPv6 (DHCPv6)", RFC 3646, December 2003. |
Let's assume home network that looks like this:
|[0]
+-----+
| CER |
+-----+
[1]/ \[2]
/ \
+-----+ +-----+
| IR1 |-| IR2 |
+-----+ +-----+
|[3]| |[4]|
We're not really interested about links [0], [1] and [2], or the links between IRs. Given the optimization described in Section 4.1, they should not produce anything to OSPF state (and therefore to DNS either) as there isn't any Multicast DNS traffic there.
The user-visible set of links are [3] and [4]; each consisting of a LAN and WLAN link. We assume that ISP provides 2001:db8::/48 prefix to be delegated in the home via [0].
Given implementation that chooses to use the second naming scheme (link).(router).(domain), and no configuration whatsoever, here's what happens (the steps are interleaved in practise but illustrated here in order):
Once there is no longer any conflict in the system, we wind up with following TLVs within OSPFv3 (RN is used as abbreviation for Router Name, and DZ for Delegated Zone TLVs):
(from CER)
DZ {s=1,zone="isp.example.com"}
(from IR1)
RN {name="router"}
DZ {address=2001:db8:1:1::1, b=1,
zone="lan.router.example.com."}
DZ {address=2001:db8:1:1::1,
zone="1.0.0.0.1.0.0.0.8.b.d.0.1.0.0.2.ip6.arpa."}
DZ {address=2001:db8:1:1::1, b=1,
zone="wlan.router.example.com."}
DZ {address=2001:db8:1:1::1,
zone="2.0.0.0.1.0.0.0.8.b.d.0.1.0.0.2.ip6.arpa."}
(from IR2)
RN {name="router-2"}
DZ {address=2001:db8:2:1::1, b=1,
zone="lan.router-2.example.com."}
DZ {address=2001:db8:2:1::1,
zone="1.0.0.0.2.0.0.0.8.b.d.0.1.0.0.2.ip6.arpa."}
DZ {address=2001:db8:2:1::1, b=1,
zone="wlan.router-2.example.com."}
DZ {address=2001:db8:2:1::1,
zone="2.0.0.0.2.0.0.0.8.b.d.0.1.0.0.2.ip6.arpa."}
In the end, we should wind up with following zone for (domain) which is example.com in this case, available at all routers, just based on dumping the delegated zone TLVs as NS+AAAA records, and optionally domain list browse entry for DNS-SD:
b._dns_sd._udp PTR lan.router
b._dns_sd._udp PTR wlan.router
b._dns_sd._udp PTR lan.router-2
b._dns_sd._udp PTR wlan.router-2
router AAAA 2001:db8:1:1::1
router-2 AAAA 2001:db8:2:1::1
router NS router
router-2 NS router-2
1.0.0.0.1.0.0.0.8.b.d.0.1.0.0.2.ip6.arpa. NS router.example.com.
2.0.0.0.1.0.0.0.8.b.d.0.1.0.0.2.ip6.arpa. NS router.example.com.
1.0.0.0.2.0.0.0.8.b.d.0.1.0.0.2.ip6.arpa. NS router-2.example.com.
2.0.0.0.2.0.0.0.8.b.d.0.1.0.0.2.ip6.arpa. NS router-2.example.com.
Internally, the router may interpret the TLVs as it chooses to, as long as externally defined behavior follows semantics of what's given in the above.
So, what do the hosts receive from the routers? Using e.g. DHCPv6 DNS options defined in [RFC3646], DNS server address should be one (or multiple) that point at DNS server that has the zone information described in Appendix A.4. Domain list provided to hosts should contain both "example.com" (the hybrid-enabled domain), as well as the externally learned domain "isp.example.com".
When hosts start using DNS-SD, they should check both b._dns-sd._udp.example.com, as well as b._dns-sd._udp.isp.example.com for list of concrete domains to browse, and as a result services from two different domains will seem to be available.
There is an prototype implementation of this draft (and transitively also of [I-D.cheshire-mdnsext-hybrid]) at hnet-core github repository which contains variety of other homenet WG-related things' implementation too.
hp.lua binary can be used to start hybrid proxy either as one-router stand-alone implementation (that can be used to e.g. use statically configured DNS zones), or as part of zeroconf OSPFv3 configured set of proxies.
Sample usage case:
# sudo lua hp.lua eth0 eth1
.. no output ..
Given the command, hybrid proxy is started for interfaces eth0 and eth1, and it will publish DNS zones l-eth0.r-router.home, l-eth1.r-router.home (and home zone with relevant DNS-SD sub-zone, and default forward behavior) in DNS port. It has -h option for seeing various options that can be set, notable one being --ospf (use OSPFv3 autoconfigured hnet infrastructure).
Disclaimer: The set-up of third-party libraries etc to get the implementation running may be painful and is omitted here. Use of ready UML NetKit-based test environment or building image for a real router using hnet github repository is recommended.
Over the time number of people have asked me about how, why, and if we should proxy (originally) link-local Multicast DNS over multiple links.
At some point I meant to write a draft about this, but I think I'm too lazy; so some notes left here for general amusement of people (and to be removed if this ever moves beyond discussion piece).
There are two main reasons why Multicast DNS is not proxyable in the general case.
First reason is the conflict resolution depends on ordering which depends on the RRsets staying constant. That is not possible across multiple links (due to e.g. link-local addresses having to be filtered). Therefore, conflict resolution breaks, or at least requires ugly hacks to work around.
A workaround for this is to make sure that in conflict resolution, propagated resources always loses. Due to conflict handling ordering logic, and the arbitrary order in which the original records may be in, this is non-trivial.
Second reason is timing, which is relatively tight in the conflict resolution phase, especially given lossy and/or high latency networks.
In general, typical stateless proxy has to involve flooding, as Multicast DNS assumes that most messages are received by every host. And it won't scale very well, as a result.
The conflict resolution is also harder without state. It may result in Multicast DNS responder being in constant probe-announce loop, when it receives altered records, notes that it's the one that should own the record. Given stateful proxying, this would be just a transient problem but designing stateless proxy that won't cause this is non-trivial exercise.
One option is to write proxy that learns state from one link, and propagates it in some way to other links in the network.
A big problem with this case lies in the fact that due to conflict resolution concerns above, it is easy to accidentally send packets that will (possibly due to host mobility) wind up at the originator of the service, who will then perform renaming. That can be alleviated, though, given clever hacks with conflict resolution order.
The stateful proxying may be also too slow to occur within the timeframe allocated for announcing, leading to excessive later renamings based on delayed finding of duplicate services with same name
A work-around exists for this though; if the game doesn't work for you, don't play it. One option would be simply not to propagate ANY records for which conflict has seen even once. This would work, but result in rather fragile, lossy service discovery infrastructure.
There are some other small nits too; for example, Passive Observation Of Failure (POOF) will not work given stateful proxying. Therefore, it leads to requiring somewhat shorter TTLs, perhaps.
Thanks to Stuart Cheshire for the original hybrid proxy draft and interesting discussion in Orlando, where I was finally convinced that stateful Multicast DNS proxying is a bad idea.
Also thanks to Mark Baugher, Ole Troan and Shwetha Bhandari for review comments.