The Mercure Protocol

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 13 July 2019.¶

Copyright Notice

Copyright (c) 2019 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 (http://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.¶

Table of Contents

1. Terminology

The keywords MUST, MUST NOT, REQUIRED, SHALL, SHALL NOT, SHOULD, SHOULD NOT, RECOMMENDED, MAY, and OPTIONAL, when they appear in this document, are to be interpreted as described in [RFC2119].¶

• Topic: An HTTP [RFC7230] or HTTPS [RFC2818] topic URL. The unit to which one can subscribe to changes.¶

• Publisher: An owner of a topic. Notifies the hub when the topic feed has been updated. As in almost all pubsub systems, the publisher is unaware of the subscribers, if any. Other pubsub systems might call the publisher the "source". Typically a website or a web API, but can also be a web browser.¶

• Subscriber: A client application that subscribes to real-time updates of topics. Typically a Progressive Web App or a Mobile App, but can also be a server.¶

• Target: A subscriber, or a group of subscribers. A publisher is able to securely dispatch updates to specific targets. Using an HTTP [RFC7230] or HTTPS [RFC2818] URL to identify targets is RECOMMENDED.¶

• Hub: A server that handles subscription requests and distributes the content to subscribers when the corresponding topics have been updated (a hub implementation is provided in this repository). Any hub MAY implement its own policies on who can use it.¶

2. Discovery

If the publisher is a server, it SHOULD advertise the URL of one or more hubs to the subscriber, allowing it to receive live updates when topics are updated. If more than one hub URL is specified, it is RECOMMENDED that the publisher notifies each hub, so the subscriber MAY subscribe to one or more of them.¶

The publisher SHOULD include at least one Link Header [RFC5988] with rel=mercure (a hub link header). The target URL of these links MUST be a hub implementing the Mercure protocol.¶

Note: this relation type has not been registered yet [RFC5988]. In the meantime, the relation type https://git.io/mercure MAY be used instead.¶

The publisher MAY provide the following target attributes in the Link Headers:¶

• last-event-id: the globally unique identifier of the last event dispatched by the publisher at the time of the generation of this resource. If provided, it MUST be passed to the hub through a query parameter called Last-Event-ID and will be used to ensure that possible updates having been made during between the resource generation time and the connection to the hub are not lost. See section #Re-Connection-and-State-Reconciliation. If this attribute is provided, the publisher MUST always set the id parameter when sending updates to the hub.¶

• content-type: the content type of the updates that will pushed by the hub. If omitted, the subscriber MUST assume that the content type will be the same as that of the original resource. Setting the content-type attribute is especially useful to hint that partial updates will be pushed, using formats such as JSON Patch [RFC6902] or JSON Merge Patch [RFC7386].¶

• key-set=<JWKS>: the key(s) to decrypt updates encoded in the JWKS (JSON Web Key Set) format (see the Encryption section).¶

All these attributes are optional.¶

The publisher MAY also include one Link Header [RFC5988] with rel=self (the self link header). It SHOULD contain the canonical URL for the topic to which subscribers are expected to use for subscriptions. If the Link with rel=self is omitted, the current URL of the resource MUST be used as a fallback.¶

Minimal example:¶

GET /books/foo.jsonld HTTP/1.1
Host: example.com

HTTP/1.1 200 Ok
Content-type: application/ld+json
Link: <https://example.com/hub>; rel="mercure"

{"@id": "/books/foo.jsonld", "foo": "bar"}

Links embedded in HTML or XML documents (as defined in the WebSub recommendation) MAY also be supported by subscribers.¶

Note: the discovery mechanism described in this section is strongly inspired from the one specified in the WebSub recommendation.¶

3. Subscriptions

The subscriber subscribes to a URL exposed by a hub to receive updates from one or many topics. To subscribe to updates, the client opens an HTTPS connection following the Server-Sent Events specification to the hub's subscription URL advertised by the publisher. The GET HTTP method must be used. The connection SHOULD use HTTP/2 to leverage mutliplexing and other advanced features of this protocol.¶

The subscriber specifies the list of topics to get updates from by using one or several query parameters named topic. The value of these query parameters MUST be URI templates [RFC6570].¶

Note: a URL is also a valid URI template.¶

The protocol doesn't specify the maximum number of topic parameters that can be sent, but the hub MAY apply an arbitrary limit.¶

The EventSource JavaScript interface MAY be used to establish the connection. Any other appropriate mechanism including, but not limited to, readable streams and XMLHttpRequest (used by popular polyfills) MAY also be used.¶

The hub sends updates concerning all subscribed resources matching the provided URI templates and the provided targets (see section #Authorization). If no targets are specified, the update is dispatched to all subscribers. The hub MUST send these updates as text/event-stream compliant events.¶

The data property MUST contain the new version of the topic. It can be the full resource, or a partial update by using formats such as JSON Patch @RFC6902 or JSON Merge Patch @RFC7386.¶

All other properties defined in the Server-Sent Events specification MAY be used and SHOULD be supported by hubs.¶

The resource SHOULD be represented in a format with hypermedia capabilities such as JSON-LD [W3C.REC-json-ld-20140116], Atom [RFC4287], XML [W3C.REC-xml-20081126] or HTML [W3C.REC-html52-20171214].¶

Web Linking [RFC5988] SHOULD be used to indicate the IRI of the resource sent in the event. When using Atom, XML or HTML as the serialization format for the resource, the document SHOULD contain a link element with a self relation containing the IRI of the resource. When using JSON-LD, the document SHOULD contain an @id property containing the IRI of the resource.¶

Example:¶

// The subscriber subscribes to updates for the https://example.com/foo topic
// and to any topic matching https://example.com/books/{name}
const url = new URL('https://example.com/hub');
url.searchParams.append('topic', 'https://example.com/foo');
url.searchParams.append('topic', 'https://example.com/bar/{id}');

const eventSource = new EventSource(url);

// The callback will be called every time an update is published
eventSource.onmessage = function ({data}) {
    console.log(data);
};

The hub MAY require that subscribers are authorized to receive updates.¶

4. Publication

The publisher send updates by issuing POST HTTPS requests on the hub URL. When it receives an update, the hub dispatches it to subscribers using the established server-sent events connections.¶

An application CAN send events directly to subscribers without using an external hub server, if it is able to do so. In this case, it MAY NOT implement the endpoint to publish updates.¶

The request MUST be encoded using the application/x-www-form-urlencoded format and contain the following data:¶

• topic: IRIs of the updated topic. If this key is present several times, the first occurrence is considered to be the canonical URL of the topic, and other ones are considered to be alternate URLs. The hub MUST dispatch this update to subscribers that are subscribed to both canonical or alternate URLs.¶

• data: The content of the new version of this topic.¶

• target (optional): Target audience of this update. This key can be present several times. See section #Authorization for further information.¶

• id (optional): The topic's revision identifier: it will be used as the SSE's id property. If omitted, the hub MUST generate a valid globally unique id. It MAY be a UUID. Even if provided, the hub MAY ignore the id provided by the client and generate its own id.¶

• type (optional): The SSE's event property (a specific event type).¶

• retry (optional): The SSE's retry property (the reconnection time).¶

In the event of success, the HTTP response's body MUST be the id associated to this update generated by the hub and a success HTTP status code MUST be returned. The publisher MUST be authorized to publish updates. See section #Authorization.¶

5. Authorization

To ensure that they are authorized, both publishers and subscribers must present a valid JWS [RFC7515] in compact serialization to the hub. This JWS SHOULD be short-lived, especially if the subscriber is a web browser. A different key MAY be used to sign subscribers' and publishers' tokens.¶

Two mechanisms are defined to present the JWS to the hub:¶

• using an Authorization HTTP header¶

• using a cookie¶

If the publisher or the subscriber is not a web browser, it SHOULD use an Authorization HTTP header. This Authorization header MUST contain the string Bearer followed by the JWS. The hub will check that the JWS conforms to the rules (defined later) ensuring that the client is authorized to publish or subscribe to updates.¶

By the EventSource specification, web browsers can not set custom HTTP headers for such connections, and they can only be estabilished using the GET HTTP method. However, cookies are supported and can be included even in cross-domain requests if the CORS credentials are set:¶

If the publisher or the subscriber is a web browser, it SHOULD send a cookie called mercureAuthorization containing the JWS when connecting to the hub.¶

Whenever possible, the mercureAuthorization cookie SHOULD be set during the discovery to improve the overall security. See section #Discovery. Consequently, if the cookie is set during the discovery, both the publisher and the hub have to share the same second level domain. The Domain attribute MAY be used to allow the publisher and the hub to use different subdomains.¶

The cookie SHOULD have the Secure, HttpOnly and SameSite attributes set. The cookie's Path attribute SHOULD also be set to the hub's URL. See section #Security-Considerations.¶

When using authorization mechanisms, the connection MUST use an encryption layer such as HTTPS.¶

If both an Authorization HTTP header and a cookie named mercureAuthorization are presented by the client, the cookie MUST be ignored. If the client tries to execute an operation it is not allowed to, a 403 HTTP status code SHOULD be returned.¶

6. Reconnection and State Reconciliation

To allow re-establishment in case of connection lost, events dispatched by the hub SHOULD include an id property. The value contained in this id property SHOULD be a globally unique identifier. To do so, a UUID [RFC4122] MAY be used.¶

According to the server-sent events specification, in case of connection lost the subscriber will try to automatically re-connect. During the re-connection, the subscriber MUST send the last received event id in a Last-Event-ID HTTP header.¶

The server-sent events specification doesn't allow this HTTP header to be set during the first connection (before a reconnection). In order to fetch any update dispatched between the initial resource generation by the publisher and the connection to the hub, the subscriber MUST send the event id provided during the discovery in the last-event-id link's attribute in a query parameter named Last-Event-ID when connecting to the hub.¶

If both the Last-Event-ID HTTP header and the query parameter are present, the HTTP header MUST take precedence.¶

If the Last-Event-ID HTTP header or query parameter exists, the hub SHOULD send all events published following the one bearing this identifier to the subscriber.¶

The hub MAY discard some messages for operational reasons. The subscriber MUST NOT assume that no update will be lost, and MUST re-fetch the original topic to ensure this (for instance, after a long disconnection time).¶

The hub MAY also specify the reconnection time using the retry key, as specified in the server-sent events format.¶

7. Encryption

Using HTTPS does not prevent the hub from accessing the update's content. Depending of the intended privacy of information contained in the update, it MAY be necessary to prevent eavesdropping by the hub.¶

To make sure that the message content can not be read by the hub, the publisher MAY encode the message before sending it to the hub. The publisher SHOULD use JSON Web Encryption [RFC7516] to encrypt the update content. The publisher MAY provide the relevant encryption key(s) in the key-set attribute of the Link HTTP header during the discovery. The key-set attribute SHOULD contain a key encoded using the JSON Web Key Set [RFC7517] format. Any other out-of-band mechanism MAY be used instead to share the key between the publisher and the subscriber.¶

Update encyption is considered a best practice to prevent mass surveillance. This is especially relevant if the hub is managed by an external provider.¶

8. Security Considerations

The confidentiality of the secret key(s) used to generate the JWTs is a primary concern. The secret key(s) MUST be stored securely. They MUST be revoked immediately in the event of compromission.¶

Possessing valid JWTs allows any client to subscribe, or to publish to the hub. Their confidentiality MUST therefore be ensured. To do so, JWTs MUST only be transmitted over secure connections.¶

Also, when the client is a web browser, the JWT SHOULD not be made accessible to JavaScript scripts for resilience against Cross-site Scription (XSS) attacks). It's the main reason why, when the client is a web browser, using HttpOnly cookies as the authorization mechanism SHOULD always be preferred.¶

In the event of compromission, revoking JWTs before their expiration is often difficult. To that end, using short-lived tokens is strongly RECOMMENDED.¶

The publish endpoint of the hub may be targeted by Cross-Site Request Forgery (CSRF) attacks) when the cookie-based authorization mechanism is used. Therefore, implementations supporting this mechanism MUST mitigate such attacks.¶

The first prevention method to implement is to set the mercureAuthorization cookie's SameSite attribute. However, some web browsers still not support this attribute and will remain vulnerable. Additionally, hub implementations SHOULD use the Origin and Referer HTTP headers set by web browsers to verify that the source origin matches the target origin. If none of these headers are available, the hub SHOULD discard the request.¶

CSRF prevention techniques, including those previously mentioned, are described in depth in OWASP's Cross-Site Request Forgery (CSRF) Prevention Cheat SheetPreventionCheat_Sheet).¶

Author's Address