| ACE Working Group | S. Gerdes |
| Internet-Draft | Universität Bremen TZI |
| Intended status: Informational | L. Seitz |
| Expires: April 25, 2019 | RISE SICS |
| G. Selander | |
| Ericsson AB | |
| C. Bormann, Ed. | |
| Universität Bremen TZI | |
| October 22, 2018 |
An architecture for authorization in constrained environments
draft-ietf-ace-actors-07
Constrained-node networks are networks where some nodes have severe constraints on code size, state memory, processing capabilities, user interface, power and communication bandwidth (RFC 7228).
This document provides terminology, and identifies the elements that an architecture needs to address, providing a problem statement, for authentication and authorization in these networks.
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 April 25, 2019.
Copyright (c) 2018 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.
As described in [RFC7228], constrained nodes are small devices with limited abilities which in many cases are made to fulfill a specific simple task. They may have limited hardware resources such as processing power, memory, non-volatile storage and transmission capacity and additionally in most cases do not have user interfaces and displays. Due to these constraints, commonly used security protocols are not always easily applicable, or may give rise to particular deployment/management challenges.
As components of the Internet of Things (IoT), constrained nodes are expected to be integrated in all aspects of everyday life and thus will be entrusted with vast amounts of data. Without appropriate security mechanisms attackers might gain control over things relevant to our lives. Authentication and authorization mechanisms are therefore prerequisites for a secure Internet of Things.
Applications generally require some degree of authentication and authorization, which gives rise to some complexity. Authorization is about who can do what to which objects (see also [RFC4949]). Authentication specifically addresses the who, but is often specific to the authorization that is required (for example, it may be sufficient to authenticate the age of an actor, so no identifier is needed or even desired). Authentication often involves credentials, only some of which need to be long-lived and generic; others may be directed towards specific authorizations (but still possibly long-lived). Authorization then makes use of these credentials, as well as other information (such as the time of day). This means that the complexity of authenticated authorization can often be moved back and forth between these two aspects.
In some cases authentication and authorization can be addressed by static configuration provisioned during manufacturing or deployment by means of fixed trust anchors and static access control lists. This is particularly applicable to siloed, fixed-purpose deployments.
However, as the need for flexible access to assets already deployed increases, the legitimate set of authorized entities as well as their specific privileges cannot be conclusively defined during deployment, without any need for change during the lifetime of the device. Moreover, several use cases illustrate the need for fine-grained access control policies, for which for instance a basic access control list concept may not be sufficiently powerful [RFC7744].
The limitations of the constrained nodes impose a need for security mechanisms which take the special characteristics of constrained environments into account; not all constituents may be able to perform all necessary tasks by themselves. To put it the other way round: the security mechanisms that protect constrained nodes must remain effective and manageable despite the limitations imposed by the constrained environment.
Therefore, in order to be able to achieve complex security objectives between actors some of which are hosted on simple (“constrained”) devices, some of the actors will make use of help from other, less constrained actors. (This offloading is not specific to networks with constrained nodes, but their constrainedness as the main motivation is.)
We therefore group the logical functional entities by whether they can be assigned to a constrained device (“constrained level”) or need higher function platforms (“less-constrained level”); the latter does not necessarily mean high-function, “server” or “cloud” platforms. Note that assigning a logical functional entity to the constrained level does not mean that the specific implementation needs to be constrained, only that it can be.
The description assumes that some form of setup (aspects of which are often called provisioning and/or commissioning) has already been performed and at least some initial security relationships important for making the system operational have already been established.
This document provides some terminology, and identifies the elements an architecture needs to address, representing the relationships between the logical functional entities involved; on this basis, a problem description for authentication and authorization in constrained-node networks is provided.
Readers are assumed to be familiar with the terms and concepts defined in [RFC4949], including “authentication”, “authorization”, “confidentiality”, “(data) integrity”, “message authentication code”, and “verify”.
REST terms including “resource”, “representation”, etc. are to be understood as used in HTTP [RFC7231] and CoAP [RFC7252]; the latter also defines additional terms such as “endpoint”.
Terminology for constrained environments including “constrained device”, “constrained-node network”, “class 1”, etc. is defined in [RFC7228].
In addition, this document uses the following terminology:
Note that other authorization architectures such as OAuth [RFC6749] or UMA [I-D.hardjono-oauth-umacore] focus on the authorization problems on the RS side, in particular what accesses to resources the RS is to allow. In this document the term authorization includes this aspect, but is also used for the client-side aspect of authorization, i.e., more generally allowing RqPs to decide what interactions clients may perform with other endpoints.
This document deals with how to control and protect resource-based interaction between potentially constrained endpoints. The following setting is assumed as a high-level problem statement:
In its simplest expression, the architecture starts with a two-layer model: the principal level (at which components are assumed to be functionally unconstrained) and the constrained level (at which some functional constraints are assumed to apply to the components).
Without loss of generality, we focus on the C functionality in one endpoint, which we therefore also call C, accessing the RS functionality in another endpoint, which we therefore also call RS.
The constrained level and its security objectives are detailed in Section 5.1.
-------------- --------------
| ------- | | ------- |
| | C | ------ requests resource -----> | RS | |
| ------- <----- provides resource ------ ------- |
| Endpoint | | Endpoint |
-------------- --------------
Figure 1: Constrained Level
The authorization decisions at the endpoints are made on behalf of the overseeing principals that control the endpoints. To reuse OAuth and UMA terminology, the present document calls the overseeing principal that is controlling C the Requesting Party (RqP), and calls the overseeing principal that is controlling RS the Resource Owner (RO). Each overseeing principal makes authorization decisions (possibly encapsulating them into security policies) which are then enforced by the endpoint it controls.
The specific security objectives will vary, but for any specific version of this scenario will include one or more of:
Objectives of type 1 require performing authorization on the Resource Server side while objectives of type 2 require performing authorization on the Client side.
More on the security objectives of the principal level in Section 5.2.
------- -------
| RqP | | RO | Principal Level
------- -------
| |
in charge of in charge of
| |
V V
------- -------
| C | -- requests resource --> | RS | Constrained Level
------- <-- provides resource-- -------
Figure 2: Constrained Level and Principal Level
The use cases defined in [RFC7744] demonstrate that constrained devices are often used for scenarios where their overseeing principals are not present at the time of the communication, are not able to communicate directly with the device because of a lack of user interfaces or displays, or may prefer the device to communicate autonomously.
Moreover, constrained endpoints may need support with tasks requiring heavy processing, large memory or storage, or interfacing to humans, such as management of security policies defined by an overseeing principal. The principal, in turn, requires some agent maintaining the policies governing how its endpoints will interact.
For these reasons, another level of nodes is introduced in the architecture, the less-constrained level (illustrated below in Figure 3). Using OAuth terminology, AS acts on behalf of the RO to control and support the RS in handling access requests, employing a pre-existing security relationship with RS. We complement this with CAS acting on behalf of RqP to control and support the C in making resource requests and acting on the responses received, employing a pre-existing security relationship with C. To further relieve the constrained level, authorization (and related authentication) mechanisms may be employed between CAS and AS (Section 6.2). (Again, both CAS and AS are conceptual entities controlled by their respective overseeing principals. Many of these entities, often acting for different overseeing principals, can be combined into a single server implementation; this of course requires proper segregation of the control information provided by each overseeing principal.)
------- -------
| RqP | | RO | Principal Level
------- -------
| |
controls controls
| |
V V
-------- -------
| CAS | <- AuthN and AuthZ -> | AS | Less-Constrained Level
-------- -------
| |
controls and supports controls and supports
authentication authentication
and authorization and authorization
| |
V V
------- -------
| C | -- requests resource --> | RS | Constrained Level
------- <-- provides resource-- -------
Figure 3: Overall architecture
Figure 3 shows all three levels considered in this document. Note that the vertical arrows point down to illustrate exerting control and providing support; this is complemented by information flows that often are bidirectional. Note also that not all entities need to be ready to communicate at any point in time; for instance, RqP may have provided enough information to CAS that CAS can autonomously negotiate access to RS with AS for C based on this information.
The elements of the architecture described above are indeed architectural; that is, they are parts of a conceptual model, and may be instantiated in various ways in practice. For example, in a given scenario, several elements might share a single device or even be combined in a single piece of software. If C is located on a more powerful device, it can be combined with CAS:
------- --------
| RqP | | RO | Principal Level
------- --------
| |
in charge of in charge of
| |
V V
------------ --------
| CAS + C | <- AuthN and AuthZ -> | AS | Less-Constrained Level
------------ --------
^ |
\__ |
\___ authentication
\___ and authorization
requests resource/ \___ support
provides resource \___ |
\___ |
V V
-------
| RS | Constrained Level
-------
Figure 4: Combined C and CAS
If RS is located on a more powerful device, it can be combined with AS:
------- -------
| RqP | | RO | Principal Level
------- -------
| |
in charge of in charge of
| |
V V
---------- -----------
| CAS | <- AuthN and AuthZ -> | RS + AS | Less-Constrained Level
---------- -----------
| ^
authentication ___/
and authorization ___/
support ___/ request resource / provides resource
| ___/
V ___/
------- /
| C | <-
-------
Figure 5: Combined AS and RS
If C and RS have the same overseeing principal, CAS and AS can be combined.
------------
| RqP = RO | Principal Level
------------
|
in charge of
|
V
--------------
| CAS + AS | Less-Constrained Level
--------------
/ \
/ \
authentication authentication
and authorization and authorization
support support
/ \
V V
------- -------
| C | -- requests resource --> | RS | Constrained Level
------- <-- provides resource -- -------
Figure 6: CAS combined with AS
We now formulate the problem statement in terms of the information flows the architecture focuses on. (While the previous section discusses the architecture in terms of abstract devices and their varying roles, the actual protocols being standardized define those information flows and the messages embodying them: “RESTful architectures focus on defining interfaces and not components” ([REST], p. 116).)
The interaction with the nodes on the principal level, RO and RqP, is not involving constrained nodes and therefore can employ an existing mechanism. The less-constrained nodes, CAS and AS, support the constrained nodes, C and RS, with control information, for example permissions of clients, conditions on resources, attributes of client and resource servers, keys and credentials. This control information may be rather different for C and RS.
The potential information flows are shown in Figure 7. The direction of the vertical arrows expresses the exertion of control; actual information flow is bidirectional.
The message flow may pass unprotected paths and thus need to be protected, potentially beyond a single REST hop (Section 3.1):
------- -------
| CAS | | AS |
------- -------
a ^ | b a = requests for control info a ^ | b
| | b = control information | |
| v | v
------- -------
| C | ------ request -------------------> | RS |
| | <----- response ------------------- | |
------- -------
Figure 7: Information flows that need to be protected
In terms of the elements of the architecture laid out above, this document’s problem statement for authorization in constrained environments can then be summarized as follows:
(Note that other aspects relevant to secure constrained node communication such as secure bootstrap or group communication are not specifically addressed by the present document.)
The security objectives that are addressed by an authorization solution include confidentiality and integrity. Additionally, an authorization solution has an impact on the availability: First, by reducing the load (only accepting selected operations by selected entities limits the burden on system resources), and second, because misconfigured or wrongly designed authorization solutions can result in availability breaches (denial of service) as users might no longer be able to use data and services as they are supposed to.
Authentication mechanisms can help achieve additional security objectives such as accountability and third-party verifiability. These additional objectives are not directly related to authorization and thus are not in scope of this draft, but may nevertheless be relevant. Accountability and third-party verifiability may require authentication on a device level, if it is necessary to determine which device performed an action. In other cases it may be more important to find out who is responsible for the device’s actions. (The ensuing requirements for logging, auditability, and the related integrity requirements are very relevant for constrained devices as well, but outside the scope of this document.) See also Section 4 for more discussion about authentication and authorization.
The security objectives and their relative importance differ for the various constrained environment applications and use cases [RFC7744].
The architecture is based on the observation that different parties may have different security objectives. There may also be a “collaborative” dimension: to achieve a security objective of one party, another party may be required to provide a service. For example, if RqP requires the integrity of representations of a resource R that RS is hosting, both C and RS need to partake in integrity-protecting the transmitted data. Moreover, RS needs to protect any write access to this resource as well as to relevant other resources (such as configuration information, firmware update resources) to prevent unauthorized users from manipulating R.
In many cases, the information flows described in Section 2.3 cross multiple client-server pairings but still need to be protected end-to-end. For example, AS may not be connected to RS (or may not want to exercise such a connection), relying on C for transferring authorization information. As the authorization information is related to the permissions granted to C, C must not be in a position to manipulate this information, which therefore requires integrity protection on the way between AS and RS.
As another example, resource representations sent between endpoints may be stored in intermediary nodes, such as caching proxies or pub-sub brokers. Where these intermediaries cannot be relied on to fulfill the security objectives of the endpoints, it is the endpoints that will need to protect the exchanges beyond a single client-server exchange.
Note that there may also be cases of intermediary nodes that very much partake in the security objectives to be achieved. The question what are the pairs of endpoints between which the communication needs end-to-end protection (and which aspect of protection) is defined by the specific use case. Two examples of intermediary nodes executing security functionality:
In order to support the required communication and application security, keying material needs to be established between the relevant nodes in the architecture.
Server-side authorization solutions aim at protecting the access to items of interest, for instance hardware or software resources or data: They enable the resource owner to control who can access it and how.
To determine if an entity is authorized to access a resource, an authentication mechanism is needed. According to the Internet Security Glossary [RFC4949], authentication is “the process of verifying a claim that a system entity or system resource has a certain attribute value.” Examples for attribute values are the ID of a device, the type of the device or the name of its owner.
The security objectives the authorization mechanism aims at can only be achieved if the authentication and the authorization mechanism work together correctly. We speak of authenticated authorization to refer to the required synthesis of mechanisms for authentication and authorization.
Where used for authorization, the set of authenticated attributes must be meaningful for this purpose, i.e., authorization decisions must be possible based on these attributes. If the authorization policy assigns permissions to an individual entity, the set of authenticated attributes must be suitable to uniquely identify this entity.
In scenarios where devices are communicating autonomously there is often less need to uniquely identify an individual device: For an overseeing principal, the fact that a device belongs to a certain company or that it has a specific type (such as a light bulb) or location may be more important than that it has a unique identifier.
Overseeing principals (RqP and RO) need to decide about the required level of granularity for the authorization. For example, we distinguish device authorization from owner authorization, and binary authorization from unrestricted authorization. In the first case different access permissions are granted to individual devices while in the second case individual owners are authorized. If binary authorization is used, all authenticated entities are implicitly authorized and have the same access permissions. Unrestricted authorization for an item of interest means that no authorization mechanism is used for accessing this resource (not even by authentication) and all entities are able to access the item as they see fit (note that an authorization mechanism may still be used to arrive at the decision to employ unrestricted authorization).
| Authorization granularity | Authorization is contingent on: |
|---|---|
| device | authentication of specific device |
| owner | (authenticated) authorization by owner |
| binary | (any) authentication |
| unrestricted | (unrestricted access; access always authorized) |
More fine-grained authorization does not necessarily provide more security but can be more flexible. Overseeing principals need to consider that an entity should only be granted the permissions it really needs (principle of least privilege), to ensure the confidentiality and integrity of resources.
Client-side authorization solutions aim at protecting the client from disclosing information to or ingesting information from resource servers RqP does not want it to interact with in the given way. Again, binary authorization (the server can be authenticated) may be sufficient, or more fine-grained authorization may be required. The client-side authorization also pertains to the level of protection required for the exchanges with the server (e.g., confidentiality). In the browser web, client-side authorization is often left to the human user that directly controls the client; a constrained client may not have that available all the time but still needs to implement the wishes of the overseeing principal controlling it, the RqP.
For the cases where an authorization solution is needed (all but unrestricted authorization), the enforcing party needs to be able to authenticate the party that is to be authorized. Authentication is therefore required for messages that contain (or otherwise update) representations of an accessed item. More precisely: The enforcing party needs to make sure that the receiver of a message containing a representation is authorized to receive it, both in the case of a client sending a representation to a server and vice versa. In addition, it needs to ensure that the actual sender of a message containing a representation is indeed the one authorized to send this message, again for both the client-to-server and server-to-client case. To achieve this, integrity protection of these messages is required: Authenticity of the message cannot be assured if it is possible for an attacker to modify it during transmission.
In some cases, only one side (client or server side) requires the integrity and / or confidentiality of a resource value. Overseeing principals may decide to omit authentication (unrestricted authorization), or use binary authorization (just employing an authentication mechanism). However, as indicated in Section 3, the security objectives of both sides must be considered, which can often only be achieved when the other side can be relied on to perform some security service.
This and the following section look at the resulting architecture from two different perspectives: This section provides a more detailed description of the various “actors” in the architecture, the logical functional entities performing the tasks required. The following section then will focus on the protocols run between these functional entities.
For the purposes of this document, an actor consists of a set of tasks and additionally has a security domain (client domain or server domain) and a level (constrained, principal, less-constrained). Tasks are assigned to actors according to their security domain and required level.
Note that actors are a concept to understand the security requirements for constrained devices. The architecture of an actual solution might differ as long as the security requirements that derive from the relationship between the identified actors are considered. Several actors might share a single device or even be combined in a single piece of software. Interfaces between actors may be realized as protocols or be internal to such a piece of software.
As described in the problem statement (see Section 2), either C or RS or both of them may be located on a constrained node. We therefore define that C and RS must be able to perform their tasks even if they are located on a constrained node. Thus, C and RS are considered to be Constrained Level Actors.
C performs the following tasks:
RS performs the following tasks:
R is an item of interest such as a sensor or actuator value. R is considered to be part of RS and not a separate actor. The device on which RS is located might contain several resources controlled by different ROs. For simplicity of exposition, these resources are described as if they had separate RS.
As C and RS do not necessarily know each other they might belong to different security domains.
(See Figure 8.)
------- --------
| C | -- requests resource ---> | RS | Constrained Level
------- <-- provides resource--- --------
Figure 8: Constrained Level Actors
Our objective is that C and RS are under control of overseeing principals in the physical world, the Requesting Party (RqP) and the Resource Owner (RO) respectively. The overseeing principals decide about the security policies of their respective endpoints; each overseeing principal belongs to the same security domain as their endpoints.
RqP is in charge of C, i.e. RqP specifies security policies for C, such as with whom C is allowed to communicate. By definition, C and RqP belong to the same security domain.
RqP must fulfill the following task:
RO is in charge of R and RS. RO specifies authorization policies for R and decides with whom RS is allowed to communicate. By definition, R, RS and RO belong to the same security domain.
RO must fulfill the following task:
(See Figure 2.)
Constrained level actors can only fulfill a limited number of tasks and may not have network connectivity all the time. To relieve them from having to manage keys for numerous endpoints and conducting computationally intensive tasks, another level of complexity for actors is introduced (and, thus, a stricter limit on their constrainedness). An actor on the less-constrained level belongs to the same security domain as its respective constrained level actor. They also have the same overseeing principal.
The Client Authorization Server (CAS) belongs to the same security domain as C and RqP. CAS acts on behalf of RqP. It assists C in authenticating RS and determining if RS is an authorized server for R. CAS can do that because for C, CAS is the authority for claims about RS.
CAS performs the following tasks:
The Authorization Server (AS) belongs to the same security domain as R, RS and RO. AS acts on behalf of RO. It supports RS by authenticating C and determining C’s permissions on R. AS can do that because for RS, AS is the authority for claims about C.
AS performs the following tasks:
Devices on the less-constrained level potentially are more powerful than constrained level devices in terms of processing power, memory, non-volatile storage. This results in different characteristics for the protocols used on these levels.
A protocol is considered to be on the constrained level if it is used between the actors C and RS which are considered to be constrained (see Section 5.1). C and RS might not belong to the same security domain. Therefore, constrained level protocols need to work between different security domains.
Commonly used Internet protocols can not in every case be applied to constrained environments. In some cases, tweaking and profiling is required. In other cases it is beneficial to define new protocols which were designed with the special characteristics of constrained environments in mind.
On the constrained level, protocols need to address the specific requirements of constrained environments. Examples for protocols that consider these requirements is the transfer protocol CoAP (Constrained Application Protocol) [RFC7252] and the Datagram Transport Layer Security Protocol (DTLS) [RFC6347] which can be used for channel security.
Constrained devices have only limited storage space and thus cannot store large numbers of keys. This is especially important because constrained networks are expected to consist of thousands of nodes. Protocols on the constrained level should keep this limitation in mind.
We refer to protocols that operate between a constrained device and its corresponding less-constrained device as cross-level support protocols. Protocols used between C and CAS or RS and AS are therefore support protocols.
Support protocols must consider the limitations of their constrained endpoint and therefore belong to the constrained level protocols.
A protocol is considered to be on the less-constrained level if it is used between the actors CAS and AS. CAS and AS might belong to different security domains.
On the less-constrained level, HTTP [RFC7230] and Transport Layer Security (TLS) [RFC8246] can be used alongside or instead of CoAP and DTLS. Moreover, existing security solutions for authentication and authorization such as the OAuth web authorization framework [RFC6749] and Kerberos [RFC4120] can likely be used without modifications and the less-constrained layer is assumed to impose no constraints that would inhibit the traditional deployment/use of, e.g., a Public Key Infrastructure (PKI).
Without anticipating specific solutions, the following considerations may be helpful in discussing them.
The core problem we are trying to solve is authorization. The following problems related to authorization need to be addressed:
The following problems need to be addressed, when considering authentication:
There are different alternatives to provide communication security, and the problem here is to choose the optimal one for each scenario. We list the available alternatives:
With respect to cryptographic keys, we see the following problems that need to be addressed:
In this section we list a set of candidate assumptions and requirements to make the problem description in the previous sections more concise and precise. Note that many of these assumptions and requirements are targeting specific solutions and not the architecture itself.
This document discusses authorization-related tasks for constrained environments and describes how these tasks can be mapped to actors in the architecture.
In this section we focus on specific security aspects related to authorization in constrained-node networks. Section 11.6 of [RFC7252], “Constrained node considerations”, discusses implications of specific constraints on the security mechanisms employed. A wider view of security in constrained-node networks is provided in [I-D.irtf-t2trg-iot-seccons].
The focus of this work is on constrained-node networks consisting of connected constrained devices such as sensors and actuators. The main function of such devices is to interact with the physical world by gathering information or performing an action. We now discuss attacks performed with physical access to such devices.
The main threats to sensors and actuator networks are:
Measuring time and keeping wall-clock time with certain accuracy is important to achieve certain security properties, for example to determine whether keying material an access token, or some other assertion, is valid. The required level of accuracy may differ for different applications.
Dynamic authorization in itself requires the ability to handle expiry or revocation of authorization decisions or to distinguish new authorization decisions from old.
For certain categories of devices we can assume that there is an internal clock which is sufficiently accurate to handle the time measurement requirements. If RS continuously measures time and can connect directly to AS, this relationship can be used to update RS in terms of time, removing some uncertainty, as well as to directly provide revocation information, removing authorizations that are no longer desired.
If RS continuously measures time but can’t connect to AS or another trusted source of time, time drift may have to be accepted and it may be harder to manage revocation. However, RS may still be able to handle short lived access rights within some margins, by measuring the time since arrival of authorization information or request.
Some categories of devices in scope may be unable to measure time with any accuracy (e.g. because of sleep cycles). This category of devices is not suitable for the use cases which require measuring validity of assertions and authorizations in terms of absolute time such as TLS certificates but require a mechanism that is specifically designed for them.
This document has no actions for IANA.
| [HUM14delegation] | Hummen, R., Shafagh, H., Raza, S., Voigt, T. and K. Wehrle, "Delegation-based Authentication and Authorization for the IP-based Internet of Things", 11th IEEE International Conference on Sensing, Communication, and Networking (SECON'14), June 30 - July 3, 2014. |
| [I-D.hardjono-oauth-umacore] | Hardjono, T., Maler, E., Machulak, M. and D. Catalano, "User-Managed Access (UMA) Profile of OAuth 2.0", Internet-Draft draft-hardjono-oauth-umacore-14, January 2016. |
| [I-D.ietf-core-coap-pubsub] | Koster, M., Keranen, A. and J. Jimenez, "Publish-Subscribe Broker for the Constrained Application Protocol (CoAP)", Internet-Draft draft-ietf-core-coap-pubsub-05, July 2018. |
| [I-D.ietf-core-object-security] | Selander, G., Mattsson, J., Palombini, F. and L. Seitz, "Object Security for Constrained RESTful Environments (OSCORE)", Internet-Draft draft-ietf-core-object-security-15, August 2018. |
| [I-D.irtf-t2trg-iot-seccons] | Garcia-Morchon, O., Kumar, S. and M. Sethi, "State-of-the-Art and Challenges for the Internet of Things Security", Internet-Draft draft-irtf-t2trg-iot-seccons-15, May 2018. |
| [REST] | Fielding, R. and R. Taylor, "Principled design of the modern Web architecture", ACM Trans. Inter. Tech. Vol. 2(2), pp. 115-150, DOI 10.1145/514183.514185, May 2002. |
| [RFC4120] | Neuman, C., Yu, T., Hartman, S. and K. Raeburn, "The Kerberos Network Authentication Service (V5)", RFC 4120, DOI 10.17487/RFC4120, July 2005. |
| [RFC4949] | Shirey, R., "Internet Security Glossary, Version 2", FYI 36, RFC 4949, DOI 10.17487/RFC4949, August 2007. |
| [RFC6347] | Rescorla, E. and N. Modadugu, "Datagram Transport Layer Security Version 1.2", RFC 6347, DOI 10.17487/RFC6347, January 2012. |
| [RFC6749] | Hardt, D., "The OAuth 2.0 Authorization Framework", RFC 6749, DOI 10.17487/RFC6749, October 2012. |
| [RFC7228] | Bormann, C., Ersue, M. and A. Keranen, "Terminology for Constrained-Node Networks", RFC 7228, DOI 10.17487/RFC7228, May 2014. |
| [RFC7230] | Fielding, R. and J. Reschke, "Hypertext Transfer Protocol (HTTP/1.1): Message Syntax and Routing", RFC 7230, DOI 10.17487/RFC7230, June 2014. |
| [RFC7231] | Fielding, R. and J. Reschke, "Hypertext Transfer Protocol (HTTP/1.1): Semantics and Content", RFC 7231, DOI 10.17487/RFC7231, June 2014. |
| [RFC7252] | Shelby, Z., Hartke, K. and C. Bormann, "The Constrained Application Protocol (CoAP)", RFC 7252, DOI 10.17487/RFC7252, June 2014. |
| [RFC7744] | Seitz, L., Gerdes, S., Selander, G., Mani, M. and S. Kumar, "Use Cases for Authentication and Authorization in Constrained Environments", RFC 7744, DOI 10.17487/RFC7744, January 2016. |
| [RFC8246] | McManus, P., "HTTP Immutable Responses", RFC 8246, DOI 10.17487/RFC8246, September 2017. |
The authors would like to thank Olaf Bergmann, Robert Cragie, Samuel Erdtman, Klaus Hartke, Sandeep Kumar, John Mattson, Corinna Schmitt, Mohit Sethi, Abhinav Somaraju, Hannes Tschofenig, Vlasios Tsiatsis and Erik Wahlstroem for contributing to the discussion, giving helpful input and commenting on previous forms of this draft. The authors would also like to specifically acknowledge input provided by Hummen and others [HUM14delegation]. Robin Wilton provided extensive editorial comments that were the basis for significant improvements of the text.