| Internet-Draft | RCM Use Cases | January 2022 |
| Henry & Lee | Expires 4 August 2022 | [Page] |
To limit the association between a device traffic and its user, client vendors have started implementing MAC address rotation. When such rotation happens, some in-network states may break, which may affect network efficiency and the user experience. At the same time, devices may continue sending other stable identifiers, defeating the MAC rotation purposes. This document lists various network environements and a set of network services that may be affected by such rotation. This document then examines settings where the user experience may be affected by in-network state disruption, and settings where other machine identifiers may expose the user privacy. Last, this document examines solutions to maintain user privacy while preserving user quality of experience and network operation efficiency.¶
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 4 August 2022.¶
Copyright (c) 2022 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 Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
It has become easier for attackers to track the activity of a personal device, particularly when traffic is sent over a wireless link. Once the association between a device and its user is made, identifying the device and its activity is sufficient to deduce information about what the user is doing, without the user consent.¶
To reduce the risks of correlation between a device activity and its owner, multiple vendors have started to implement Randomized and Changing MAC addresses (RCM). With this scheme, an end-device implements a different RCM over time when exchanging traffic over a wireless network. By randomizing the MAC address, the association between a given traffic flow and a single device is made more difficult, assuming no other visible unique identifiers are in use.¶
However, such address change may affect the user experience and the efficiency of legitimate network operations. For a long time, the unicity of the association between a device and a MAC address was assumed, despite the emergence of tools to flush out the MAC address to bypass some network policies. When this association is broken, elements of network communication may also break. For example, sessions established between the end-device and network services may be lost and packets in translation may suddenly be without clear source or destination. As multiple clients implements fast-paced RCM rotations, network services may be over-solicited by a small number of stations that appear as many clients.¶
At the same time, some network services rely on the client station providing an identifier, which can be the MAC address or another value. If the client implements MAC rotation but continues sending the same static identifier, then the association between a stable identifier identifier and the station continues despite the RCM scheme. There may be environements where such continued association is desirable, but others where the user privacy has more value than any continuity of network service state.¶
There is a need to enumerate services that may be affected by RCM, and evaluate possible solutions to maintain both the quality of user experience and network efficiency while RCM happens and user privacy is reinforced. This document presents such assessment and recommendations.¶
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 BCP 14 RFC 2119 [RFC2119] RFC 8174 [RFC8174] when, and only when, they appear in all capitals, as shown here.¶
Any device member of an IEEE 802 network [IEEE.802-1D.1993] includes several operating layers. Among them, the Media Access Control (MAC) layer defines rules to control how the device accesses the shared medium. In a network where a machine can communicate with one or more other machines, one such rule is that each machine needs to be identified, either as the target destination of a message, or as the source of a message (and thus the target destination of the answer). Initially intended as a 48-bit (6 octets) value, later versions of the Standard [IEEE.802.15.4P_2014] allowed this address to take an extended format of 64 bits (8 octets), thus enabling a larger number of MAC addresses to coexist as the 802 technologies became widely adopted.¶
Regardless of the address length, different networks have different needs, and several bits of the first octet are reserved for specific purposes. In particular, the first bit is used to identify the destination address either as an individual (bit set to 0) or a group address (bit set to 1). The second bit, called the Universally or Locally Administered (U/L) Address Bit, indicates whether the address has been assigned by a local or universal administrator. Universally administered addresses have this bit set to 0. If this bit is set to 1, the entire address (i.e., 48 bits) has been locally administered [IEEE.802-1Y.1990] Section 5.2.1.¶
The intent of this provision is important for the present document. The 802 Standard recognized that some devices may never travel and thus, always attaching to the same network, would not need a globally unique MAC address. To accommodate for this relaxed requirement, the second bit of the MAC address first octet the MAC address format was designed to express whether the address was intended to be globally unique, or if significance was only local. The address allocation method was not defined in the Standard in this later case, but the mechanism was defined in the same clause that defined that an address should be unique so as to avoid collision.¶
It is also important to note that the purpose of the Universal version of the address was to avoid collisions and confusion, as any machine could connect to any network, and each machine needs to determine if it is the intended destination of a message or its response. The same clause 5.2.1 reminds network designers and operators that all potential members of a network need to have a unique identifier (if they are going to coexist in the network). The advantage of a universal address is that a node with such an address can be attached to any Local Area Network (LAN) in the world with an assurance that its address is unique.¶
With the rapid development of wireless technologies and mobile devices, this scenario became very common. With a vast majority of 802 networks implementing radio technologies at the access, the MAC address of a wireless device can appear anywhere on the planet and collisions should still be avoided. However, the same evolution brought the distinction between two types of devices that the 802 Standard generally referred to as 'nodes in a network'. Their definition is found in the 802E Recommended Practice (clause 6.2). One type is a shared service device, which functions are used by a number of people large enough that the device itself, its functions or its traffic cannot be associated with a single or small group of people. Examples of such devices include switches in a dense network, 802.11 (WLAN) access points in a crowded airport, task-specific (e.g. barcode scanners) devices, etc. Another type is a personal device, which is a machine, a node, primarily used by a single person or small group of people, and so that any identification of the device or its traffic can also be associated to the identification of the primary user or their traffic. Quite naturally, the unique identification of the device is trivial if the device expresses a universally unique MAC address. Then, the detection of elements directly or indirectly identifying the user of the device (Personally Identifiable Information, or PII) is sufficient to tie the universal MAC address to a user. Then, any detection of traffic that can be associated to the device becomes also associated with the known user of that device (Personally Correlated Information, or PCI).¶
This possible identification or association presents a serious privacy issue, especially with wireless technologies. For most of them, and in particular for 802.11, the source and destination MAC addresses are not encrypted even in networks that implement encryption (so that each machine can easily detect if it is the intended target of the message before attempting to decrypt its content, and also identify the transmitter, so as to use the right key when multiple unicast keys are in effect).¶
This unique identification of the user associated to a node was clearly not the intent of the 802 MAC address. A logical solution to remove this association is to use a locally administered address instead, and change the address in a fashion that prevents a temporal association between one MAC address and some PII to be maintained fruitfully. However, other network devices on the same LAN implementing a MAC layer also expect the unicity of the MAC address. When a device changes its MAC address, other devices on the same LAN may fail to recognize that the same machine is attempting to communicate with them. Additionally, multiple layers implemented at upper OSI layers have been designed with the assumption that each node on the LAN, using these services, would have a unique MAC address. This assumption sometimes adds to the PII confusion, for example in the case of Authentication, Association and Accounting (AAA) services authenticating the user of a machine and associating the authenticated user to the device MAC address. Other services solely focus on the machine (e.g. DHCP), but still expect each device to use a single MAC address. Changing the MAC address may disrupt these services.¶
The risk of service disruption is thus weighted against the privacy benefits. However, the plurality of actors involved in the exchanges tends to blurry the boundaries of what privacy should be protected against. It might therefore be useful to list the actors to the network exchanges. Some actors are functional entities, some others are humans (or related) entities.¶
Wireless access network infrastructure devices (e.g. WLAN access points or controllers): these devices participate in 802 LAN operations. As such, they need to uniquely identify machines as a source or destination so as to successfully continue exchanging frames. Part of the identification includes recording, and adapting to, devices communication capabilities (e.g. support for specific protocols). As a device changes its network attachment (roams) from one access point to another, the access points can exchange contextual information (e.g. device MAC, keying material) allowing the device session to continue seamlessly. These access points can also inform devices further in the wired network about the roam, to ensure that OSI model Layer 2 frames are redirected to the new device access point.¶
Other network devices operating at the MAC layer: many wireless network access devices (e.g., 802.11 access points) are conceived as Layer 2 devices, and as such they bridge a frame from one medium (e.g., 802.11 or Wi-Fi) to another (e.g., 802.3 or Ethernet). This means that a wireless device MAC address often exists on the wire beyond the wireless access device. Devices connected to this wire also implement 802 technologies, and as such operate on the expectation that each device is associated to a unique MAC address for the duration of continuous exchanges. For example, switches and bridges associate MAC addresses to individual ports (so as to know which port to send a frame intended for a particular MAC address). Similarly, authentication, authorization and accounting (AAA) services can validate the identity of a device and use the device MAC address as a first pointer to the device identity (before operating further verification). Similarly, some networking devices offer Layer-2 filtering policies that may rely on the connected MAC addresses. 802.1X-enabled devices may also selectively block the data portion of a port until a connecting device is authenticated. These services then use the MAC address as a first pointer to the device identity to allow or block data traffic. This list is not exhaustive. Multiple services are defined for 802.3 networks, and multiple services defined by the IEEE 802.1 working group are also applicable to 802.3 networks. Wireless access points may also connect to other mediums than 802.3, which also implements mechanism under the umbrella of the general 802 Standard, and therefore expect the unique association of a MAC address to a device.¶
Network devices operating at upper layers: some network devices provide functions and services above the MAC layer. Some of them also operate a MAC layer function: for example, routers provide IP forwarding services, but rely on the device MAC address to create the appropriate frame structure. Other devices and services operate at upper layers, but also rely upon the 802 principle of unique MAC-to-device mapping. For example, DHCPv4 services commonly provide a single IP address per MAC address (they do not assign more than one IPv4 address per MAC address, and assign a new IPv4 address to each new requesting MAC address). ARP and reverse-ARP services commonly expect that, once an IP-to-MAC mapping has been established, this mapping is valid and unlikely to change for the cache lifetime. DHCPv6 services commonly do not assign the same IPv6 address to two different requesting MAC addresses. Hybrid services, such as EoIP, also assume stability of the device-to-MAC-and-IP mapping for the duration of a given session.¶
Over the air (OTA) observers: as the transmitting or receiving MAC address is usually not encrypted in wireless 802-technologies exchanges, and as any protocol-compatible device in range of the signal can read the frame header, OTA observers are able to read individual transmissions MAC addresses. Some wireless technologies also support techniques to establish distances or positions, allowing the observer, in some cases, to uniquely associate the MAC address to a physical device and it associated location. It can happen that an OTA observer has a legitimate reason to monitor a particular device, for example for IT support operations. However, it is difficult to control if another actor also monitors the same station with the goal of obtaining PII or PCI.¶
Wireless access network operators: some wireless access networks are only offered to users or devices matching specific requirements, such as device type (e.g., IoT-only networks, factory operational networks). Therefore, operators can attempt to identify the devices (or the users) connecting to the networks under their care. They can use the MAC address to represent an identified device.¶
Network access providers: wireless access networks are often considered beyond the first 2 layers of the OSI model. For example, several regulatory or legislative bodies can group all OSI layers into their functional effect of allowing network communication between machines. In this context, entities operating access networks can see their liability associated to the activity of devices communicating through the networks that these entities operate. In other contexts, operators assign network resources based on contractual conditions (e.g., fee, bandwidth fair share). In these scenarios, these operators may attempt to identify the devices and the users of their networks. They can use the MAC address to represent an identified device.¶
Over the wire internal (OTWi) observers: because the device wireless MAC address continues to be present over the wire if the infrastructure connection device (e.g. access point) functions as a Layer 2 bridge, observers may be positioned over the wire and read transmission MAC addresses. Such capability supposes that the observer has access to the wired segment of the broadcast domain where the frames are exchanged. In most networks, such capability requires physical access to an infrastructure wired device in the broadcast domain (e.g. switch closet), and is therefore not accessible to all.¶
Over the wired external (OTWe) observers: beyond the broadcast domain, frames headers are removed by a routing device, and a new Layer 2 header is added before the frame is transmitted to the next segment. The personal device MAC address is not visible anymore, unless a mechanism copies the MAC address into a field that can be read while the packet travels onto the next segment (e.g. pre- [RFC4941] and pre- [RFC7217] IPv6 addresses built from the MAC address). Therefore, unless this last condition exists, OTWe observers are not able to see the device MAC address.¶
The surface of PII exposures that can drive MAC address randomization depends on the environment where the device operates, on the presence and nature of other devices in the environment, and on the type of network the device is communicating through. Therefore, a device can express an identity (such as a MAC address) that can be stable over time if trust with the environment is established, or that can be temporal if an identity is required for a service in an environment where trust has not been established. Trust is not a binary currency. Thus it is useful to distinguish what trust a personal device may establish with the different entities at play in a L2 domain:¶
This trust relationship naturally depends on the relationship between the user of the personal device and the operator of the service. Thus, it is useful to observe the typical trust structure of common environments:¶
Many network functional devices offering a service to a personal device use the device MAC address to maintain service continuity.¶
Wireless access points and controllers use the MAC address to validate the device connection context, including protocol capabilities, confirmation that authentication was completed, QoS or security profiles, encryption key material. Some advanced access points and controllers also include upper layer functions which purpose is covered below. A device changing its MAC address, without another recorded device identity, would cause the access point and the controller to lose these parameters. As such, the Layer 2 infrastructure does not know that the device (with its new MAC address) is authorized to communicate through the network. The encryption keying material is not identified anymore (causing the access point to fail decrypting the device traffic, and fail selecting the right key to send encrypted traffic to the device). In short, the entire context needs to be rebuilt, and a new session restarted. The time consumed by this procedure breaks any flow that needs continuity or short delay between packets on the device (e.g. real-time audio, video, AR/VR etc.) The 802.11i Standard recognizes that a device may leave the network and come back after a short time window. As such, the standard suggests that the infrastructure should keep the context for a device up to one hour after the device was last seen. MAC address rotation in this context can cause resource exhaustion on the wireless infrastructure and the flush of contexts, including for devices that are simply in temporal sleep mode.¶
Other devices in the Layer 2 broadcast domain also use the MAC address to know whether and where to forward frames. MAC rotation can cause these devices to exhaust their resources, holding in memory traffic for a device which port location can no longer be found. As these infrastructure devices also implement a cache (to remember the port position of each known device), too frequent MAC rotation can cause resources exhaustion and the flush of older MAC addresses, including for devices that did not rotate their MAC. For the RCM device, these effects translate into session discontinuity and return traffic losses.¶
In wireless contexts, 802.1X authenticators rely on the device and user identity validation provided by a AAA server to open their port to data transmission. The MAC address is used to verify that the device is in the authorized list, and the associated key used to decrypt the device traffic. A change in MAC address causes the port to be closed to the device data traffic until the AAA server confirms the validity of the new MAC address. Therefore, MAC rotation can interrupt the device traffic, and cause a strain on the AAA server.¶
DHCP servers, without a unique identification of the device, lose track of which IP address is validly assigned. Unless the RCM device releases the IP address before the rotation occurs, DHCP servers are at risk of scope exhaustion, causing new devices (and RCM devices) to fail to obtain a new IP address. Even if the RCM device releases the IP address before the rotation occurs, the DHCP server typically holds the released IP address for a certain duration, in case the leaving MAC would return. As the DHCP server cannot know if the release is due to a temporal disconnection or a MAC rotation, the risk of scope address exhaustion exists even in cases where the IP address is released.¶
Routers keep track of which MAC address is on which interface. MAC rotation can cause MAC address cache exhaustion, but also the need for frequent ARP and inverse ARP exchanges.¶
In residential settings (environments type A), policies can be in place to control the traffic of some devices (e.g. parental control, block-list devices). These policies are often based on the device MAC address. Rotation of the MAC address removes the possibility for such control.¶
In residential settings (environments type A) and in enterprises (environments types D and E), device recognition and ranging may be used for IoT-related functionalities (door unlock, preferred light and temperature configuration, etc.) These functions often rely on the detection of the device wireless MAC address. MAC address rotation breaks the services based on such model.¶
In managed residential settings (environments types B) and in enterprises (environments types D and E), the network operator is often requested to provide IT support. With MAC address rotation, real time support is only possible if the user is able to provide the current MAC address. Service improvement support is not possible if the MAC address that the device had at the (past) time of the reported issue is not known at the time the issue is reported.¶
Section 3.4discusses different environments, different settings, and the expectations of users and network operators. Table 1 summarizes the expected degree of trust, network admin responsibility, complexity of supported network services and network support expectation from the user¶
| Network Location | Trust Degree | Network Admin | Network Services | Network Support Expectation |
|---|---|---|---|---|
| Home | High | User | Medium | Low |
| Campus (BYOD) | Medium | IT | Complex | Medium |
| Enterprise (MDM) | High | IT | Complex | High |
| Hospitality | Low | IT | Simple | Medium |
| Public WiFi | Low | ISP | Simple | Low |
For example: a Home network is considered to be trusted and safe. Users expect a simple procedure to connect to their home network. All devices in the home network trust each other. Privacy within the Layer 2 domain is not a major concern. The Home network can also include many IoT devices, which need to be simple to onboard and manage. The home user commonly expects the network operator to protect the home network from external threats (attacks from the Internet). The home user also commonly expects simple policy features (e.g., Parental Control). Most home users do not expect to need networking skills to manage their home network.¶
On the other end of the spectrum, Public Wi-Fi is often considered to be untrusted. Privacy is the number one concern for the user. Most users connect to Public Wi-Fi only require imple Internet connectivity service, and expect only limited to no technical support.¶
The section describes the requirements for Randomized MAC-Address Changes:¶
This memo includes no request to IANA.¶
Privacy considerations are discussed throughout this document.¶