mDNS X-link review
draft-otis-dnssd-mdns-xlink-04

Abstract

Multicast DNS will not normally extend beyond the MAC Bridge. This limitation is problematic when desired services are beyond the reach of multicast mDNS. This document explores security considerations when overcoming this limitation.

Requirements Language

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].

Status of This Memo

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 3, 2015.

Copyright Notice

Copyright (c) 2014 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.

Table of Contents

• 1. Introduction
• 2. IANA Considerations
• 3. Security Considerations
• 3.1. Multiple Link Strategies
• 3.2. Scope of Discovery
• 3.3. Multiple Namespaces
• 3.4. Authorization
• 3.5. Authentication
• 3.6. Privacy Considerations
• 4. Acknowledgements
• 5. References
• 5.1. References - Informative
• 5.2. References - Normative
• Author's Address

1. Introduction

On Bridged LANs, as described by [IEEE.802-1D.1993], MAC entities make their services known via multicast. Multicast forms a basis for networking and layer 3 protocol initialization where [mDNS] together with [DNS-SD] provides a structure for multicast announcements within the LAN environment.

A Bridge acts as an interconnect mechanism transparent to end stations on LANs. Bridges designated to forward frames is normally accomplished by participation in a Spanning Tree Algorithm. Many expect [mDNS] resource records can be safely and automatically placed into [DNS] to overcome Bridge to Bridge multicast limitations. Nevertheless, such a process must operate in conjunction with requisite controls necessary to retain network security.

A Bridge forwards frames based on prior source MAC associations with incoming frames on different LAN ports. Source MAC and LAN port associations are recommended to expire in 300 seconds. Frames containing source multicast MACs are silently discarded as invalid. Frames containing a destination MAC on the same LAN port already associated with the MAC are silently discarded. A valid incoming frame with a destination not previously associated with a different LAN port is forwarded (flooded) to all other LAN ports, otherwise when a MAC destination address is associated with a different LAN port from which the frame was received, the frame is selectively forwarded to this port. All broadcast and multicast MACs are flooded to all other LAN ports because they do not represent a valid source. Flooding operations may create a storm of replicated frames having an unknown MAC destination whenever forwarding is enabled on LAN ports connected in a loop.

In [IEEE.802-11.1999] wireless networks, multicast frames are transmitted at a low data rate supported by all receivers. Multicast on wireless networks may thereby lower overall network throughput. Some network administrators block some multicast traffic or convert it to a series of link-layer unicast frames.

Wireless links may be orders of magnitude less reliable than their wired counterparts. To improve transmission reliability, [IEEE.802-11.1999] requires positive acknowledgement of unicast frames. It does not, however, support positive acknowledgement of multicast frames. As a result, it is common to observe much higher loss of multicast frames on wireless compared against wired network technologies.

2. IANA Considerations

This document requires no IANA consideration.

3. Security Considerations

Scalable DNS-SD (SSD) proposes to automatically gather autonomously named [mDNS] resource records by observing announcement traffic to then make routable resources visible and accessible from other networks via unicast [DNS] structured per [DNS-SD]. When doing so, address translation using ULAs [RFC4193] can offer a significant level of protection since typical link-local addresses are not usable from other networks. Not all devices exposed by publishing their resources in DNS can be expected able to handle resulting direct Internet access. For either home or enterprise networks, ULAs permit a local routing plane isolated from that of the Internet. As such, ULAs also permit local communications to remain unaffected by Internet related link failures.

ULAs avoid a need to renumber internal-only private nodes when changing ISPs, or when ISPs restructure their address allocations. In these situations, use of ULA offers an effective tool for protecting internal-only nodes. As such, more than just the security considerations discussed in [mDNS] and [DNS-SD] are needed. For example, [DNS-SD] states the following: "Since DNS-SD is just a specification for how to name and use records in the existing DNS, it has no specific additional security requirements over and above those that already apply to DNS queries and DNS updates." This simply overlooks many devices are not automatically published in DNS nor can it be assumed they are able to handle the direct access that DNS might permit.

Current BTMM [RFC6281] only publishes ULAs of hosts in DNS able to authenticate when setting up an overlay network. Remaining devices, such as printers, are only accessed as shared elements offered by the authenticating hosts. DNS resources should never be considered to offer privacy even in split-horizon configurations. DNS is unable to authenticate incoming queries nor can it offer application layer protection. Since many prefixes are expected to be in use within environments served by [I-D.cheshire-dnssd-hybrid], errors related to network boundary detections becomes critical. As such, DNS SHOULD NOT publish addresses of devices unable to authenticate sessions that traverse the Internet.

[DNS-SD] should not be viewed as only a catalog structure of desired services. [I-D.cheshire-dnssd-hybrid] is to be used to bridge adjacent networks, which risks conveying resources of hosts that are unable to safely facilitate Internet access. Since [I-D.cheshire-dnssd-hybrid] only expects to disclose routable addresses while also ignoring use of ULAs, this clearly expects conveyance of globally routable addresses, GUA. Use of ULAs instead of GUAs represents a significant change in strategy that permits limited devices to remain isolated from the Internet while still allowing packet routing between local network realms in much the same manner as are MACs employed by [RFC6325].

[I-D.cheshire-dnssd-hybrid] lacks a process able to limit resources being gathered, resolved, and propagated to those that can be administrated. As such, an [I-D.cheshire-dnssd-hybrid] scheme represents a profound change to network security. The following sections highlight potential threats posed by deploying [DNS-SD] over multiple links through the automated collection and publication of [mDNS] resources into [DNS] as proposed by [I-D.cheshire-dnssd-hybrid]. This conveyance expands namespaces into .local., .sitelocal., and [DNS] which may also cache Internet namespace.

This new routable namespace also lacks the benefit of registrar involvement and may not afford an administrator an ability to mitigate nefarious activity, such as spoofing and phishing, without requisite controls having been first carefully established. When a device has access to different realms on multiple interfaces, it is not even clear how simple conflict resolution avoids threatening network stability while resolving names conveyed over disparate technologies.

Managing autonomously named resources becomes especially salient since visually selected names are not ensured uniquely represented nor quickly resolved due to latency uncertainties. For example, [DNS] recommends 5 second timeouts with a doubling on two subsequent retries for a total of 35 seconds. [mDNS] only requires compliance with [RFC5198] rather than IDNA2008 [RFC5895]. This less restrictive use of the name space may impair the defense of critical services from look-alike attack. [mDNS] does not ensure instances are visually unique and allows spaces and punctuation not permitted by IDNA2008.

It is imperative for SSD to include requisite filtering necessary to prevent data ex-filtration or the interception of sensitive services. Any exchanged data must first ensure locality, limit the resources gathered, resolved, and propagated to just those elements that can be effectively administrated. It is critical to ensure normal network protection is not lost for hosts that depend on link-local addressing and exclusion of routable traffic. A printer would be one such example of a host that can not be upgraded.

3.1. Multiple Link Strategies

3.1.1. Selective Forwarding based on IGMP or MLD snooping

Internet Group Management Protocol (IGMP) [RFC3376] supports multicast on IPv4 networks. Multicast Listener Discovery (MLD) [RFC3810] supports multicast management on IPv6 networks using ICMPv6 messaging in contrast to IGMP's bare IP encapsulation. This management allows routers to announce their multicast membership to neighboring routers. To optimize which LANs receive forwarded multicast frames, IGMP or MLD snooping can be used to determine the presence of listeners as a means to permit selective forwarding of multicast frames as well.

3.1.2. RBridge

RBridges [RFC6325] are compatible with previous [IEEE.802-1D.1993] customer bridges as well as IPv4 and IPv6 routers and end nodes. RBridges may support either [IEEE.802-3.1998] or other link technologies. RBridges are invisible to current IP routers as bridges are and, like routers, terminate the Bridge spanning tree protocol. The RBridge design supports VLANs and optimization of the distribution of multi-destination frames based on VLAN ID or on IP-derived multicast groups. It also allows unicast forwarding tables at transit RBridges to be sized according to the number of RBridges (rather than the number of end nodes), which allows their forwarding tables to be substantially smaller than in conventional customer bridges.

Consolidating MAC/LAN port association at individual RBridge nodes does not necessarily combine all LANs into a flat MAC space. It is possible to filter RBridge to RBridge traffic, however such filtering is unlikely needed for a home or small office. In addition, RBridge also supports PPP [RFC6361] to facilitate the merger of other physical transport technologies. Use of RBridge can incrementally simplify the LAN environment while also eliminating bottlenecks imposed by a Spanning Tree Algorithm. Use of RBridge can also satisfy normal link-local network restrictions when necessary.

[RFC3927] provides an overview of IPv4 address complexities related to dealing with multiple segments and interfaces. IPv6 introduces new paradigms in respect to interface address assignments which offer scoping as explained in [RFC4291].

3.1.3. VLAN

Use of VLAN such as [RFC5517] can selectively extend multicast forwarding beyond Bridge limitations. While not a general solution, use of VLAN can both isolate and unite specific networks.

3.1.4. DHCP

IP address assignment and host registration might use a single or forwarded DHCP [RFC2131] or [RFC3315] server for IPv4 and IPv6 respectively that responds to interconnected networks as a means to register hosts and addresses. DHCP does not ensure against name or address conflict nor is it intended to configure routers.

3.1.5. Automated placement of mDNS resources into DNS

IP addresses made visible by [DNSSEC] or [DNS] that conform with [DNS-SD] might be used, but the automated population of information into [DNS] should be limited to administrative systems.

Automated conversion of [mDNS] into unicast [DNS] can be problematic from a security standpoint as can the widespread propagation of multicast frames. [mDNS] only requires compliance with [RFC5198] rather than IDNA2008 [RFC5895]. This means [mDNS] does not ensure instances are visually unique and may contain spaces and punctuation not permitted by IDNA2008. As such, this might allow users into becoming misled about the scope of a name.

Public Suffix lists might help simplify creation of A-Labels from UTF-8 user input by offering matching items for user selection. A [PublicSuffix] list represents DNS domain names reserved for registrations by appropriate authorities. This still leaves the domain registered above the public suffix, but its validation and encoding should involve fewer transactions.

Replacing ASCII punctuation and spaces in the label with the '_' character, except when located as the leftmost character, may reduce some handling issues related to end of string parsing, since labels in [DNS] normally do not contain spaces or punctuation. Nevertheless, [DNS] is able to handle such labels within sub-domains of registered domains.

Services outside the ".local." domain may have applications obtaining domain search lists provided by DHCP ([RFC2131] and [RFC3315] for IPv4 and IPv6 respectively or RA DNSSL [RFC6106] also for IPv6. Internet domains need to be published in [DNS] as A-Labels [RFC3492] because IDNA2008 compliance depends on A-label enforcement by registrars. Therefore A-Labels and not U-Labels must be published in DNS for Internet domains at this time.

The SRV scheme used by [mDNS] has also been widely adopted in the Windows OS since it offered a functional replacement for Windows Internet Name Service (WINS) as their initial attempt which lacked sufficient name hierarchy. Such common use may represent security considerations whenever these records can be automatically published.

It is unknown whether sufficient filtering of [mDNS] to expose just those services likely needed will sufficiently protect wireless networks. The extent of RBridge use and something analogous to IGMP or MLD for selective forwarding might help to mitigate otherwise spurious traffic is unknown.

3.2. Scope of Discovery

As [mDNS] is currently restricted to a single link, the scope of the advertisement is limited, by design, to the shared link between client and the device offering a service. In a multi-link scenario, the owner of the advertised service may not have a clear indication of the scope of its advertisement.

If the advertisement propagates to a larger set of links than expected, this may result in unauthorized clients (from the perspective of the owner) connecting to the advertised service. It also discloses information (about the host and service) to a larger set of potential attackers.

If the scope of the discovery is not properly setup or constrained, then information leaks will happen beyond the appropriate network which may also expose the network to various forms of attack as well.

3.3. Multiple Namespaces

There is a possibility of conflicts between local, multi-realm, and global [DNS] namespaces. Without adequate feedback, a client may not know whether the target service is the correct one, which can therefore enabling potential attacks.

A Host unable to recognize when it is in conflict with itself over multiple realms also represents a potential network stability threat.

3.4. Authorization

[DNSSEC] can assert the validity but not the veracity of records in a zone file. The trust model of the global [DNS] relies on the fact that human administrators either a) manually enter resource records into a zone file, or b) configure the [DNS] server to authenticate a trusted device (e.g., a DHCP server) that can automatically maintain such records.

An imposter may register on the local link and appear as a legitimate service. Such "rogue" services may then be automatically registered in wide area [DNS-SD].

3.5. Authentication

Up to now, the "plug-and-play" nature of [mDNS] devices have relied only on physical connectivity to the local network. If a device is visible via [mDNS], it had been assumed to be trusted. When multiple networks are involved, verifying a host is local using [mDNS] is no longer possible so other verification schemes must be used.

3.6. Privacy Considerations

Mobile devices such as smart phones that can expose the location of their owners by registering services in arbitrary zones pose a risk to privacy. Such devices must not register their services in arbitrary zones without the approval of their operators. However, it should be possible to configure one or more "safe" zones, e.g., based on subnet prefix, in which mobile devices may automatically register their services.

4. Acknowledgements

The authors wish to acknowledge valuable contributions from the following: Dave Rand, Michael Tuexen, Hosnieh Rafiee

5. References

5.1. References - Informative

5.2. References - Normative

Author's Address