Internet-Draft | Multicast Transport Security | August 2021 |
Rose & Holland | Expires 7 February 2022 | [Page] |
Interdomain multicast has unique potential to solve delivery scalability for popular content, but it carries a set of security and privacy issues that differ from those in unicast delivery. This document analyzes the security threats unique to multicast-based delivery for Internet and Web traffic under the Internet and Web threat models.¶
This note is to be removed before publishing as an RFC.¶
Discussion of this document takes place on the mailing list (), which is archived at .¶
Source for this draft and an issue tracker can be found at https://github.com/squarooticus/draft-krose-multicast-security.¶
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 7 February 2022.¶
Copyright (c) 2021 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.¶
This document examines the security considerations relevant to the use of multicast for scalable one-to-many delivery of application traffic over the Internet, along with special considerations for multicast delivery to clients constrained by the Web security model.¶
This document assumes readers have a basic understanding of some background topics, specifically:¶
The Web security model, while not yet documented authoritatively in a single reference, nevertheless strongly influences Web client implementations, and has generally been interpreted to require certain properties of underlying transports such as:¶
For unicast transport, TLS [RFC8446] satisfies these requirements, therefore Web Transport [webtrans] proposes to require qualifying transport protocols to use "TLS or a semantically equivalent security protocol".¶
For unicast communication this is sensible and meaningful (if imprecise) for an engineer with a grounding in security, but it is unclear how or whether 'semantic equivalence to TLS' can be directly interpreted in any meaningful way for multicast transport protocols. This document instead explicitly describes a security and privacy threat model for multicast transports in order to extend the Web security model to accommodate multicast delivery in a way that fits within the spirit of how that model is generally interpreted for unicast.¶
Although defining the security protections necessary to make multicast traffic suitable for Web Transport is a key goal for this document, many of the security considerations described here would be equally necessary to consider if a higher level multicast transport protocol were to be made available via a different interface within clients constrained by the Web security model.¶
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 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶
Fundamentally, multicast is simply an addressing scheme in which the destination address identifies more than one unique receiver; that said, this has implications for protocol design that differ greatly from those for unicast addressing.¶
Given the virtually unbounded potential for attacks targeting data confidentiality and user privacy, we attempt to make the description of a multicast threat model tractable by taking the approach of highlighting areas in which multicast differs from unicast or poses novel challenges that are not addressed at a layer unconcerned with the addressing scheme.¶
Unlike typical unicast transport protocols, multicast transports are naturally unidirectional. Use cases for multicast transports typically involve one or a small number of senders transmitting data to a large number of receivers. The sender may not know who the receivers are, or even how many of them there are, although a sender may require a pre-existing out-of-band relationship with receivers for the received data to be useful, such as via distribution of decryption keys only to authorized receivers.¶
Applications built atop multicast IP or UDP must provide a mechanism for congestion control, just as those built atop unicast IP or UDP must. Although multicast applications compliant with Section 4.1 of [RFC8085] will implement congestion control, in the context of a threat model it is important to note that malicious clients might attempt to use non-compliant subscriptions to multicast traffic as part of a DoS attack where possible, and that some applications might not be compliant with the recommendations for congestion control implementations.¶
IP and UDP provide no native reliability mechanism, whether for unicast or multicast transmission. Protocols leveraging multicast may add mechanisms for reliable delivery (see [RFC5740], [RFC5775], and [quic-http-mcast] for examples), but this may expand the attack surface against content providers if per-packet authenticity is not provided. For example, in an application with unicast recovery for objects constructed out of multiple packets and which is limited to object-level authentication, if a packet is injected into the multicast stream receivers will fail to authenticate an entire object, necessitating unicast recovery by every receiver for the entire object. Care must be taken to avoid such amplification attack vectors.¶
The Web security model requires that content be authenticated cryptographically. In the context of unicast transport security, authentication means that content is known to have originated from the trusted peer, something that is typically enforced via a cryptographic authentication tag:¶
In both cases:¶
Instead of directly authenticating the protected content, the tag may protect a root of trust that itself protects cryptographically-linked content. Examples include:¶
Asymmetric verification of content delivered through multicast is conceptually identical to the unicast case, owing to the asymmetry of access to the signing key; but the symmetric case does not directly apply given that multiple receivers need access to the same key used for both signing and verification, which in a naive implementation opens up the possibility of forgery by a receiver on-path or with the ability to spoof the source.¶
Multiple mechanisms providing for reliable asymmetric authentication of data delivered by multicast have been proposed over the years.¶
Regardless of mechanism, the primary goal of authentication in the multicast context is identical to that for unicast: that the content delivered to the application originated from the trusted source. Semantic equivalence to (D)TLS in this respect is therefore straightforwardly achieved by any number of potential mechanisms.¶
Integrity in the Web security model for unicast is closely tied to the features provided by transports that enabled the Web from its earliest days. TCP, the transport substrate for the original HTTP, provides in-order delivery, reliability via retransmission, packet de-duplication, and modest protection against replay and forgery by certain classes of adversaries. SSL and TLS later greatly strengthened those protections. Web applications universally rely on these integrity assumptions for even the most basic operations. It is no surprise, then, that when QUIC was subsequently designed with HTTP as the model application, initial requirements included the integrity guarantees provided by TCP at the granularity of an individual stream.¶
Multicast applications by contrast have different integrity assumptions owing to the multicast transport legacy. UDP, the transport protocol atop which multicast applications are typically built, provides no native reliability, in-order delivery, de-duplication, or protection against replay or forgery. Additionally, UDP by itself provides no protection against off-path spoofing or injection. Multicast has therefore traditionally been used for applications that can deal with a modest loss of integrity through application-layer mitigations such as:¶
A baseline for multicast transport integrity that makes sense within the Web security model requires that we first define the minimally acceptable integrity requirements for data that may be presented to a user or otherwise input to the browser's trusted computing base. We propose that the proper minimal standard given the variety of potential use cases, including many that have no need for reliable or in-order delivery, is to require protection against replay, injection, and modification and the ability to detect deletion, loss, or reordering. This standard will necessarily constrain conformant application-layer protocol design, just as the Web security model adds constraints to vanilla TCP.¶
Integrity in multicast, as in the unicast case, is partially provided by the authentication mechanism: for example, if authentication is provided at packet granularity, modified or forged packets will fail to authenticate and will thus not be delivered to the application. Lacking a bidirectional relationship at the transport layer, however, applications relying on multicast must otherwise provide for detection of and/or recovery from packet duplication/replay, loss/deletion, and reordering. Some of these functions, too, may be provided by the authentication layer. For instance:¶
In the unicast transport security context, confidentiality implies that an observer (passive or active) without pre-existing access to keying material must not be able to decrypt the bytes on the wire or identify the content being transferred, even if that adversary has access to the decrypted content via other means. In practice, the former is trivially achieved through the use of authenticated key exchange and modern symmetric ciphers, but the latter is an ideal that is rarely possible owing to the substantial metadata in the clear on the public Internet: traffic analysis can make use of packet sizes and timing, endpoint identities, biases in application-layer protocol designs, side channels, and other such metadata to reveal an often surprising amount of information about the encrypted payload without needing access to any keying material. (Conceptually, one could make many streams appear identical to a passive observer: video streams, for example, could be bucketed into a small number of bitrates with identical packet sizes and pacing via padding of the actual content. This would increase overhead for servers and networks, primarily in terms of bandwidth utilization, that may be operationally unacceptable.)¶
Multicast additionally introduces the complication that all receivers of a stream, even if such a stream is encrypted, receive the same payload (loss and duplication notwithstanding). This introduces novel privacy concerns that do not apply to unicast transports.¶
In contrast to (say) unicast TLS, on-path monitoring can trivially prove that identical content was delivered to multiple receivers, irrespective of payload encryption. Furthermore, since those receivers all require the same keying material to decrypt the received payload, a compromise of any single receiver's device exposes decryption keys, and therefore the plaintext content, to the attacker.¶
That having been said, however, there are factors and practices that help mitigate these additional risks:¶
A sender has responsibility not to expose personal information broadly. This is not a consideration unique to multicast delivery: an irresponsible service could publish a web page with Social Security numbers or push its server TLS private key into the certificate transparency log as easily as it could multicast personal data to a large set of receivers.¶
The Web security model partially mitigates negligence on the part of senders by mandating the use of secure transports: prohibiting the fetching of mixed content on a single page prevents a server from sending private data to a browser in the clear. The main effect is to raise the bar closer to requiring bad faith or willful irresponsibility on the part of senders in revealing personal information.¶
Multicast by its very nature is not generally suitable for transport of personal data: since the main value of leveraging a multicast transport is to deliver the same data to a large pool of receivers, such content must not include confidential personal information. Senders already have a responsibility to handle private information in a way that respects the privacy of users: the availability of multicast transports does not further complicate this responsibility.¶
Forward secrecy (also called "perfect forward secrecy" or "PFS" and defined in [RFC4949]) is a countermeasure against attackers that record encrypted traffic with the intent of later decrypting it should the communicating parties' long-term keys be compromised. Forward secrecy for protocols leveraging time-limited keys to protect a communication session ("session keys") requires that such session keys be unrecoverable by an attacker that later compromises the long-term keys used to negotiate or deliver those session keys.¶
As noted earlier, confidential content delivered via multicast will necessarily imply delivery of the same keying material to multiple receivers, rather than negotiation of a unique key as is typical in the unicast case. Presumably, such receivers will need to be individually authenticated and authorized by the content provider prior to delivery of decryption keys. If this authorization and key delivery mechanism employs a forward secret unicast transport such as TLS 1.3, then so long as these encryption keys are ephemeral (that is, rotated periodically and discarded after rotation) the multicast payloads will also effectively be forward secret beyond the time interval of rotation, which we can consider to be the session duration.¶
Protocols should be designed to discourage implementations from making use of unauthenticated data. The usual approach to enforcing this is to entangle decryption and authentication where possible, for example via use of primitives such as authenticated encryption. While ultimately authentication checks are independent of decryption (at least in classical cryptography), use of such primitives to minimize the number of places in which an incomplete or lazy implementation can avoid such checks constitutes best practice. TLS 1.3, for instance, mandates AE for all symmetric cryptographic operations: without writing one's own AE cipher implementation that purposely skips the authentication tag check, this leaves establishment of trust in the peer certificate as the only practical step an implementation can skip without impacting the ability to make use of the decrypted content.¶
The situation in multicast is complicated by the need for more than two parties to have access to symmetric keys that would used to secure payloads via AE in the unicast case. As discussed in Section 3.2, it is imperative for protocols to provide, and for receivers to leverage, some kind of asymmetry in authentication of each content unit prior to any use of said content to eliminate the ability for an attacker in possession of a shared symmetric key (possibly including an authorized receiver) to inject forged data into a stream that other receivers would then validate and deliver to applications. This requirement to perform authentication checks throughout the lifetime of a stream that are separate from, and orthogonal to, content decryption adds an extra dimension of risk from implementation incorrectness, because such authentication becomes an on-going process rather than the result of a one-time certificate check at connection establishment. Protocol designers and implementors are thus strongly encouraged to simplify or even black box such on-going authentication to minimize the potential for implementors or users to skip such checks.¶
Concern about pervasive monitoring of users culminated in the publication of [RFC7258], which states that "the IETF will work to mitigate the technical aspects of [pervasive monitoring]." One area of particular concern is the ability for pervasive monitoring to track individual clients across changes in network connectivity, such as being able to tell when a device or connection migrates from a wired home network to a cell network. This has motivated mitigations in subsequent protocol designs, such as those discussed in section 9.5 of [RFC9000]. Migration of multicast group subscriptions across network connections carries the potential for correlation of metadata between multicast group subscriptions and unicast control channels, even when control channels are encrypted, so care must be taken to design protocols to avoid such correlations.¶
The security requirements for multicast transport to a browser follow directly from the requirement that the browser's job is to protect the user. Huang et al. [huang-w2sp] summarize the core browser security guarantee as follows:¶
The reader will find the full discussion of the browser threat model in section 3 of [RFC8826] helpful in understanding what follows.¶
This document covers only unidirectional multicast from a server to (potentially many) clients, as well as associated control channels used to manage that communication and access to the content delivered via multicast. As a result, local resource access can be presumed to be limited to that already available within web applications. Note that these resources may include fingerprint information that can be used to identify or track individuals, such as information about the user agent, viewport size, display resolution, a concern covered in extensive detail in [RFC8942].¶
In the absence of any specific mitigations, network attackers have the ability to inject or modify packets in a multicast stream. On-path injection and modification are trivial, but even off-path injection is feasible for many networks, such as those that implement no protections against source address spoofing. Consequently, it is critical that a browser prevent any such injected or modified traffic from reaching large attack surfaces in the browser, such as the rendering code.¶
A hostile origin could serve a Web application that attempts to join many multicast channels, overwhelming the provider's network with undesired traffic.¶
The first line of defense is the browser itself: the browser should at a minimum prevent joining of channels not associated with the hosting site. In the general case, this implies the need for a CORS-like mechanism for cross-origin authorization of multicast channel sharing.¶
The second line of defense is the network. The user's upstream router can and should monitor the user's multicast behavior, implementing circuit breakers that will target unpopular content when overloaded or when an abusive subscription pattern is detected.¶
Browsers that offer a private browsing mode, designed both to bypass access to client-side persistent state and to prevent broad classes of data leakage that can be leveraged by passive and active attackers alike, should require explicit user approval for joining a multicast group given the metadata exposure to network elements of IGMP and MLD messages.¶
This entire document is about security.¶
This document has no IANA actions.¶
TODO acks¶