HTTP Working Group | M. Nottingham |
Internet-Draft | M. Thomson |
Intended status: Experimental | Mozilla |
Expires: December 23, 2016 | June 21, 2016 |
This document describes how http URIs can be accessed using Transport Layer Security (TLS) to mitigate pervasive monitoring attacks.¶
Discussion of this draft takes place on the HTTP working group mailing list (ietf-http-wg@w3.org), which is archived at https://lists.w3.org/Archives/Public/ietf-http-wg/.¶
Working Group information can be found at http://httpwg.github.io/; source code and issues list for this draft can be found at https://github.com/httpwg/http-extensions/labels/opp-sec.¶
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 http://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 December 23, 2016.¶
Copyright (c) 2016 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.¶
This document describes a use of HTTP Alternative Services [RFC7838] to decouple the URI scheme from the use and configuration of underlying encryption, allowing a http URI [RFC7230] to be accessed using Transport Layer Security (TLS) [RFC5246] opportunistically.¶
Serving https URIs require acquiring and configuring a valid certificate, which means that some deployments find supporting TLS difficult. This document describes a usage model whereby sites can serve http URIs over TLS without being required to support strong server authentication.¶
Opportunistic Security [RFC7435] does not provide the same guarantees as using TLS with https URIs; it is vulnerable to active attacks, and does not change the security context of the connection. Normally, users will not be able to tell that it is in use (i.e., there will be no “lock icon”).¶
The immediate goal is to make the use of HTTP more robust in the face of pervasive passive monitoring [RFC7258].¶
A secondary goal is to limit the potential for active attacks. It is not intended to offer the same level of protection as afforded to https URIs, but instead to increase the likelihood that an active attack can be detected.¶
A final (but significant) goal is to provide for ease of implementation, deployment and operation. This mechanism is expected to have a minimal impact upon performance, and require a trivial administrative effort to configure.¶
An origin server that supports the resolution of http URIs can indicate support for this specification by providing an alternative service advertisement [RFC7838] for a protocol identifier that uses TLS, such as h2 [RFC7540].¶
A client that receives such an advertisement MAY make future requests intended for the associated origin ([RFC6454]) to the identified service (as specified by [RFC7838]).¶
A client that places the importance of protection against passive attacks over performance might choose to withhold requests until an encrypted connection is available. However, if such a connection cannot be successfully established, the client can resume its use of the cleartext connection.¶
A client can also explicitly probe for an alternative service advertisement by sending a request that bears little or no sensitive information, such as one with the OPTIONS method. Likewise, clients with existing alternative services information could make such a request before they expire, in order minimize the delays that might be incurred.¶
Client certificates are not meaningful for URLs with the “http” scheme, and therefore clients creating new TLS connections to alternative services for the purposes of this specification MUST NOT present them. Established connections with client certificates MAY be reused, however.¶
[RFC7838] requires that an alternative service only be used when there are “reasonable assurances” that it is under control of and valid for the whole origin.¶
As defined in that specification, a client can establish reasonable assurances when using a TLS-based protocol with the certificate checks defined in [RFC2818].¶
For the purposes of this specification, an additional way of establishing reasonable assurances is available when the alternative is on the same host as the origin, using the “http-opportunistic” well-known URI defined in Section 6.¶
This allows deployment without the use of valid certificates, to encourage deployment of opportunistic security. When it is in use, the alternative service can provide any certificate, or even select TLS cipher suites that do not include authentication.¶
When a client has a valid http-opportunistic response for an origin (as per Section 6), it MAY consider there to be reasonable assurances as long as:¶
For example, this request/response pair would constitute reasonable assurances for the origin “http://www.example.com” for an alternative service on port 443 or 8000 of the host “www.example.com”:¶
GET /.well-known/http-opportunistic HTTP/1.1 Host: www.example.com HTTP/1.1 200 OK Content-Type: application/json Connection: close { "http://www.example.com": { "tls-ports": [443, 8000], "lifetime": 2592000 } }
Note that this mechanism is only defined to establish reasonable assurances for the purposes of this specification; it does not apply to other uses of alternative services unless they explicitly invoke it.¶
When using alternative services, requests for resources identified by both http and https URIs might use the same connection, because HTTP/2 permits requests for multiple origins on the same connection.¶
Since https URIs rely on server authentication, a connection that is initially created for http URIs without authenticating the server cannot be used for https URIs until the server certificate is successfully authenticated. Section 3.1 of [RFC2818] describes the basic mechanism, though the authentication considerations in Section 2.1 of [RFC7838] also apply.¶
Connections that are established without any means of server authentication (for instance, the purely anonymous TLS cipher suites) cannot be used for https URIs.¶
Because of the risk of server confusion about individual requests’ schemes (see Section 8.4), clients MUST NOT mix “https” and “http” requests on the same connection unless the http-opportunistic response’s origin object Section 6 has a “mixed-scheme” member whose value is “true”.¶
Even when the alternative service is strongly authenticated, opportunistically upgrading cleartext HTTP connections to use TLS is subject to active attacks. In particular:¶
Given that the primary goal of this specification is to prevent passive attacks, these are not critical failings (especially considering the alternative - HTTP over cleartext). However, a modest form of protection against active attacks can be provided for clients on subsequent connections.¶
When an origin is able to commit to providing service for a particular origin over TLS for a bounded period of time, clients can choose to rely upon its availability, failing when it cannot be contacted. Effectively, this makes the choice to use a secured protocol “sticky”.¶
An origin can reduce the risk of attacks on opportunistically secured connections by committing to provide a secured, authenticated alternative service. This is done by including the optional tls-commit member in the origin object of the http-opportunistic well-known response (see Section 6).¶
This feature is optional due to the requirement for server authentication and the potential risk entailed (see Section 5.3).¶
When the value of the tls-commit member is “true” ([RFC7159], Section 3), it indicates that the origin makes such a commitment for the duration of the origin object lifetime.¶
{ "http://www.example.com": { "tls-ports": [443,8080], "tls-commit": true, "lifetime": 3600 } }
Including tls-commit creates a commitment to provide a secured alternative service for the advertised period. Clients that receive this commitment can assume that a secured alternative service will be available for the origin object lifetime. Clients might however choose to limit this time (see Section 5.3).¶
The value of the tls-commit member MUST be ignored unless the alternative service can be strongly authenticated. The same authentication requirements that apply to https:// resources SHOULD be applied to authenticating the alternative. Minimum authentication requirements for HTTP over TLS are described in Section 2.1 of [RFC7838] and Section 3.1 of [RFC2818]. As noted in [RFC7838], clients can impose other checks in addition to this minimum set. For instance, a client might choose to apply key pinning [RFC7469].¶
A client that receives a commitment and that successfully authenticates the alternative service can assume that a secured alternative will remain available for the origin object lifetime.¶
A client SHOULD avoid sending requests via cleartext protocols or to unauthenticated alternative services for the duration of the origin object lifetime, except to discover new potential alternatives.¶
A commitment is not bound to a particular alternative service. Clients are able to use alternative services that they become aware of. However, once a valid and authenticated commitment has been received, clients SHOULD NOT use an alternative service without both reasonable assurances (see Section 3) and strong authentication. Where there is an active commitment, clients SHOULD ignore advertisements for unsecured alternative services.¶
A client MAY send requests to an unauthenticated origin in an attempt to discover potential alternative services, but these requests SHOULD be entirely generic and avoid including credentials.¶
Errors in configuration of commitments has the potential to render even the unsecured origin inaccessible for the duration of a commitment. Initial deployments are encouraged to use short duration commitments so that errors can be detected without causing the origin to become inaccessible to clients for extended periods.¶
To avoid situations where a commitment causes errors, clients MAY limit the time over which a commitment is respected for a given origin. A lower limit might be appropriate for initial commitments; the certainty that a site has set a correct value - and the corresponding limit on persistence - might increase as a commitment is renewed multiple times.¶
This specification defines the “http-opportunistic” well-known URI [RFC5785]. A client is said to have a valid http-opportunistic response for a given origin when:¶
Note that origin object lifetime might differ from the freshness lifetime of the response.¶
User Agents MUST NOT provide any special security indicia when an http resource is acquired using TLS. In particular, indicators that might suggest the same level of security as https MUST NOT be used (e.g., a “lock device”).¶
A downgrade attack against the negotiation for TLS is possible. With commitment (see Section 5), this is limited to occasions where clients have no prior information (see Section 8.3), or when persisted commitments have expired.¶
For example, because the Alt-Svc header field [RFC7838] likely appears in an unauthenticated and unencrypted channel, it is subject to downgrade by network attackers. In its simplest form, an attacker that wants the connection to remain in the clear need only strip the Alt-Svc header field from responses.¶
Downgrade attacks can be partially mitigated using the tls-commit member of the http-opportunistic well-known resource, because when it is used, a client can avoid using cleartext to contact a supporting server. However, this only works when a previous connection has been established without an active attacker present; a continuously present active attacker can either prevent the client from ever using TLS, or offer its own certificate.¶
Cached alternative services can be used to track clients over time; e.g., using a user-specific hostname. Clearing the cache reduces the ability of servers to track clients; therefore clients MUST clear cached alternative service information when clearing other origin-based state (i.e., cookies).¶
HTTP implementations and applications sometimes use ambient signals to determine if a request is for an https resource; for example, they might look for TLS on the stack, or a server port number of 443.¶
This might be due to limitations in the protocol (the most common HTTP/1.1 request form does not carry an explicit indication of the URI scheme), or it may be because how the server and application are implemented (often, they are two separate entities, with a variety of possible interfaces between them).¶
Any security decisions based upon this information could be misled by the deployment of this specification, because it violates the assumption that the use of TLS (or port 443) means that the client is accessing a HTTPS URI, and operating in the security context implied by HTTPS.¶
Therefore, servers need to carefully examine the use of such signals before deploying this specification.¶
Because this specification allows “reasonable assurances” to be established by the content of a well-known URI, servers SHOULD take suitable measures to assure that its content remains under their control. Likewise, because the Alt-Svc header field is used to describe policies across an entire origin, servers SHOULD NOT permit user content to set or modify the value of this header.¶
Mike Bishop contributed significant text to this document.¶
Thanks to Patrick McManus, Stefan Eissing, Eliot Lear, Stephen Farrell, Guy Podjarny, Stephen Ludin, Erik Nygren, Paul Hoffman, Adam Langley, Eric Rescorla, Julian Reschke, Kari Hurtta, and Richard Barnes for their feedback and suggestions.¶