| dnsext | C. Contavalli |
| Internet-Draft | W.W. van der Gaast |
| Intended status: Experimental | |
| Expires: January 06, 2014 | S. Leach |
| VeriSign | |
| E. Lewis | |
| Neustar | |
| July 05, 2013 |
Client Subnet in DNS Requests
draft-vandergaast-edns-client-subnet-02
This draft defines an EDNS0 extension to carry information about the network that originated a DNS query, and the network for which the subsequent reply can be cached.
This document describes an experimental EDNS0 option. The purpose of this experiment is to discover if the information carried in an edns-client-subnet option is sufficiently helpful to Authoritative Nameservers to give better Optimized Replies, and to measure the latency improvement from this better reply, taking into account a likely lower cache hit rate at (and generally higher load on) the Recursive Resolver, and possibly other operational aspects. Details on how this experiment will be carried out can be found in Section 12.
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 January 06, 2014.
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.
Many Authoritative Nameservers today return different replies based on the perceived topological location of the user. These servers use the IP address of the incoming query to identify that location. Since most queries come from intermediate recursive resolvers, the source address is that of the Recursive Resolver rather than of the query originator.
Traditionally and probably still in the majority of instances, recursive resolvers are reasonably close in the topological sense to the stub resolvers or forwarders that are the source of queries. For these resolvers, using their own IP address is sufficient for authority servers that tailor responses based upon location of the querier.
Increasingly though a class of Recursive Resolvers has arisen that serves query sources without regard to topology. The motivation for a query source to use such a Third-party Resolver varies but is usually because of some enhanced experience, such as greater cache security or applying policies regarding where users may connect. (Although political censorship usually comes to mind here, the same actions may be used by a parent when setting controls on where a minor may connect.) When using a Third-party Resolver, there can no longer be any assumption of close proximity between the originator and the recursive resolver, leading to less than optimal replies from the authority servers.
A similar situation exists within some ISPs where the Recursive Resolvers are topologically distant from some edges of the ISP network, resulting in less than optimal replies from the authority servers.
This draft defines an EDNS0 option to convey network information that is relevant to the message but not otherwise included in the datagram. This will provide the mechanism to carry sufficient network information about the originator for the authority server to tailor responses. It also provides for the authority server to indicate the scope of network addresses that the tailored answer is intended. This EDNS0 option is intended for those recursive and authority servers that would benefit from the extension and not for general purpose deployment. It is completely optional and can safely be ignored by servers that choose not to implement it or enable it.
This draft also includes guidelines on how to best cache those results and provides recommendations on when this protocol extension should be used.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in [RFC2119].
The general idea of this document is to provide an EDNS0 option so that Recursive Resolvers can, if they are willing to, forward details about the network a query is coming from when talking to other Nameservers.
The format of this option is described in Section 4, and is meant to be added in queries sent by Intermediate Nameservers in a way transparent to Stub Resolvers and end users, as described in Section 5.1.
As described in Section 5.2, an Authoritative Nameserver could use this EDNS0 option as a hint to better locate the network of the end user, and provide a better answer.
Its reply would contain an EDNS0 client-subnet option, clearly indicating that (1) the server made use of this information and (2) the answer is tied to the network of the client.
As described in Section 5.3, Intermediate Nameservers would use this information to cache the reply.
Some Intermediate Nameservers may also have to be able to forward edns-client-subnet queries they receive. This is described in Section 5.4.
The mechanisms provided by edns-client-subnet raise various security related concerns, related to cache growth, the ability to spoof EDNS0 options, and privacy. Section 9 explores various mitigation techniques.
The expectation, however, is that this option will only be enabled (and used) by Recursive Resolvers and Authoritative Nameserver that incur geolocation issues.
Most Recursive Resolvers, Authoritative Nameservers and Stub Resolver will never know about this option, and keep working as usual.
Failure to support this option or its improper handling will at worst cause sub-optimal geolocation, which is a pretty common occurrence in current CDN setups and not a cause of concern.
Section 5.1 also provides a mechanism for Stub Resolvers to signal Recursive Resolvers that they do not want an edns-client-subnet with their network to be added.
Additionally, owners of resolvers with edns-client-subnet enabled are allowed to choose how many bits of the address of received queries to forward, or to reduce the number of bits forwarded for queries already including an edns-client-subnet option.
This draft uses an EDNS0 ([RFC2671]) option to include client IP information in DNS messages. The option is structured as follows:
+0 (MSB) +1 (LSB)
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
0: | OPTION-CODE |
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
2: | OPTION-LENGTH |
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
4: | FAMILY |
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
6: | SOURCE NETMASK | SCOPE NETMASK |
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
7: | ADDRESS... /
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
The format of the address part depends on the value of FAMILY. This document only defines the format for FAMILY 1 (IP version 4) and 2 (IP version 6), which are as follows:
All fields are in network byte order. Throughout the document, we will often refer to "longer" or "shorter" netmasks, corresponding to netmasks that have a "higher" or "lower" value when represented as integers.
The edns-client-subnet option should generally be added by Recursive Resolvers when querying other servers, as described in Section 10.
In this option, the server should include the IP of the client that caused the query to be generated, truncated to a number of bits specified in the SOURCE NETMASK field.
The IP of the client can generally be determined by looking at the source IP indicated in the IP header of the request.
A Stub Resolver MAY generate DNS queries with an edns-client-subnet option with SOURCE NETMASK set to 0 (i.e. 0.0.0.0/0) to indicate that the Recursive Resolver MUST NOT add address information of the client to its queries. The Stub Resolver may also add non-empty edns-client-subnet options to its queries, but Recursive Resolvers are not required to accept/use this information.
For privacy reasons, and because the whole IP address is rarely required to determine an optimized reply, the ADDRESS field in the option SHOULD be truncated to a certain number of bits, chosen by the administrators of the server, as described in Section 9.
When a query containing an edns-client-subnet option is received, an Authoritative Nameserver supporting edns-client-subnet MAY use the address information specified in the option in order to generate an optimized reply.
Authoritative servers that have not implemented or enabled support for the edns-client-subnet may safely ignore the option within incoming queries. Such a server MUST NOT include an edns-client-subnet option within replies, to indicate lack of support for the option.
Requests with wrongly formatted options (i.e. wrong size) MUST be rejected and a FORMERR response must be returned to the sender, as described by [RFC2671], Transport Considerations.
If the Authoritative Nameserver decides to use information from the edns-client-subnet option to calculate a response, it MUST include the option in the response to indicate that the information was used (and has to be cached accordingly). If the option was not included in a query, it MUST NOT be included in the response.
The FAMILY, ADDRESS and SOURCE NETMASK in the response MUST match those in the request. Echoing back the address and netmask helps to mitigate certain attack vectors, as described in Section 9.
The SCOPE NETMASK in the reply indicates the netmask of the network that the answer is intended for.
A SCOPE NETMASK value larger than the SOURCE NETMASK indicates that the address and netmask provided in the query was not specific enough to select a single, best response, and that an optimal reply would require at least SCOPE NETMASK bits of address information.
Conversely, a shorter SCOPE NETMASK indicates that more bits than necessary were provided.
As not all netblocks are the same size, an Authoritative Nameserver may return different values of SCOPE NETMASK for different networks.
In both cases, the value of the SCOPE NETMASK in the reply has strong implications with regard to how the reply will be cached by Intermediate Nameservers, as described in Section 5.3.
If the edns-client-subnet option in the request is not used at all (for example if an optimized reply was temporarily unavailable or not supported for the requested domain name), a server supporting edns-client-subnet MUST indicate that no bits of the ADDRESS in the request have been used by specifying a SCOPE NETMASK of 0 (equivalent to the networks 0.0.0.0/0 or ::/0).
If no optimized answer could be found at all for the FAMILY, ADDRESS and SOURCE NETMASK indicated in the query, the Authoritative Nameserver SHOULD still return the best result it knows of (i.e. by using the query source IP address instead, or a sensible default), and indicate that this result should only be cached for the FAMILY, ADDRESS and SOURCE NETMASK indicated in the request. The server will indicate this by copying the SOURCE NETMASK into the SCOPE NETMASK field.
When an Intermediate Nameserver receives a reply containing an edns-client-subnet option, it will return a reply to its client and may cache the result.
If the FAMILY, ADDRESS and SOURCE NETMASK fields in the reply don't match the fields in the corresponding request, the full reply MUST be dropped, as described in Section 9.
In the cache, any resource record in the answer section will be tied to the network specified by the FAMILY, ADDRESS and SCOPE NETMASK fields, as detailed below. Note that the additional and authority sections from a DNS response message are specifically excluded here.
If another query is received matching the entry in the cache, the resolver will verify that the FAMILY and ADDRESS that represent the client match any of the networks in the cache for that entry.
If the address of the client is within any of the networks in the cache, then the cached response MUST be returned as usual. In case the address of the client matches multiple networks in the cache, the entry with the highest SCOPE NETMASK value MUST be returned, as with most route-matching algorithms.
If the address of the client does not match any network in the cache, then the Recursive Resolver MUST behave as if no match was found and perform resolution as usual. This is necessary to avoid sub-optimal replies in the cache from being returned to the wrong clients, and to avoid a single request coming from a client on a different network from polluting the cache with a sub-optimal reply for all the users of that resolver.
Note that every time a Recursive Resolver queries an Authoritative Nameserver by forwarding the edns-client-subnet option that it received from another client, a low SOURCE NETMASK in the original request could cause a sub-optimal reply to be returned by the Authoritative Nameserver.
To avoid this sub-optimal reply from being served from cache for clients for which a better reply would be available, the Recursive Resolver MUST check the SCOPE NETMASK that was returned by the Authoritative Nameserver:
When another request is performed, the existing entries SHOULD be kept in the cache until their TTL expires, as per standard behavior.
As another reply is received, the reply will be tied to a different network. The server SHOULD keep in cache both replies, and return the most appropriate one depending on the address of the client.
Although omitting this behaviour will significantly simplify an implementation, the resulting drop in cache hits is very likely to defeat most latency benefits provided by edns-client-subnet. Therefore, when implementing this option for latency purposes, implementing full caching support as described in this section is STRONGLY RECOMMENDED.
Any reply containing an edns-client-subnet option considered invalid should be treated as if no edns-client-subnet option was specified at all.
Replies coming from servers not supporting edns-client-subnet or otherwise not containing an edns-client-subnet option SHOULD be considered as containing a SCOPE NETMASK of 0 (e.g., cache the result for 0.0.0.0/0 or ::/0) for all the supported families.
In any case, the response from the resolver to the client MUST NOT contain the edns-client-subnet option if none was present in the client's original request. If the original client request contained a valid edns-client-subnet option that was used during recursion, the Recursive Resolver MUST include the edns-client-subnet option from the Authoritative Nameserver response in the response to the client.
Enabling support for edns-client-subnet in a recursive resolver will significantly increase the size of the cache, reduce the number of results that can be served from cache, and increase the load on the server. Implementing the mitigation techniques described in Section 9 is strongly recommended.
Generally, edns-client-subnet options will only be present in DNS messages between a Recursive Resolver and an Authoritative Nameserver, i.e. one hop. In certain configurations however (for example multi-tier nameserver setups), it may be necessary to implement transitive behaviour on Intermediate Nameservers.
It is important that any Intermediate Nameserver that implements transitive behaviour (i.e. forward edns-client-subnet options received from their clients) MUST fully implement the caching behaviour described in Section 5.3.
Intermediate Nameservers (including Recursive Resolvers) supporting edns-client-subnet MUST forward options with SOURCE NETMASK set to 0 (i.e. anonymized), such an option MUST NOT be replaced with an option with more accurate address information.
An Intermediate Nameserver MAY also forward edns-client-subnet options with actual address information. This information MAY match the source IP address of the incoming query, and MAY have more or less address bits than the Nameserver would normally include in a locally originated edns-client-subnet option.
If for any reason the Intermediate Nameserver does not want to use the information in an edns-client-subnet option it receives (too little address information, network address from an IP range not authorized to use the server, private/unroutable address space, ...) it SHOULD drop the query and return a REFUSED response. Note again that an edns-client-subnet option with 0 address bits MUST NOT be refused.
IANA has assigned option code 8 in the "DNS EDNS0 Option Codes (OPT)" registry to edns-client-subnet.
The presence or absence of an OPT resource record containing an edns-client-subnet option in a DNS query does not change the usage of those resource records and mechanisms used to provide data origin authentication and data integrity to the DNS, as described in [RFC4033], [RFC4034] and [RFC4035].
Special awareness of edns-client-subnet in devices that perform NAT as described in [RFC2663] is not required, queries can be passed through as-is. The client's network address SHOULD NOT be added, and existing edns-client-subnet options, if present, SHOULD NOT be modified by NAT devices.
In large-scale global networks behind NAT (but for example with centralized DNS infrastructure), an internal Intermediate Nameserver may have detailed network layout information, and may know which external subnets are used for egress traffic by each internal network. In such cases, the Intermediate Nameserver MAY use that information when originating edns-client-subnet options.
In other cases, Recursive Resolvers sited behind NAT SHOULD NOT originate edns-client-subnet options with their external IP address, and instead rely on downstream Intermediate Nameservers doing so.
With the edns-client-subnet option, the network address of the client that initiated the resolution becomes visible to all servers involved in the resolution process. Additionally, it will be visible from any network traversed by the DNS packets.
To protect users' privacy, Recursive Resolvers are strongly encouraged to conceal part of the IP address of the user by truncating IPv4 addresses to 24 bits. No recommendation is provided for IPv6 at this time, but IPv6 addresses should be similarly truncated in order to not allow to uniquely identify the client.
ISPs will often have more detailed knowledge of their own networks. I.e. they will know if all 24-bit prefixes in a /20 are in the same area. In those cases, for optimal cache utilization and improved privacy, the ISP's Recursive Resolver SHOULD truncate IP addresses in this /20 to just 20 bits, instead of 24 as recommended above.
Users who wish their full IP address to be hidden can include an edns-client-subnet option specifying the wildcard address 0.0.0.0/0 (i.e. FAMILY set to 1 (IPv4), SOURCE NETMASK to 0 and no ADDRESS). As described in previous sections, this option will be forwarded across all the Recursive Resolvers supporting edns-client-subnet, which MUST NOT modify it to include the network address of the client.
Note that even without edns-client-subnet options, any server queried directly by the user will be able to see the full client IP address. Recursive Resolvers or Authoritative Nameservers MAY use the source IP address of requests to return a cached entry or to generate an optimized reply that best matches the request.
edns-client-subnet adds information to the q-tuple. This allows an attacker to send a caching Intermediate Nameserver multiple queries with spoofed IP addresses either in the edns-client-subnet option or as the source IP. These queries will trigger multiple outgoing queries with the same name, type and class, just different address information in the edns-client-subnet option.
With multiple queries for the same name in flight, the attacker has a higher chance of success in sending a matching response (with the address 0.0.0.0/0 to still get it cached for many hosts).
To counter this, every edns-client-subnet option in a response packet MUST contain the full FAMILY, ADDRESS and SOURCE NETMASK fields from the corresponding request. Intermediate Nameservers processing a response MUST verify that these match, and MUST discard the entire reply if they do not.
It is simple for an arbitrary resolver or client to provide false information in the edns-client-subnet option, or to send UDP packets with forged source IP addresses.
This could be used to:
Even without malicious intent, Third-party Resolvers providing answers to clients in multiple networks will need to cache different replies for different networks, putting more pressure on the cache.
To mitigate those problems:
When implementing a Recursive Resolver, there are two strategies on deciding when to include an edns-client-subnet option in a query. At this stage it's not clear which strategy is best.
A Recursive Resolver can send the edns-client-subnet option with every outgoing query. However, it is RECOMMENDED that Resolvers remember which Authoritative Nameservers did not return the option with their response, and omit client address information from subsequent queries to those Nameservers.
Additionally, Recursive Resolvers MAY be configured to never send the option when querying root and TLD servers, as these are unlikely to generate different replies based on the IP of the client.
When probing, it is important that several things are probed: support for edns-client-subnet, support for EDNS0, support for EDNS0 options, or possibly an unreachable Nameserver. Various implementations are known to drop DNS packets with OPT RRs (with or without options), thus several probes are required to discover what is supported.
Probing, if implemented, MUST be repeated periodically (i.e. daily). If an Authoritative Nameserver indicates edns-client-subnet support for one zone, it is to be expected that the Nameserver supports edns-client-subnet for all its zones. Likewise, an Authoritative Nameserver that uses edns-client-subnet information for one of its zones, MUST indicate support for the option in all its responses. If the option is supported but not actually used for generating a response, its SCOPE NETMASK value SHOULD be set to 0.
As described previously, it is expected that only a few Recursive Resolvers will need to use edns-client-subnet, and that it will generally be enabled only if it offers a clear benefit to the users.
To avoid the complexity of implementing a probing and detection mechanism (and the possible query loss/delay that may come with it), an implementation could decide to use a statically configured whitelist of Authoritative Namesevers to send the option to. Implementations MAY also allow additionally configuring this based on other criteria (i.e. zone, qtype).
An additional advantage of using a whitelist is that partial client address information is only disclosed to Nameservers that are known to use the information, improving privacy.
A major drawback is scalability. The operator needs to track which Nameservers support edns-client-subnet, making it harder for new Authoritative Nameservers to start using the option.
This document describes an experiment to be conducted on the Internet. Participation requires the careful eye on the proposed EDNS0 option. What is described in this document may need to be altered or adjusted as experience dictates.
During the experiment, participants will enable edns-client-subnet support on their Nameservers (both Intermediate Nameservers and Authoritative Nameservers).
Intermediate Nameservers will be configured with whitelists so edns-client-subnet options will be sent only to Authoritative Nameservers participating in the experiment. Some participants might also choose to experiment with the probing behaviour described in Section 10.1.
Authoritative Nameservers could be configured with a whitelist but maintainers are strongly encouraged to accept edns-client-subnet options from anywhere to encourage openness. ns1-ns4.google.com are configured as such and can be used to test client/resolver implementations.
Pending official code point allocation (as described in Section 6), a temporary EDNS0 option code will be in use. This code is not guaranteed to be used in further stages of this effort - assuming the are further stages.
The effects of edns-client-subnet will be measured using data of 24-hour periods with and without the option enabled on a Recursive Resolver. Metrics to be compared include:
Success of the experiment will depend on the level of improvement in the last mentioned metric. Others are for operational reasons (i.e. does one need to provision for extra Nameservers?).
Possible side-effects from using edns-client-subnet will also be investigated. These may include interoperability problems within the DNS and decreased (or possibly increased) ease of troubleshooting.
The authors wish to thank Darryl Rodden for his work as a co-author on previous versions, and the following people for reviewing early drafts of this document and for providing useful feedback: Paul S. R. Chisholm, B. Narendran, Leonidas Kontothanassis, David Presotto, Philip Rowlands, Chris Morrow, Kara Moscoe, Alex Nizhner, Warren Kumari, Richard Rabbat from Google, Terry Farmer, Mark Teodoro, Edward Lewis, Eric Burger from Neustar, David Ulevitch, Matthew Dempsky from OpenDNS, Patrick W. Gilmore from Akamai, Colm MacCarthaigh, Richard Sheehan and all the other people that replied to our emails on various mailing lists.
| [RFC2663] | Srisuresh, P. and M. Holdrege, "IP Network Address Translator (NAT) Terminology and Considerations", RFC 2663, |