Internet-Draft CoRE Resource Directory Extensions July 2022
Amsüss Expires 11 January 2023 [Page]
Workgroup:
CoRE
Internet-Draft:
draft-amsuess-core-resource-directory-extensions-06
Published:
Intended Status:
Experimental
Expires:
Author:
C. Amsüss

CoRE Resource Directory Extensions

Abstract

A collection of extensions to the Resource Directory [rfc9176] that can stand on their own, and have no clear future in specification yet.

Discussion Venues

This note is to be removed before publishing as an RFC.

Discussion of this document takes place on the Constrained RESTful Environments Working Group mailing list (core@ietf.org), which is archived at https://mailarchive.ietf.org/arch/browse/core/.

Source for this draft and an issue tracker can be found at https://gitlab.com/chrysn/resource-directory-extensions.

Status of This Memo

This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.

Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at 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 11 January 2023.

Table of Contents

1. Introduction

This document pools some extensions to the Resource Directory [rfc9176] that might be useful but have no place in the original document.

They might become individual documents for IETF submission, simple registrations in the RD Parameter Registry at IANA, or grow into a shape where they can be submitted as a collection of tools.

At its current state, this draft is a collection of ideas.

2. Reverse Proxy requests

When a registrant registers at a Resource Directory, it might not have a suitable address it can use as a base address. Typical reasons include being inside a NAT without control over port forwarding, or only being able to open outgoing connections (as program running inside a web browser utilizing CoAP over WebSocket [RFC8323] might be).

[rfc9176] suggests (in the Cellular M2M use case) that proxy access to such endpoints can be provided, it gives no concrete mechanism to do that; this is such a mechanism.

This mechanism is intended to be a last-resort option to provide connectivity. Where possible, direct connections are preferred. Before registering for proxying, the registrant should attempt to obtain a publicly available port, for example using PCP ([RFC6887]).

The same mechanism can also be employed by registrants that want to conceal their network address from its clients.

A deployed application where this is implicitly done is LwM2M [citation missing]. Notable differences are that the protocol used between the client and the proxying RD is not CoAP but application specific, and that the RD (depending on some configuration) eagerly populates its proxy caches by sending requests and starting observations at registration time.

2.1. Discovery

An RD that provides proxying functionality advertises it by announcing the additional resource type "TBD1" on its directory resource.

2.2. Registration

A client passes the "proxy=yes" or "proxy=ondemand" query parameter in addition to (but typically instead of) a "base" query parameter.

A server that receives a "proxy=yes" query parameter in a registration (or receives "proxy=ondemand" and decides it needs to proxy) MUST come up with a "Proxy URL" on which it accepts requests, and which it uses as a Registration Base URI for lookups on the present registration.

The Proxy URL SHOULD have no path component, as acting as a reverse proxy in such a scenario means that any relative references in all representations that are proxied must be recognized and possibly rewritten.

The RD MAY accept connections also on alternative Registration Base URIs using different protocols; it can advertise them using the mechanisms of [I-D.ietf-core-transport-indication].

The registrant is not informed of the chosen public name by the RD.

This mechanism is applicable to all transports that can be used to register. If proxying is active, the restrictions on when the base parameter needs to be present ([rfc9176] Registration template) are relaxed: The base parameter may also be absent if the connection originates from an ephemeral port, as long as the underlying protocol supports role reversal, and link-local IPv6 addresses may be used without any concerns of expressibility.

If the client uses the role reversal rule relaxation, both it and the server keep that connection open for as long as it wants to be reachable. When the connection terminates, the RD SHOULD treat the registration as having timed out (even if its lifetime has not been exceeded) and MAY eventually remove the registration. It is yet to be decided whether the RD's announced ability to do proxying should imply that infinite lifetimes are necessarily supported for such registrations; at least, it is RECOMMENDED.

2.2.1. Registration updates

The "proxy" query parameter can not be changed or repeated in a registration update; RD servers MUST answer 4.00 Bad Request to any registration update that has a "proxy" query parameter.

As always, registration updates can explicitly or implicitly update the Registration Base URI. In proxied registrations, those changes are not propagated to lookup, but do change the forwarding address of the proxy.

For example, if a registration is established over TCP, an update can come along in a new TCP connection. Starting then, proxied requests are forwarded along that new connection.

2.2.2. Proxy behavior

The RD operates as a reverse-proxy as described in [RFC7252] Section 5.7.3 at the announced Proxy URL(s), where it decides based on the requested host and port to which registrant endpoint to forward the request.

The address the incoming request are forwarded to is the base address of the registration. If an explicit "base" paremter is given, the RD will forward requests to that location. Otherwise, it forwards to the registration's source address (which is the implied base parameter).

When an implicit base is used, the requests forwarded by the RD to the EP contain no Uri-Host option. EPs that want to run multiple parallel registrations (especially gateway-like devices) can either open up separate connections, or use an additional (to-be-specified) mechanism to set the "virtual host name" for that registration in a separate argument.

2.2.2.1. Limitations from using a reverse proxy

The registrant requesting the reverse proxying needs to ensure that all services it provides are compatible with being operated behind a reverse proxy with an unknown name. In particular, this rules out all applications that refer to local resources by a full URI (as opposed to relative references without scheme and host). Applications behind a reverse proxy can not use role reversal.

Some of these limitations can be mitigated if the application knows its advertised address.

2.2.3. On-Demand proxying

If an endpoint is deployed in an unknown network, it might not know whether it is behind a NAT that would require it to configure an explicit base address, and ask the RD to assist by proxying if necessary by registering with the "proxy=ondemand" query parameter.

A server receiving that SHOULD use a different IP address to try to access the registrant's .well-known/core file using a GET request under the Registration Base URI. If that succeeds, it may assume that no NAT is present, and ignore the proxying request. Otherwise, it configures proxying as if "proxy=yes" were requested.

Note that this is only a heuristic [ and not tested in deployments yet ].

2.2.4. Multiple upstreams

When a proxying RD is operating behind a router that has uplinks with multiple provisioning domains (see [RFC7556]) or a similar setup, it MAY mint multiple addresses that are reachable on the respective provisioning domains. When possible, it is preferred to keep the number of URIs handed out low (avoiding URI aliasing); this can be achieved by announcing both the proxy's public addresses under the same wildcard name.

If RDs are announced by the uplinks using RDAO, the proxy may use the methods of [I-D.amsuess-core-rd-replication] to distribute its registrations to all the announced upstream RDs.

In such setups, the router can forward the upstream RDs using the PvD option ([RFC8801]) to PvD-aware hosts and only announce the local RD to PvD-unaware ones (which then forwards their registrations). It can be expected that PvD-aware endpoints are capable of registering with multiple RDs simultaneously.

2.2.5. Examples

2.2.5.1. Registration through a firewall
Req from [2001:db8:42::9876]:5683:
POST coap://rd.example.net/rd?ep=node9876&proxy=ondemand
</some-resource>;rt="example.x"

Req from other-address.rd.example.net:
GET coap://[2001:db8:42::9876]/.well-known/core

Request blocked by stateful firewall around [2001:db8:42::]

RD decides that proxying is necessary

Res: 2.04 Created
Location: /reg/abcd

Later, lookup of that registration might say:

Req: GET coap://rd.example.net/lookup/res?rt=example.x

Res: 2.05 Content
<coap://node987.rd.example.net/some-resource>;rt="example.x

A request to that resource will end up at an IP address of the RD, which will forward it using its the IP and port on which the registrant had registered as source port, thus reaching the registrant through the stateful firewall.

2.2.5.2. Registration from a browser context
Req: POST coaps+ws://rd.example.net/rd?ep=node1234&proxy=yes
</gyroscope>;rt="core.s"

Res: 2.04 Created
Location: /reg/123

The gyroscope can now not only be looked up in the RD, but also be reached:

Req: GET coap://rd.example.net/lookup/res?rt=core.s

Res: 2.05 Content
<coap://[2001:db8:1::1]:10123/gyroscope>;rt="core.s"

In this example, the RD has chosen to do port-based rather than host-based virtual hosting and announces its literal IP address as that allows clients to not send the lengthy Uri-Host option with all requests.

2.2.6. Notes on stability and maturity

Using this with UDP can be quite fragile; the author only draws on own experience that this can work across cell-phone NATs and does not claim that this will work over generic firewalls.

[ It may make sense to have the example as TCP right away. ]

2.2.7. Security considerations

An RD MAY impose additional restrictions on which endpoints can register for proxying, and thus respond 4.01 Unauthorized to request that would pass had they not requested proxying.

Attackers could do third party registrations with an attacked device's address as base URI, though the RD would probably not amplify any attacks in that case.

The RD MUST NOT reveal the address at which it reaches the registrant except for adaequately authenticated and authorized debugging purposes, as that address could reveal sensitive location data the registrant may wish to hide by using a proxy.

Usual caveats for proxies apply.

3. Infinite lifetime

An RD can indicate support for infinite lifetimes by adding the resoruce type "TBD2" to its list of resource types.

A registrant that wishes to keep its registration alive indefinitely can set the lifetime value as "lt=inf".

Registrations with infinite lifetimes never time out. Unlike regular registrations, they are not "soft state"; the registrant can expect the RD to persist the registrations across network changes, reboots, softare updates and that like.

Typical use cases for infinite life times are:

3.1. Example

Had the example of Section 2.2.5.2 discovered support for infinite lifetimes during lookup like this:

Req: GET coaps+ws://rd.example.net/.well-known/core?rt=core.rd*

Res: 2.05 Content
</rd>;rt="core.rd TBD1 TBD2";ct=40

it could register like that:

Req: POST coaps+ws://rd.example.net/rd?ep=node1234&proxy=yes&lt=inf
</gyroscope>;rt="core.s"

Res: 2.04 Created
Location: /reg/123

and never need to update the registration for as long as the websocket connection is open.

(When it gets terminated, it could try renewing the registration, but needs to be prepared for the RD to already have removed the original registration.)

4. Limited lifetimes

Even if an RD supports infinite lifetimes, it may not accept them from just any registrant. Even more, an RD may have policies in place that require a certain frequency of updates and thus place an upper limit on lt lower than the technical limit of 136 years.

This document does not define any means of communicating lifetime limits, but explores a few options:

6. Lifetime Age

This extension is described in [I-D.amsuess-core-rd-replication] Section 5.2.

The "provenance" extension in Section 5.1 of the same document should probably be expressed differently to avoid using non-target link attributes.

7. Zone identifier introspection

The 'split-horizon' mechanism of [rfc9176] (that registrations with link-local bases can only be read from the zone they registered on) reduces the usability of the endpoint lookup interface for debugging purposes.

To allow an administrator to read out the "show-zone-id" query parameter for endpoint and resource lookup is introduced.

A Resource Directory that understands this parameter MUST NOT limit lookup results to registrations from the lookup's zone, and MUST use [RFC6874] zone identifiers to annotate which zone those registrations are valid on.

The RD MUST limit such requests to authenticated and authorized debugging requests, as registrants may rely on the RD to keep their presence secret from other links.

7.1. Example

Req: GET /rd-lookup/ep?show-zone-id&et=printer

Res: 2.05 Content
</reg/1>;base="coap://[2001:db8::1]";et=printer;ep="bigprinter",
</reg/2>;base="coap://[fe80::99%wlan0]";et=printer;ep="localprinter-1234",
</reg/3>;base="coap://[fe80::99%eth2]";et=printer;ep="localprinter-5678",

8. Proxying multicast requests

Multicast requests are hard to forward at a proxy: Even if a media type is used in which multiple responses can be aggregated transparently, the proxy can not reliably know when all responses have come in. [RFC7390] Section 2.9 describes the difficulties in more detail.

Note that [I-D.tiloca-core-groupcomm-proxy] provides a mechanism that does allow the forwarding of multicast requests. It is yet to be determined what the respective pros and cons are. Conversely, that lookup mechanism may also serve as an alternative to resource lookup on an RD.

A proxy MAY expose an interface compatible with the RD lookup interface, which SHOULD be advertised by a link to it that indicates the resource types core.rd-lookup-res and TBD4.

The proxy sends multicast requests to All CoAP Nodes ([RFC7252] Section 12.8) requesting their .well-known/core files either eagerly (ie. in regular intervals independent of queries) or on demand (in which case it SHOULD limit the results by applying [RFC6690] query filtering; if it has received multiple query parameters it should forward the one it deems most likely to limit the results, as .well-known/core only supports a single query parameter).

In comparison to classical RD operation, this RD behaves roughly as if it had received a simple registration with a All CoAP Nodes address as the source address, if such behavior were specified. The individual registrations that result from this neither have an explicit registration resource nor an explicit endpoint name; given that the endpoint lookup interface is not present on such proxies, neither can be queried.

Clients that would intend to do run a multicast discovery operation behind the proxy can then instead query that resource lookup interface. They SHOULD use observation on lookups, as an on-demand implementation MAY return the first result before others have arrived, or MAY even return an empty link set immediately.

8.1. Example

Req: GET coap+ws://gateway.example.com/.well-known/core?rt=TBD4

Res: 2.05 Content
</discover>;rt="core.rd-lookup-res TBD4";ct=40

Req: GET coap+ws://gateway.example.com/discover?rt=core.s
Observe: 0

Res: 2.05 Content
Observe: 0
Content-Format: 40
(empty payload)

At the same time, the proxy sends out multicast requests on its interfaces:

Req: GET coap://ff05::fd/.well-known/core?rt=core.s

Res (from [2001:db8::1]:5683): 2.05 Content
</temp>;ct="0 112";rt="core.s"

Res (from [2001:db8::2]:5683): 2.05 Content
</light>;ct="0 112";rt="core.s"

upon receipt of which it sends out a notification to the websocket client:

Res: 2.05 Content
Observe: 1
Content-Format: 40
<coap://[2001:db8::1]/temp>;ct="0 112";rt="core.s";anchor="coap://[2001:db8::1]",
<coap://[2001:db8::2]/light>;ct="0 112";rt="core.s";anchro="coap://[2001:db8::2]"

9. Opportunistic RD

An application that wants to advertise its resources in Resource Directory can find itself in a network that has no RD deployed. It may be able to start an RD on its own to fill in that gap until an explicitly configured one gets installed.

This bears the risk of having competing RDs on the same network, where resources registered at one can not be discovered on the other. To mitigate that, such Opportunistic Resource Directories should follow those steps:

Future iterations of this document may want to cut down on the possibilities listed above.

Some ideas are around for making the shutdown of a commissioned or otherwise high-capability RD more graceful, but they still have some problems

Installations of Opportunistic RDs are at special risk of resource exhaustion because they are not sized with their actual deployment in mind, but rely on defaults set by the application that starts the RD. Opportunistic RDs should only be started if the application's administrator can be informed in a timely fashion when the RD's resources are nearing exhaustion; guidance towards installing a more capable RD on the network should be provided in that case.

9.1. Applications

  • Group managers using [I-D.tiloca-core-oscore-discovery] can ship its own low-priority Opportunistic RD to announce its join resources. This provides benefits over announcing them on multicast discovery if the network can efficiently route requests to the All CoAP Resource Directories multicast address (so group members get a response back from an early focused request to all RDs rather than falling back to multicasting All CoAP Nodes for ?rt=osc.j&...), or if discovery of the group's multicast address is used.
  • Administrative tools that try to provide a broad overview of a network's CoAP devices could offer to open an Opportunistic RD if they find no active RD on the network (but should ask the user in interactive scenarios).

    That allows them to see devices that newly join the network quickly (by observing their own or the found RD), rather than relying periodic multicasts.

10. References

10.1. Normative References

[I-D.amsuess-core-rd-replication]
Amsüss, C., "Resource Directory Replication", Work in Progress, Internet-Draft, draft-amsuess-core-rd-replication-02, , <https://datatracker.ietf.org/doc/html/draft-amsuess-core-rd-replication-02>.
[RFC6874]
Carpenter, B., Cheshire, S., and R. Hinden, "Representing IPv6 Zone Identifiers in Address Literals and Uniform Resource Identifiers", RFC 6874, DOI 10.17487/RFC6874, , <https://www.rfc-editor.org/rfc/rfc6874>.
[RFC7252]
Shelby, Z., Hartke, K., and C. Bormann, "The Constrained Application Protocol (CoAP)", RFC 7252, DOI 10.17487/RFC7252, , <https://www.rfc-editor.org/rfc/rfc7252>.
[rfc9176]
Amsüss, C., Ed., Shelby, Z., Koster, M., Bormann, C., and P. van der Stok, "Constrained RESTful Environments (CoRE) Resource Directory", RFC 9176, DOI 10.17487/RFC9176, , <https://www.rfc-editor.org/rfc/rfc9176>.

10.2. Informative References

[I-D.ietf-core-problem-details]
Fossati, T. and C. Bormann, "Concise Problem Details For CoAP APIs", Work in Progress, Internet-Draft, draft-ietf-core-problem-details-08, , <https://datatracker.ietf.org/doc/html/draft-ietf-core-problem-details-08>.
[I-D.ietf-core-transport-indication]
Amsüss, C., "CoAP Protocol Indication", Work in Progress, Internet-Draft, draft-ietf-core-transport-indication-00, , <https://datatracker.ietf.org/doc/html/draft-ietf-core-transport-indication-00>.
[I-D.tiloca-core-groupcomm-proxy]
Tiloca, M. and E. Dijk, "Proxy Operations for CoAP Group Communication", Work in Progress, Internet-Draft, draft-tiloca-core-groupcomm-proxy-06, , <https://datatracker.ietf.org/doc/html/draft-tiloca-core-groupcomm-proxy-06>.
[I-D.tiloca-core-oscore-discovery]
Tiloca, M., Amsuess, C., and P. V. D. Stok, "Discovery of OSCORE Groups with the CoRE Resource Directory", Work in Progress, Internet-Draft, draft-tiloca-core-oscore-discovery-11, , <https://datatracker.ietf.org/doc/html/draft-tiloca-core-oscore-discovery-11>.
[RFC6690]
Shelby, Z., "Constrained RESTful Environments (CoRE) Link Format", RFC 6690, DOI 10.17487/RFC6690, , <https://www.rfc-editor.org/rfc/rfc6690>.
[RFC6887]
Wing, D., Ed., Cheshire, S., Boucadair, M., Penno, R., and P. Selkirk, "Port Control Protocol (PCP)", RFC 6887, DOI 10.17487/RFC6887, , <https://www.rfc-editor.org/rfc/rfc6887>.
[RFC7390]
Rahman, A., Ed. and E. Dijk, Ed., "Group Communication for the Constrained Application Protocol (CoAP)", RFC 7390, DOI 10.17487/RFC7390, , <https://www.rfc-editor.org/rfc/rfc7390>.
[RFC7556]
Anipko, D., Ed., "Multiple Provisioning Domain Architecture", RFC 7556, DOI 10.17487/RFC7556, , <https://www.rfc-editor.org/rfc/rfc7556>.
[RFC8323]
Bormann, C., Lemay, S., Tschofenig, H., Hartke, K., Silverajan, B., and B. Raymor, Ed., "CoAP (Constrained Application Protocol) over TCP, TLS, and WebSockets", RFC 8323, DOI 10.17487/RFC8323, , <https://www.rfc-editor.org/rfc/rfc8323>.
[RFC8801]
Pfister, P., Vyncke, É., Pauly, T., Schinazi, D., and W. Shao, "Discovering Provisioning Domain Names and Data", RFC 8801, DOI 10.17487/RFC8801, , <https://www.rfc-editor.org/rfc/rfc8801>.

Appendix A. Change log

Since -05:

Since -04:

Since -03:

Since -02:

Since -01:

Since -00:

Appendix B. Acknowledgements

[ Reviews from: Jaime Jimenez ]

Section 4 was inspired by Ben Kaduk's comments from reviewing [rfc9176].

Author's Address

Christian Amsüss