| Internet Engineering Task Force | S. Cheshire |
| Internet-Draft | Apple Inc. |
| Intended status: Standards Track | September 13, 2017 |
| Expires: March 17, 2018 |
Discovery Proxy for Multicast DNS-Based Service Discovery
draft-ietf-dnssd-hybrid-07
This document specifies a mechanism that uses Multicast DNS to automatically populate the wide-area unicast Domain Name System namespace with records describing devices and services found on the local link.
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 https://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 March 17, 2018.
Copyright (c) 2017 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 (https://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.
Multicast DNS [RFC6762] and its companion technology DNS-based Service Discovery [RFC6763] were created to provide IP networking with the ease-of-use and autoconfiguration for which AppleTalk was well known [RFC6760] [ZC].
For a small home network consisting of just a single link (or a few physical links bridged together to appear as a single logical link from the point of view of IP) Multicast DNS is sufficient for client devices to look up the ".local" host names of peers on the same home network, and to use Multicast DNS-Based Service Discovery (DNS-SD) to discover services offered on that home network.
For a larger network consisting of multiple links that are interconnected using IP-layer routing instead of link-layer bridging, link-local Multicast DNS alone is insufficient because link-local Multicast DNS packets, by design, are not propagated onto other links.
Using link-local multicast packets for Multicast DNS was a conscious design choice [RFC6762]. Even when limited to a single link, multicast traffic is still generally considered to be more expensive than unicast, because multicast traffic impacts many devices, instead of just a single recipient. In addition, with some technologies like Wi-Fi, multicast traffic is inherently less efficient and less reliable than unicast, because Wi-Fi multicast traffic is sent using the lower data rates, and is not acknowledged. Multiplying the amount of expensive multicast traffic by flooding it across multiple links would make the traffic load even worse.
Partitioning the network into many small links curtails the spread of expensive multicast traffic, but limits the discoverability of services. Using a very large local link with thousands of hosts enables better service discovery, but at the cost of larger amounts of multicast traffic.
Performing DNS-Based Service Discovery using purely Unicast DNS is more efficient and doesn't require excessively large multicast domains, but requires that the relevant data be available in the Unicast DNS namespace. The Unicast DNS namespace in question could fall within a traditionally assigned globally unique domain name, or could use a private local unicast domain name such as ".home.arpa" [HOME].)
In the DNS-SD specification [RFC6763], Section 10 ("Populating the DNS with Information") discusses various possible ways that a service's PTR, SRV, TXT and address records can make their way into the Unicast DNS namespace, including manual zone file configuration [RFC1034] [RFC1035], DNS Update [RFC2136] [RFC3007] and proxies of various kinds.
Making the relevant data available in the Unicast DNS namespace by manual DNS configuration (as has been done for many years at IETF meetings to advertise the IETF Terminal Room printer) is labor intensive, error prone, and requires a reasonable degree of DNS expertise.
Populating the Unicast DNS namespace via DNS Update by the devices offering the services themselves requires configuration of DNS Update keys on those devices, which has proven onerous and impractical for simple devices like printers and network cameras.
Hence, to facilitate efficient and reliable DNS-Based Service Discovery, a compromise is needed that combines the ease-of-use of Multicast DNS with the efficiency and scalability of Unicast DNS.
This document specifies a type of proxy called a "Multicast Discovery Proxy" (or just "Discovery Proxy") that uses Multicast DNS [RFC6762] to discover Multicast DNS records on its local link, and makes corresponding DNS records visible in the Unicast DNS namespace.
In principle, similar mechanisms could be defined using other local service discovery protocols, to discover local information and then make corresponding DNS records visible in the Unicast DNS namespace. Such mechanisms for other local service discovery protocols could be addressed in future documents.
The design of the Discovery Proxy is guided by the previously published Requirements for Scalable DNS-Based Service.
In simple terms, a descriptive DNS name is chosen for each link in an organization. Using a DNS NS record, responsibility for that DNS name is delegated to a Discovery Proxy physically attached to that link. Now, when a remote client issues a unicast query for a name falling within the delegated subdomain, the normal DNS delegation mechanism results in the unicast query arriving at the Discovery Proxy, since it has been declared authoritative for those names. Now, instead of consulting a textual zone file on disk to discover the answer to the query, as a traditional DNS server would, a Discovery Proxy consults its local link, using Multicast DNS, to find the answer to the question.
For fault tolerance reasons there may be more than one Discovery Proxy serving a given link.
Note that the Discovery Proxy uses a "pull" model. The local link is not queried using Multicast DNS until some remote client has requested that data. In the idle state, in the absence of client requests, the Discovery Proxy sends no packets and imposes no burden on the network. It operates purely "on demand".
An alternative proposal that has been suggested is a proxy that performs DNS updates to a remote DNS server on behalf of the Multicast DNS devices on the local network. The difficulty of this is that the proxy would have to be issuing all possible Multicast DNS queries all the time, to discover all the answers it needed to push up to the remote DNS server using DNS Update. It would thus generate very high load on the network continuously, even when there were no clients with any interest in that data.
Hence, having a model where the query comes to the Discovery Proxy is much more efficient than a model where the Discovery Proxy pushes the answers out to some other remote DNS server.
A client seeking to discover services and other information achieves this by sending traditional DNS queries to the Discovery Proxy, or by sending DNS Push Notification subscription requests.
A Discovery Proxy does not operate as a multicast relay, or multicast forwarder. There is no danger of multicast forwarding loops that result in traffic storms, because no multicast packets are forwarded. A Discovery Proxy operates as a *proxy* for a remote client, performing queries on its behalf and reporting the results back.
A reasonable analogy would be making a telephone call to a colleague at your workplace and saying, "I'm out of the office right now. Would you mind bringing up a printer browser window and telling me the names of the printers you see?" That entails no risk of a forwarding loop causing a traffic storm, because no multicast packets are sent over the telephone call.
A similar analogy, instead of enlisting another human being to initiate the service discovery operation on your behalf, would be to log into your own desktop work computer using screen sharing, and then run the printer browser yourself to see the list of printers. Or log in using ssh and type "dns-sd -B _ipp._tcp" and observe the list of discovered printer names. In neither case is there any risk of a forwarding loop causing a traffic storm, because no multicast packets are being sent over the screen sharing or ssh connection.
The Discovery Proxy provides another way of performing remote queries, just using a different protocol instead of screen sharing or ssh.
When the Discovery Proxy software performs Multicast DNS operations, the exact same Multicast DNS caching mechanisms are applied as when any other client software on that Discovery Proxy device performs Multicast DNS operations, whether that be running a printer browser client locally, or a remote user running the printer browser client via a screen sharing connection, or a remote user logged in via ssh running a command-line tool like "dns-sd".
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in "Key words for use in RFCs to Indicate Requirement Levels" [RFC2119].
The Discovery Proxy builds on Multicast DNS, which works between hosts on the same link. A set of hosts is considered to be "on the same link" if: Token Ring Source Routing, but not the link-layer *payload*. In particular, if any device forwarding a packet modifies any part of the IP header or IP payload then the packet is no longer considered to be on the same link. This means that the packet may pass through devices such as repeaters, bridges, hubs or switches and still be considered to be on the same link for the purpose of this document, but not through a device such as an IP router that decrements the IP TTL or otherwise modifies the IP header.
The link-layer *header* may be modified, such as in
No changes to existing devices are required to work with a Discovery Proxy.
Existing devices that advertise services using Multicast DNS work with Discovery Proxy.
Existing clients that support DNS-Based Service Discovery over Unicast DNS work with Discovery Proxy. Service Discovery over Unicast DNS was introduced in Mac OS X 10.4 in April 2005, as is included in Apple products introduced since then, including iPhone and iPad, as well as products from other vendors, such as Microsoft Windows 10.
In a typical configuration, a Discovery Proxy is configured to be authoritative [RFC1034] [RFC1035] for four DNS subdomains, and authority for these subdomains is delegated to it via NS records:
In an enterprise network the naming and delegation of these subdomains is typically performed by conscious action of the network administrator. In a home network naming and delegation would typically be performed using some automatic configuration mechanism such as HNCP [RFC7788].
These three varieties of delegated subdomains (service discovery, host names, and reverse mapping) are described below in sections Section 5.1, Section 5.3 and Section 5.4.
How a client discovers where to issue its service discovery queries is described below in section Section 5.2.
+---------------+--------------+-------------+-------------------+ | *example.com* | Building 1 | 1st Floor | Alice's printer | | | Building 2 | *2nd Floor* | Bob's printer | | | *Building 3* | 3rd Floor | Charlie's printer | | | Building 4 | 4th Floor | | | | Building 5 | | | | | Building 6 | | | +---------------+--------------+-------------+-------------------+
Figure 1: Illustrative GUI
In its simplest form, each link in an organization is assigned a unique Unicast DNS domain name, such as "Building 1.example.com" or "2nd Floor.Building 3.example.com". Grouping multiple links under a single Unicast DNS domain name is to be specified in a future companion document, but for the purposes of this document, assume that each link has its own unique Unicast DNS domain name. In a graphical user interface these names are not displayed as strings with dots as shown above, but something more akin to a typical file browser graphical user interface (which is harder to illustrate in a text-only document) showing folders, subfolders and files in a file system.
Each named link in an organization has one or more Discovery Proxies which serve it. This Discovery Proxy function for each link could be performed by a device like a router or switch that is physically attached to that link. In the parent domain, NS records are used to delegate ownership of each defined link name (e.g., "Building 1.example.com") to the one or more Discovery Proxies that serve the named link. In other words, the Discovery Proxies are the authoritative name servers for that subdomain.
With appropriate VLAN configuration a single Discovery Proxy device could have a logical presence on many links, and serve as the Discovery Proxy for all those links. In such a configuration the Discovery Proxy device would have a single physical Ethernet port, configured as a VLAN trunk port, which would appear to software on that device as multiple virtual Ethernet interfaces, one connected to each of the VLAN links.
When a DNS-SD client issues a Unicast DNS query to discover services in a particular Unicast DNS subdomain (e.g., "_printer._tcp.Building 1.example.com. PTR ?") the normal DNS delegation mechanism results in that query being forwarded until it reaches the delegated authoritative name server for that subdomain, namely the Discovery Proxy on the link in question. Like a conventional Unicast DNS server, a Discovery Proxy implements the usual Unicast DNS protocol [RFC1034] [RFC1035] over UDP and TCP. However, unlike a conventional Unicast DNS server that generates answers from the data in its manually-configured zone file, a Discovery Proxy generates answers using Multicast DNS. A Discovery Proxy does this by consulting its Multicast DNS cache and/or issuing Multicast DNS queries for the corresponding Multicast DNS name, type and class, (e.g., in this case, "_printer._tcp.local. PTR ?"). Then, from the received Multicast DNS data, the Discovery Proxy synthesizes the appropriate Unicast DNS response. How long the Discovery Proxy should wait to accumulate Multicast DNS responses is described below in section Section 5.6.
Naturally, the existing Multicast DNS caching mechanism is used to minimize unnecessary Multicast DNS queries on the wire. The Discovery Proxy is acting as a client of the underlying Multicast DNS subsystem, and benefits from the same caching and efficiency measures as any other client using that subsystem.
A DNS-SD client performs Domain Enumeration [RFC6763] via certain PTR queries, using both unicast and multicast. If it receives a Domain Name configuration via DHCP option 15, then it issues unicast queries using this domain. It issues unicast queries using names derived from its IPv6 prefix(es) and IPv4 subnet address(es). These are described below in Section 5.2.1. It also issues multicast Domain Enumeration queries in the "local" domain [RFC6762]. These are described below in Section 5.2.2. The results of all the Domain Enumeration queries are combined for Service Discovery purposes.
The administrator creates Domain Enumeration PTR records [RFC6763] to inform clients of available service discovery domains, e.g.,:
b._dns-sd._udp.example.com. PTR Building 1.example.com.
PTR Building 2.example.com.
PTR Building 3.example.com.
PTR Building 4.example.com.
db._dns-sd._udp.example.com. PTR Building 1.example.com.
lb._dns-sd._udp.example.com. PTR Building 1.example.com.
The "b" ("browse") records tell the client device the list of browsing domains to display for the user to select from and the "db" ("default browse") record tells the client device which domain in that list should be selected by default. The "lb" ("legacy browse") record tells the client device which domain to automatically browse on behalf of applications that don't implement UI for multi-domain browsing (which is most of them, as of 2017). The "lb" domain is often the same as the "db" domain, or sometimes the "db" domain plus one or more others that should be included in the list of automatic browsing domains for legacy clients.
DNS responses are limited to a maximum size of 65535 bytes. This limits the maximum number of domains that can be returned for a Domain Enumeration query, as follows:
A DNS response header is 12 bytes. That's typically followed by a single qname (up to 256 bytes) plus qtype (2 bytes) and qclass (2 bytes), leaving 65275 for the Answer Section.
An Answer Section Resource Record consists of:
This means that each Resource Record in the Answer Section can take up to 268 bytes total, which means that the Answer Section can contain, in the worst case, no more than 243 domains.
In a more typical scenario, where the domain names are not all maximum-sized names, and there is some similarity between names so that reasonable name compression is possible, each Answer Section Resource Record may average 140 bytes, which means that the Answer Section can contain up to 466 domains.
It is anticipated that this should be sufficient for even a large corporate network or university campus.
Since a Discovery Proxy exists on many, if not all, the links in an enterprise, it offers an additional way to provide Domain Enumeration data for clients.
A Discovery Proxy can be configured to generate Multicast DNS responses for the following Multicast DNS Domain Enumeration queries issued by clients:
b._dns-sd._udp.local. PTR ?
db._dns-sd._udp.local. PTR ?
lb._dns-sd._udp.local. PTR ?
This provides the ability for Discovery Proxies to indicate recommended browsing domains to DNS-SD clients on a per-link granularity. In some enterprises it may be preferable to provide this per-link configuration data in the form of Discovery Proxy configuration, rather than populating the Unicast DNS servers with the same data (in the "ip6.arpa" or "in-addr.arpa" domains).
Regardless of how the network operator chooses to provide this configuration data, clients will perform Domain Enumeration via both unicast and multicast queries, and then combine the results of these queries.
DNS-SD service instance names and domains are allowed to contain arbitrary Net-Unicode text, encoded as precomposed UTF-8.
Users typically interact with service discovery software by viewing a list of discovered service instance names on a display, and selecting one of them by pointing, touching, or clicking. Similarly, in software that provides a multi-domain DNS-SD user interface, users view a list of offered domains on the display and select one of them by pointing, touching, or clicking. To use a service, users don't have to remember domain or instance names, or type them; users just have to be able to recognize what they see on the display and touch or click on the thing they want.
In contrast, host names are often remembered and typed. Also, host names have historically been used in command-line interfaces where spaces can be inconvenient. For this reason, host names have traditionally been restricted to letters, digits and hyphens (LDH), with no spaces or other punctuation.
While we still want to allow rich text for DNS-SD service instance names and domains, it is advisable, for maximum compatibility with existing usage, to restrict host names to the traditional letter-digit-hyphen rules. This means that while a service name "My Printer._ipp._tcp.Building 1.example.com" is acceptable and desirable (it is displayed in a graphical user interface as an instance called "My Printer" in the domain "Building 1" at "example.com"), a host name "My-Printer.Building 1.example.com" is less desirable (because of the space in "Building 1").
To accomodate this difference in allowable characters, a Discovery Proxy SHOULD support having two separate subdomains delegated to it for each link it serves, one whose name is allowed to contain arbitrary Net-Unicode text [RFC5198], and a second more constrained subdomain whose name is restricted to contain only letters, digits, and hyphens, to be used for host name records (names of 'A' and 'AAAA' address records).
For example, a Discovery Proxy could have the two subdomains "Building 1.example.com" and "bldg1.example.com" delegated to it. The Discovery Proxy would then translate these two Multicast DNS records:
My Printer._ipp._tcp.local. SRV 0 0 631 prnt.local. prnt.local. A 203.0.113.2
into Unicast DNS records as follows:
My Printer._ipp._tcp.Building 1.example.com.
SRV 0 0 631 prnt.bldg1.example.com.
prnt.bldg1.example.com. A 203.0.113.2
Note that the SRV record name is translated using the rich-text domain name ("Building 1.example.com") and the address record name is translated using the LDH domain ("bldg1.example.com").
A Discovery Proxy MAY support only a single rich text Net-Unicode domain, and use that domain for all records, including 'A' and 'AAAA' address records, but implementers choosing this option should be aware that this choice may produce host names that are awkward to use in command-line environments. Whether this is an issue depends on whether users in the target environment are expected to be using command-line interfaces.
A Discovery Proxy MUST NOT be restricted to support only a letter-digit-hyphen subdomain, because that results in an unnecessarily poor user experience.
A Discovery Proxy can facilitate easier management of reverse mapping domains, particularly for IPv6 addresses where manual management may be more onerous than it is for IPv4 addresses.
To achieve this, in the parent domain, NS records are used to delegate ownership of the appropriate reverse mapping domain to the Discovery Proxy. In other words, the Discovery Proxy becomes the authoritative name server for the reverse mapping domain. For fault tolerance reasons there may be more than one Discovery Proxy serving a given link.
For example, if a given link is using the
IPv6 prefix 2001:0DB8:1234:5678/64,
then the domain "8.7.6.5.4.3.2.1.8.b.d.0.1.0.0.2.ip6.arpa"
is delegated to the Discovery Proxy for that link.
If a given link is using the IPv4 subnet 203.0.113/24,
then the domain "113.0.203.in-addr.arpa"
is delegated to the Discovery Proxy for that link.
When a reverse mapping query arrives at the Discovery Proxy, it issues the identical query on its local link as a Multicast DNS query. The mechanism to force an apparently unicast name to be resolved using link-local Multicast DNS varies depending on the API set being used. For example, in the "/usr/include/dns_sd.h" APIs
(available on macOS, iOS, Bonjour for Windows, Linux and Android), using kDNSServiceFlagsForceMulticast indicates that the DNSServiceQueryRecord() call should perform the query using Multicast DNS. Other APIs sets have different ways of forcing multicast queries. When the host owning that IPv6 or IPv4 address responds with a name of the form "something.local", the Discovery Proxy rewrites that to use its configured LDH host name domain instead of "local", and returns the response to the caller.
For example, a Discovery Proxy with the two subdomains "113.0.203.in‑addr.arpa" and "bldg1.example.com" delegated to it would translate this Multicast DNS record:
2.113.0.203.in-addr.arpa. PTR prnt.local.
into this Unicast DNS response:
2.113.0.203.in-addr.arpa. PTR prnt.bldg1.example.com.
Subsequent queries for the prnt.bldg1.example.com address record, falling as it does within the bldg1.example.com domain, which is delegated to the Discovery Proxy, will arrive at the Discovery Proxy, where they are answered by issuing Multicast DNS queries and using the received Multicast DNS answers to synthesize Unicast DNS responses, as described above.
Generating the appropriate Multicast DNS queries involves,
at the very least, translating from the configured DNS domain (e.g., "Building 1.example.com") on the Unicast DNS side to "local" on the Multicast DNS side.
Generating the appropriate Unicast DNS responses involves translating back from "local" to the appropriate configured DNS Unicast domain.
Other beneficial translation and filtering operations are described below.
For efficiency, Multicast DNS typically uses moderately high DNS TTL values. For example, the typical TTL on DNS-SD PTR records is 75 minutes. What makes these moderately high TTLs acceptable is the cache coherency mechanisms built in to the Multicast DNS protocol which protect against stale data persisting for too long. When a service shuts down gracefully, it sends goodbye packets to remove its PTR records immediately from neighbouring caches. If a service shuts down abruptly without sending goodbye packets, the Passive Observation Of Failures (POOF) mechanism described in Section 10.5 of the Multicast DNS specification [RFC6762] comes into play to purge the cache of stale data.
A traditional Unicast DNS client on a remote link does not get to participate in these Multicast DNS cache coherency mechanisms on the local link. For traditional Unicast DNS queries (those received without using Long-Lived Query [LLQ] or DNS Push Notification [PUSH]) the DNS TTLs reported in the resulting Unicast DNS response SHOULD be capped to be no more than ten seconds.
Similarly, for negative responses, the negative caching TTL indicated in the SOA record [RFC2308] should also be ten seconds (Section 6.1).
This value of ten seconds is chosen based on user-experience considerations.
For negative caching, suppose a user is attempting to access a remote device (e.g., a printer), and they are unsuccessful because that device is powered off. Suppose they then place a telephone call and ask for the device to be powered on. We want the device to become available to the user within a reasonable time period. It is reasonable to expect it to take on the order of ten seconds for a simple device with a simple embedded operating system to power on. Once the device is powered on and has announced its presence on the network via Multicast DNS, we would like it to take no more than a further ten seconds for stale negative cache entries to expire from Unicast DNS caches, making the device available to the user desiring to access it.
Similar reasoning applies to capping positive TTLs at ten seconds. In the event of a device moving location, getting a new DHCP address, or other renumbering events, we would like the updated information to be available to remote clients in a relatively timely fashion.
However, network administrators should be aware that many recursive (caching) DNS servers by default are configured to impose a minimum TTL of 30 seconds. If stale data appears to be persisting in the network to the extent that it adversely impacts user experience, network administrators are advised to check the configuration of their recursive DNS servers.
For received Unicast DNS queries that use LLQ or DNS Push Notification, the Multicast DNS record's TTL SHOULD be returned unmodified, because the Push Notification channel exists to inform the remote client as records come and go. For further details about Long-Lived Queries, and its newer replacement, DNS Push Notifications, see Section 5.6.
A Discovery Proxy SHOULD suppress Unicast DNS answers for records that are not useful outside the local link. For example, DNS AAAA and A records for IPv6 link-local addresses [RFC4862] and IPv4 link-local addresses [RFC3927] SHOULD be suppressed. Similarly, for sites that have multiple private address realms [RFC1918], in cases where the Discovery Proxy can determine that the querying client is in a different address realm, private addresses MUST NOT be communicated to that client. IPv6 Unique Local Addresses SHOULD be suppressed in cases where the Discovery Proxy can determine that the querying client is in a different IPv6 address realm.
By the same logic, DNS SRV records that reference target host names that have no addresses usable by the requester should be suppressed, and likewise, DNS PTR records that point to unusable SRV records should be similarly be suppressed.
Since a Discovery Proxy only knows what names exist on the local link by issuing queries for them, and since it would be impractical to issue queries for every possible name just to find out which names exist and which do not, a Discovery Proxy cannot programatically generate the traditional NSEC and NSEC3 records which assert the nonexistence of a large range of names.
When queried for an NSEC or NSEC3 record type, the Discovery Proxy issues a qtype "ANY" query using Multicast DNS on the local link, and then generates an NSEC or NSEC3 response signifying which record types do and do not exist just the specific name queried, and no others.
Multicast DNS NSEC records received on the local link MUST NOT be forwarded unmodified to a unicast querier, because there are slight differences in the NSEC record data. In particular, Multicast DNS NSEC records do not have the NSEC bit set in the Type Bit Map, whereas conventional Unicast DNS NSEC records do have the NSEC bit set.
A Discovery Proxy does no translation between text encodings. Specifically, a Discovery Proxy does no translation between Punycode and UTF-8, either in the owner name of DNS records, or anywhere in the RDATA of DNS records (such as the RDATA of PTR records, SRV records, NS records, or other record types like TXT, where it is ambiguous whether the RDATA may contain DNS names). All bytes are treated as-is, with no attempt at text encoding translation. A client implementing DNS-based Service Discovery [RFC6763] will use UTF-8 encoding for its service discovery queries, which the Discovery Proxy passes through without any text encoding translation to the Multicast DNS subsystem. Responses from the Multicast DNS subsystem are similarly returned, without any text encoding translation, back to the requesting client.
There may be cases where Application-Specific Data Translation is appropriate.
For example, AirPrint printers tend to advertise fairly verbose information about their capabilities in their DNS-SD TXT record. TXT record sizes in the range 500-1000 bytes are not uncommon. This information is a legacy from LPR printing, because LPR does not have in-band capability negotiation, so all of this information is conveyed using the DNS-SD TXT record instead. IPP printing does have in-band capability negotiation, but for convenience printers tend to include the same capability information in their IPP DNS-SD TXT records as well. For local mDNS use this extra TXT record information is inefficient, but not fatal. However, when a Discovery Proxy aggregates data from multiple printers on a link, and sends it via unicast (via UDP or TCP) this amount of unnecessary TXT record information can result in large responses. A DNS reply over TCP carrying information about 70 printers with an average of 700 bytes per printer adds up to about 50 kilobytes of data. Therefore, a Discovery Proxy that is aware of the specifics of an application-layer protocol such as AirPrint (which uses IPP) can elide unnecessary key/value pairs from the DNS-SD TXT record for better network efficiency.
Also, the DNS-SD TXT record for many printers contains an "adminurl" key something like "adminurl=http://printername.local/status.html". For this URL to be useful outside the local link, the embedded ".local" hostname needs to be translated to an appropriate name with larger scope. It is easy to translate ".local" names when they appear in well-defined places, either as a record's name, or in the rdata of record types like PTR and SRV. In the printing case, some application-specific knowledge about the semantics of the "adminurl" key is needed for the Discovery Proxy to know that it contains a name that needs to be translated. This is somewhat analogous to the need for NAT gateways to contain ALGs (Application-Specific Gateways) to facilitate the correct translation of protocols that embed addresses in unexpected places.
As is the case with NAT ALGs, protocol designers are advised to avoid communicating names and addresses in nonstandard locations, because those "hidden" names and addresses are at risk of not being translated when necessary, resulting in operational failures. In the printing case, the operational failure of failing to translate the "adminurl" key correctly is that, when accessed from a different link, printing will still work, but clicking the "Admin" UI button will fail to open the printer's administration page. Rather than duplicating the host name from the service's SRV record in its "adminurl" key, thereby having the same host name appear in two places, a better design might have been to omit the host name from the "adminurl" key, and instead have the client implicitly substitute the target host name from the service's SRV record in place of a missing host name in the "adminurl" key. That way the desired host name only appears once, and it is in a well-defined place where software like the Discovery Proxy is expecting to find it.
Note that this kind of Application-Specific Data Translation is expected to be very rare. It is the exception, rather than the rule. This is an example of a common theme in computing. It is frequently the case that it is wise to start with a clean, layered design, with clear boundaries. Then, in certain special cases, those layer boundaries may be violated, where the performance and efficiency benefits outweigh the inelegance of the layer violation.
These layer violations are optional. They are done primarily for efficiency reasons, and generally should not be required for correct operation. A Discovery Proxy MAY operate solely at the mDNS layer, without any knowledge of semantics at the DNS-SD layer or above.
In a simple analysis, simply gathering multicast answers and forwarding them in a unicast response seems adequate, but it raises the question of how long the Discovery Proxy should wait to be sure that it has received all the Multicast DNS answers it needs to form a complete Unicast DNS response. If it waits too little time, then it risks its Unicast DNS response being incomplete. If it waits too long, then it creates a poor user experience at the client end. In fact, there may be no time which is both short enough to produce a good user experience and at the same time long enough to reliably produce complete results.
Similarly, the Discovery Proxy -- the authoritative name server for the subdomain in question -- needs to decide what DNS TTL to report for these records. If the TTL is too long then the recursive (caching) name servers issuing queries on behalf of their clients risk caching stale data for too long. If the TTL is too short then the amount of network traffic will be more than necessary. In fact, there may be no TTL which is both short enough to avoid undesirable stale data and at the same time long enough to be efficient on the network.
Both these dilemmas are solved by use of DNS Long-Lived Queries (DNS LLQ) [LLQ] or its newer replacement, DNS Push Notifications.
Clients supporting unicast DNS Service Discovery SHOULD implement DNS Push Notifications for improved user experience.
Clients and Discovery Proxies MAY support both DNS LLQ and DNS Push, and when talking to a Discovery Proxy that supports both, the client may use either protocol, as it chooses, though it is expected that only DNS Push will continue to be supported in the long run.
When a Discovery Proxy receives a query using DNS LLQ or DNS Push Notification, it responds immediately using the Multicast DNS records it already has in its cache (if any). This provides a good client user experience by providing a near-instantaneous response. Simultaneously, the Discovery Proxy issues a Multicast DNS query on the local link to discover if there are any additional Multicast DNS records it did not already know about. Should additional Multicast DNS responses be received, these are then delivered to the client using additional DNS LLQ or DNS Push Notification update messages. The timeliness of such update messages is limited only by the timeliness of the device responding to the Multicast DNS query. If the Multicast DNS device responds quickly, then the update message is delivered quickly. If the Multicast DNS device responds slowly, then the update message is delivered slowly. The benefit of using update messages is that the Discovery Proxy can respond promptly because it doesn't have to delay its unicast response to allow for the expected worst-case delay for receiving all the Multicast DNS responses. Even if a proxy were to try to provide reliability by assuming an excessively pessimistic worst-case time (thereby giving a very poor user experience) there would still be the risk of a slow Multicast DNS device taking even longer than that (e.g., a device that is not even powered on until ten seconds after the initial query is received) resulting in incomplete responses. Using update message solves this dilemma: even very late responses are not lost; they are delivered in subsequent update messages.
There are two factors that determine specifically how responses are generated:
The first factor is whether the query from the client used LLQ or DNS Push Notification (typical with long-lived service browsing PTR queries) or not (typical with one-shot operations like SRV or address record queries). Note that queries using LLQ or DNS Push Notification are received directly from the client. Queries not using LLQ or DNS Push Notification are generally received via the client's configured recursive (caching) name server.
The second factor is whether the Discovery Proxy already has at least one record in its cache that positively answers the question.
Note that the "negative responses" referred to above are "no error no answer" negative responses, not NXDOMAIN. This is because the Discovery Proxy cannot know all the Multicast DNS domain names that may exist on a link at any given time, so any name with no answers may have child names that do exist, making it an "empty nonterminal" name.
The MNAME field SHOULD contain the host name of the Discovery Proxy device (i.e., the same domain name as the rdata of the NS record delegating the relevant zone(s) to this Discovery Proxy device).
The RNAME field SHOULD contain the mailbox of the person responsible for administering this Discovery Proxy device.
The SERIAL field MUST be zero.
Zone transfers are undefined for Discovery Proxy zones, and consequently the REFRESH, RETRY and EXPIRE fields have no useful meaning for Discovery Proxy zones. These fields SHOULD contain reasonable default values. The RECOMMENDED values are: REFRESH 7200, RETRY 3600, EXPIRE 86400.
The MINIMUM field (used to control the lifetime of negative cache entries) SHOULD contain the value 10. The value of ten seconds is chosen based on user-experience considerations (see Section 5.5.1).
In the event that there are multiple Discovery Proxy devices on a link for fault tolerance reasons, this will result in clients receiving inconsistent SOA records (different MNAME, and possibly RNAME) depending on which Discovery Proxy answers their SOA query. However, since clients generally have no reason to use the MNAME or RNAME data, this is unlikely to cause any problems.
In the event that there are multiple Discovery Proxy devices on a link for fault tolerance reasons, the parent zone MUST be configured with glue records giving the names and addresses of all the Discovery Proxy devices on the link.
Each Discovery Proxy device MUST be configured with its own NS record, and with the NS records of its fellow Discovery Proxy devices on the same link, so that it can return the correct answers for NS queries.
In the event that a Discovery Proxy implements Long-Lived Queries and/or DNS Push Notifications (as most SHOULD) they MUST generate answers for the appropriate corresponding _dns‑llq._udp.<zone> and/or _dns‑push‑tls._tcp.<zone> SRV record queries. These records are conceptually inserted into the namespace of the corresponding zones. They do not exist in the ".local" namespace of the local link.
The Discovery Proxy acts as the authoritative name server for designated subdomains, and if DNSSEC is to be used, the Discovery Proxy needs to possess a copy of the signing keys, in order to generate authoritative signed data from the local Multicast DNS responses it receives. Off-line signing not applicable to Discovery Proxy.
In DNSSEC, NSEC and NSEC3 records are used to assert the nonexistence of certain names, also described as "authenticated denial of existence".
Since a Discovery Proxy only knows what names exist on the local link by issuing queries for them, and since it would be impractical to issue queries for every possible name just to find out which names exist and which do not, a Discovery Proxy cannot programatically synthesize the traditional NSEC and NSEC3 records which assert the nonexistence of a large range names. Instead, when generating a negative response, a Discovery Proxy programatically synthesizes a single NSEC record assert the nonexistence of just the specific name queried, and no others. Since the Discovery Proxy has the zone signing key, it can do this on demand. Since the NSEC record asserts the nonexistence of only a single name, zone walking is not a concern, so NSEC3 is not necessary.
Note that this applies only to traditional immediate DNS queries, which may return immediate negative answers when no immediate positive answer is available. When used with a DNS Push Notification subscription there are no negative answers, merely the absence of answers so far, which may change in the future if answers become available.
An IPv6-only host and an IPv4-only host behave as "ships that pass in the night". Even if they are on the same Ethernet, neither is aware of the other's traffic. For this reason, each link may have *two* unrelated ".local." zones, one for IPv6 and one for IPv4. Since for practical purposes, a group of IPv6-only hosts and a group of IPv4-only hosts on the same Ethernet act as if they were on two entirely separate Ethernet segments, it is unsurprising that their use of the ".local." zone should occur exactly as it would if they really were on two entirely separate Ethernet segments.
It will be desirable to have a mechanism to 'stitch' together these two unrelated ".local." zones so that they appear as one. Such mechanism will need to be able to differentiate between a dual-stack (v4/v6) host participating in both ".local." zones, and two different hosts, one IPv6-only and the other IPv4-only, which are both trying to use the same name(s). Such a mechanism will be specified in a future companion document.
At present, it is RECOMMENDED that a Discovery Proxy be configured with a single domain name for both the IPv4 and IPv6 ".local." zones on the local link, and when a unicast query is received, it should issue Multicast DNS queries using both IPv4 and IPv6 on the local link, and then combine the results.
A service proves its presence on a link by its ability to answer link-local multicast queries on that link. If greater security is desired, then the Discovery Proxy mechanism should not be used, and something with stronger security should be used instead, such as authenticated secure DNS Update [RFC2136] [RFC3007].
The Domain Name System is, generally speaking, a global public database. Records that exist in the Domain Name System name hierarchy can be queried by name from, in principle, anywhere in the world. If services on a mobile device (like a laptop computer) are made visible via the Discovery Proxy mechanism, then when those services become visible in a domain such as "My House.example.com" that might indicate to (potentially hostile) observers that the mobile device is in my house. When those services disappear from "My House.example.com" that change could be used by observers to infer when the mobile device (and possibly its owner) may have left the house. The privacy of this information may be protected using techniques like firewalls, split-view DNS, and Virtual Private Networks (VPNs), as are customarily used today to protect the privacy of corporate DNS information.
The Discovery Proxy could also provide sensitive records only to authenticated users. This is a general DNS problem, not specific to the Discovery Proxy. Work is underway in the IETF to tackle this problem [RFC7626].
A remote attacker could use a rapid series of unique Unicast DNS queries to induce a Discovery Proxy to generate a rapid series of corresponding Multicast DNS queries on one or more of its local links. Multicast traffic is generally more expensive than unicast traffic -- especially on Wi-Fi links -- which makes this attack particularly serious. To limit the damage that can be caused by such attacks, a Discovery Proxy (or the underlying Multicast DNS subsystem which it utilizes) MUST implement Multicast DNS query rate limiting appropriate to the link technology in question. For today's 802.11b/g/n/ac Wi-Fi links (for which approximately 200 multicast packets per second is sufficient to consume approximately 100% of the wireless spectrum) a limit of 20 Multicast DNS query packets per second is RECOMMENDED. On other link technologies like Gigabit Ethernet higher limits may be appropriate. A consequence of this rate limiting is that a rogue remote client could issue an excessive number of queries, resuling in denial of service to other remote clients attempting to use that Discovery Proxy. However, this is preferable to a rogue remote client being able to inflict even greater harm on the local network, which could impact the correct operation of all local clients on that network.
Apple has submitted an IPR disclosure concerning the technique proposed in this document. Details are available on the IETF IPR disclosure page.
This document has no IANA Considerations.
Thanks to Markus Stenberg for helping develop the policy regarding the four styles of unicast response according to what data is immediately available in the cache. Thanks to Anders Brandt, Tim Chown, Ralph Droms, Ray Hunter, Ted Lemon, Tom Pusateri, Markus Stenberg, Dave Thaler, and Andrew Yourtchenko for their comments.
| [RFC1034] | Mockapetris, P., "Domain names - concepts and facilities", STD 13, RFC 1034, DOI 10.17487/RFC1034, November 1987. |
| [RFC1035] | Mockapetris, P., "Domain names - implementation and specification", STD 13, RFC 1035, DOI 10.17487/RFC1035, November 1987. |
| [RFC1918] | Rekhter, Y., Moskowitz, B., Karrenberg, D., de Groot, G. and E. Lear, "Address Allocation for Private Internets", BCP 5, RFC 1918, DOI 10.17487/RFC1918, February 1996. |
| [RFC2119] | Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997. |
| [RFC2308] | Andrews, M., "Negative Caching of DNS Queries (DNS NCACHE)", RFC 2308, DOI 10.17487/RFC2308, March 1998. |
| [RFC3629] | Yergeau, F., "UTF-8, a transformation format of ISO 10646", STD 63, RFC 3629, DOI 10.17487/RFC3629, November 2003. |
| [RFC3927] | Cheshire, S., Aboba, B. and E. Guttman, "Dynamic Configuration of IPv4 Link-Local Addresses", RFC 3927, DOI 10.17487/RFC3927, May 2005. |
| [RFC4862] | Thomson, S., Narten, T. and T. Jinmei, "IPv6 Stateless Address Autoconfiguration", RFC 4862, DOI 10.17487/RFC4862, September 2007. |
| [RFC5198] | Klensin, J. and M. Padlipsky, "Unicode Format for Network Interchange", RFC 5198, DOI 10.17487/RFC5198, March 2008. |
| [RFC6762] | Cheshire, S. and M. Krochmal, "Multicast DNS", RFC 6762, December 2012. |
| [RFC6763] | Cheshire, S. and M. Krochmal, "DNS-Based Service Discovery", RFC 6763, December 2012. |
| [PUSH] | Pusateri, T. and S. Cheshire, "DNS Push Notifications", Internet-Draft draft-ietf-dnssd-push-12, July 2017. |
Some aspects of the mechanism specified in this document already exist in deployed software. Some aspects are new. This section outlines which aspects already exist and which are new.
Domain enumeration by the client (the "b._dns-sd._udp" queries) is already implemented and deployed.
Unicast queries to the indicated discovery domain is already implemented and deployed.
These are implemented and deployed in Mac OS X 10.4 and later (including all versions of Apple iOS, on all iPhone and iPads), in Bonjour for Windows, and in Android 4.1 "Jelly Bean" (API Level 16) and later.
Domain enumeration and unicast querying have been used for several years at IETF meetings to make Terminal Room printers discoverable from outside the Terminal room. When an IETF attendee presses Cmd-P on a Mac, or selects AirPrint on an iPad or iPhone, and the Terminal room printers appear, that is because the client is sending unicast DNS queries to the IETF DNS servers.
A minimal portable Discovery Proxy implementation has been produced by Markus Stenberg and Steven Barth, which runs on OS X and several Linux variants including OpenWrt [ohp]. It was demonstrated at the Berlin IETF in July 2013.
Tom Pusateri also has an implementation that runs on any Unix/Linux. It has a RESTful interface for management and an experimental demo CLI and web interface.
The current APIs make multiple domains visible to client software, but most client UI today lumps all discovered services into a single flat list. This is largely a chicken-and-egg problem. Application writers were naturally reluctant to spend time writing domain-aware UI code when few customers today would benefit from it. If Discovery Proxy deployment becomes common, then application writers will have a reason to provide better UI. Existing applications will work with the Discovery Proxy, but will show all services in a single flat list. Applications with improved UI will group services by domain.
The Long-Lived Query mechanism [LLQ] referred to in this specification exists and is deployed, but has not been standardized by the IETF. The IETF is considering standardizing a superior Long-Lived Query mechanism called DNS Push Notifications [PUSH]. The pragmatic short-term deployment approach is for vendors to produce Discovery Proxies that implement both the deployed Long-Lived Query mechanism [LLQ] (for today's clients) and the new DNS Push Notifications mechanism [PUSH] as the preferred long-term direction.
The translating/filtering Discovery Proxy specified in this document. Implementations are under development, and operational experience with these implementations has guided updates to this document.
Client implementations of the new DNS Push Notifications mechanism [PUSH] are currently underway.
A mechanism to 'stitch' together multiple ".local." zones so that they appear as one. Such a stitching mechanism will be specified in a future companion document. This stitching mechanism addresses the issue that if a printer is physically moved from one link to another, then conceptually the old service has disappeared from the DNS namespace, and a new service with a similar name has appeared. This stitching mechanism will allow a service to change its point of attachment without changing the name by which it can be found.