Zero-configuration networking

Computer networks use numeric network addresses to identify communications endpoints in a network of participating devices. This is similar to the telephone network which assigns a string of digits to identify each telephone. In modern networking protocols, information to be transmitted is divided into a series of network packets. Every packet contains the source and destination addresses for the transmission. Network routers examine these addresses to determine the best network path in forwarding the data packet at each step toward its destination.

Similarly to telephones being labeled with their telephone number, it was a common practice in early networks to attach an address label to networked devices. The dynamic nature of modern networks, especially residential networks in which devices are powered up only when needed, desire dynamic address assignment mechanisms that do not require user involvement for initialization and management. These systems automatically give themselves common names chosen either by the equipment manufacturer, such as a brand and model number or chosen by users for identifying their equipment. The names and addresses are then automatically entered into a directory service.

Early computer networking was built upon technologies of the telecommunications networks and thus protocols tended to fall into two groups: those intended to connect local devices into a local area network (LAN), and those intended primarily for long-distance communications. The latter wide area network (WAN) systems tended to have centralized setup, where a network administrator would manually assign addresses and names. LAN systems tended to provide more automation of these tasks so that new equipment could be added to a LAN with a minimum of operator and administrator intervention.

An early example of a zero-configuration LAN system is AppleTalk, a protocol introduced by Apple Inc. for the early Macintosh computers in the 1980s. Macs, as well as other devices supporting the protocol, could be added to the network by simply plugging them in; all further configuration was automated. Network addresses were automatically selected by each device using a protocol known as AppleTalk Address Resolution Protocol (AARP), while each machine built its own local directory service using a protocol known as Name Binding Protocol (NBP). NBP included not only a name but the type of device and any additional user-provided information such as its physical location or availability. Users could look up any device on the network with the application Chooser, which filtered names based on the device type.

On Internet Protocol (IP) networks, the Domain Name System database for a network was initially maintained manually by a network administrator. Efforts to automate maintenance of this database, led to the introduction of a number of new protocols providing automated services, such as the Dynamic Host Configuration Protocol (DHCP).

Hosts on a network must be assigned IP addresses that uniquely identify them to other devices on the same network. On some networks, there is a central authority that assigns these addresses as new devices are added. Mechanisms were introduced to handle this task automatically, and both IPv4 and IPv6 now include systems for address autoconfiguration, which allows a device to determine a safe address to use through simple mechanisms. For link-local addressing, IPv4 uses the special block 169.254.0.0/16,[1] while IPv6 hosts use the prefix fe80::/10. More commonly addresses are assigned by a DHCP server, often built into common networking hardware like computer hosts or routers.

Most IPv4 hosts use link-local addressing only as a last resort when a DHCP server is unavailable. An IPv4 host otherwise uses its DHCP-assigned address for all communications, global or link-local. One reason is that IPv4 hosts are not required to support multiple addresses per interface, although many do. Another is that not every IPv4 host implements distributed name resolution (e.g., multicast DNS), so discovering the autoconfigured link-local address of another host on the network can be difficult. Discovering the DHCP-assigned address of another host requires either distributed name resolution or a unicast DNS server with this information; Some networks feature DNS servers that are automatically updated with DHCP-assigned host and address information.

IPv6 hosts are required to support multiple addresses per interface; moreover, every IPv6 host is required to configure a link-local address even when global addresses are available. IPv6 hosts may additionally self-configure additional addresses on receipt of router advertisement messages, thus eliminating the need for a DHCP server.[2]

Both IPv4 and IPv6 hosts may randomly generate the host-specific part of an autoconfigured address. IPv6 hosts generally combine a prefix of up to 64 bits with a 64-bit EUI-64 derived from the factory-assigned 48-bit IEEE MAC address. The MAC address has the advantage of being globally unique, a basic property of the EUI-64. The IPv6 protocol stack also includes duplicate address detection to avoid conflicts with other hosts. In IPv4, the method is called link-local address autoconfiguration.[1] However, Microsoft refers to this as Automatic Private IP Addressing (APIPA)[3] or Internet Protocol Automatic Configuration (IPAC). The feature is supported in Windows since at least Windows 98.[4]

Internet protocols use IP addresses for communications, but these are not easy for humans to use; IPv6 in particular uses very long strings of digits that are not easily entered manually. To address this issue, the internet has long used DNS, which allows human-readable names to be associated with IP addresses, and includes code for looking up these names from a hierarchical database system. Users type in domain names, such as example.org, which the computer's DNS software looks up in the DNS databases to retrieve an IP address, and then hands off that address to the protocol stack for further communications.[5]

Looking up an address using DNS requires the IP address of the DNS server to be known. This has normally been accomplished by typing in the address of a known server into a field in one of the devices on the network. In early systems, this was normally required on every device, but this has been pushed up one layer in the hierarchy to the DHCP servers or broadband devices like cable modems that receive this information from their internet service provider. This has reduced the user-side administration requirements and provides a key element of zero-configuration access.[5]

DNS was intended to provide uniform names to groups of devices within the same administration realm, such as example.org, provided by a name service. Assigning an address to a local device, e.g., thirdfloorprinter.example.org, normally requires administrator access to the DNS server and is often accomplished manually. Additionally, traditional DNS servers are not expected to automatically correct for changes in configuration. For instance, if a printer is moved from one floor to another it might be assigned a new IP address by the local DHCP server.[5]

To address the need for automatic configuration, Microsoft implemented NetBIOS Name Service, part of which is the Computer Browser Service already in Microsoft Windows for Workgroups 3.11[6] as early as 1992. NetBIOS Name Service is zero-configuration on networks with a single subnet and may be used in conjunction with a WINS server or a Microsoft DNS server that supports secure automatic registration of addresses. This system has small, but not zero, management overhead even on very large enterprise networks. The protocols NetBIOS can use are part of the Server Message Block (SMB) suite of open protocols[6] which are also available on Linux and iOS, although Windows typically supports a wider range of so-called dialects which can be negotiated between Windows clients that support it. For example, Computer Browser Services running on server operating systems or later versions of Windows are elected as so-called master browser over those that are not running a server operating system or running older versions of Windows.[6]

In 2000, Bill Manning and Bill Woodcock described the Multicast Domain Name Service[7] which spawned the implementations by Apple and Microsoft. Both implementations are very similar. Apple's Multicast DNS (mDNS) is published as a standards track proposal RFC 6762, while Microsoft's Link-local Multicast Name Resolution (LLMNR) is published as informational RFC 4795. LLMNR is included in every Windows version from Windows Vista onwards[8] and acts as a side-by-side alternative for Microsoft's NetBIOS Name Service over IPv4 and as a replacement over IPv6, since NetBIOS is not available over IPv6. Apple's implementation is available as the Bonjour service since 2002 in Mac OS X v10.2. The Bonjour implementation (mDNSResponder) is available under the Apache 2 Open Source License[9] and is included in Android Jelly Bean and later[10] under the same license.

Use of either NetBIOS or LLMNR services on Windows is essentially automatic, since using standard DNS client APIs will result in the use of either NetBIOS or LLMNR depending on what name is being resolved (whether the name is a local name or not), the network configuration in effect (e.g. DNS suffixes in effect) and (in corporate networks) the policies in effect (whether LLMNR or NetBIOS are disabled), although developers may opt into bypassing these services for individual address lookups.

The mDNS and LLMNR protocols have minor differences in their approach to name resolution. mDNS allows a network device to choose a domain name in the local DNS namespace and announce it using a special multicast IP address. This introduces special semantics for the domain local,[11] which is considered a problem by some members of the IETF.[12] The current LLMNR draft allows a network device to choose any domain name, which is considered a security risk by some members of the IETF.[13] mDNS is compatible with DNS-SD as described in the next section, while LLMNR is not.[14]

Name services such as mDNS, LLMNR and others do not provide information about the type of device or its status. A user looking for a nearby printer, for instance, might be hindered if the printer was given the name "Bob". Service discovery provides additional information about devices. Service discovery is sometimes combined with a name service, as in Apple's Name Binding Protocol and Microsoft's NetBIOS.

RFC 2608, the SLP standard for figuring out where to get services, was published in June 1999 by the SVRLOC IETF working group.[27]

RFC 3927, a standard for choosing addresses for networked items, was published in March 2005 by the IETF Zeroconf working group. The group included individuals from Apple, Sun, and Microsoft.[28]

LLMNR was submitted for official adoption in the IETF DNSEXT working group, however, failed to gain consensus and thus was published as informational RFC 4795 in January 2007.[29]

Following the failure of LLMNR to become an Internet standard and given that mDNS/DNS-SD is used much more widely than LLMNR, Apple was asked by the IETF to submit the mDNS/DNS-SD specs for publication as Informational RFC as well.

In February 2013 mDNS and DNS-SD were published as Standards Track Proposals RFC 6762 and RFC 6763.

Because mDNS operates under a different trust model than unicast DNS—trusting the entire network rather than a designated DNS server, it is vulnerable to spoofing attacks by any system within the same broadcast domain. Like SNMP and many other network management protocols, it can also be used by attackers to quickly gain detailed knowledge of the network and its machines.[30] Because of this, applications should still authenticate and encrypt traffic to remote hosts (e.g. via RSA, SSH, etc.) after discovering and resolving them through DNS-SD/mDNS. LLMNR suffers from similar vulnerabilities.[31]

• Bonjour Sleep Proxy
• Wireless Zero Configuration
• Peer Name Resolution Protocol (PNRP)

Notes

Sources

• JmDNS, Source forge, a pure Java implementation of mDNS/DNS-SD.
• pyZeroConf, Source forge, 11 July 2015, a pure Python implementation of mDNS/DNS-SD.
• Mono.Zeroconf, Mono project, a cross platform (Linux, MS Windows, Apple Mac), unified Mono/.NET library for Zeroconf, supporting both Bonjour and Avahi.
• WxServDisc, Source forge, 13 June 2013, a cross-platform wxWidgets-based service discovery module without external dependencies.
• Cheshire, Stuart, Multicast DNS (draft).
• Cheshire, Stuart, DNS-Based Service Discovery Specification (draft), DNS‐SD.
• Cheshire, Stuart, Zeroconf (tech talk), archived from the original (video) on 2008-03-02, retrieved 2006-03-08.
• Cheshire, Stuart, Zeroconf, including Internet drafts.
• DNS-SD, DNS based Service Discovery
• "Multicast DNS".
• Service Location Protocol, version 2. doi:10.17487/RFC2608. RFC 2608.
• Steinberg, Daniel; Cheshire, Stuart, Zero Configuration Networking: The Definitive Guide, O'Reilly.