title: Caching Secure HTTP Content using Blind Caches abbrev: Blind Cache docname: draft-thomson-http-bc date: 2016 category: std ipr: trust200902
stand_alone: yes pi: [toc, sortrefs, symrefs, docmapping]
ins: M. Thomson
name: Martin Thomson
org: Mozilla
email: martin.thomson@gmail.com
- ins: G. Eriksson name: Göran AP Eriksson org: Ericsson email: goran.ap.eriksson@ericsson.com
- ins: C. Holmberg name: Christer Holmberg org: Ericsson email: christer.holmberg@ericsson.com
normative: RFC2119: RFC2818: RFC7230: RFC7234: RFC7540: SCD: title: An Architecture for Secure Content Delegation using HTTP author: - ins: G. Ericsson - ins: C. Holmberg - ins: M. Thomson date: 2016-02-07 I-D.ietf-httpbis-encryption-encoding: I-D.thomson-http-mice:
informative: RFC7231: I-D.reschke-http-oob-encoding: HINTS: title: "Resource Hints" author: - ins: I. Grigorik date: 2015-05-27 seriesinfo: W3C TR
--- abstract
A mechanism is described whereby a server can use client-selected shared cache.
--- middle
Shared caches allow an HTTP server to offload the responsibility for delivering certain content. Content in the shared cache can be accessed efficiently by multiple clients, saving the origin server from having to serve those requests and ensuring that clients receive responses to cached requests more quickly.
Proxy caching is the most common configuration for shared caching. A proxy cache is either explicitly configured by a client, discovered as a result of being automatically configured.
HTTPS [RFC2818] prevents the use of proxies by creating an authenticated end-to-end connection to the origin server or its gateway that is authenticated. This provides a critical protection against man-in-the-middle attacks, but it also prevents the proxy from acting as a shared cache.
Clients do not direct queries for https URIs to proxies. Clients configured
with a proxy use the CONNECT pseudo-method (Section 4.3.6 of [RFC7231]) with any
explicitly configured or discovered proxies to create an end-to-end tunnel.
Transparent proxies are unable to intercept connections that are protected with
TLS.
This document describes a method that enables shared caching for a limited set
of https resources, as selected by the server. The server conditionally
delegates the hosting of secure content to itself. This delegation includes a
marker that signals permission for a client to send a request for an https
resource via a proxy rather than insisting on an end-to-end TLS connection.
The words "MUST", "MUST NOT", "SHOULD", and "MAY" are used in this document. It's not shouting; when they are capitalized, they have the special meaning defined in [RFC2119].
This document uses the term "proxy cache" to refer to a proxy [RFC7230] that operates an HTTP cache [RFC7234].
The secure content delegation mechanism defined in [SCD] is used to create a separate resource that contains encrypted and integrity protected content. To enable caching, the primary and secondary servers can be the same server.
A client that signals a willingness to support delegation is provided with a response that uses a proxy-enabled out-of-band encoding that behaves identically to the out-of-band encoding defined in [I-D.reschke-http-oob-encoding]. The out-of-band encoding identifies a secondary resource and implicitly indicates that the client is willing to use a proxy and that the server allows this use. The client is then able to request the secondary resource from a proxy cache rather than directly to the origin server.
In this document, the origin server is able to act in the role of the secondary server in [SCD]. However, all of the considerations that apply to having a secondary server host content apply instead to the proxy cache. Thus, integrity and confidentiality protections against the proxy cache are the primary consideration.
Without a clear signal from the client that a caching proxy is present, an
origin server is unable to send a response with out-of-band encoding. A value
of out-of-band in the Accept-Encoding header field only indicates
willingness to use the secure content delegation mechanism.
A new oobp content encoding is defined. The oobp content encoding is
identical to the out-of-band content encoding, with the following additional
conditions:
-
A client MUST NOT signal support for
oobpcontent encoding unless it is using a proxy cache and it is willing to direct requests to that proxy. -
A server MUST NOT encode a response using the
oobpcontent encoding unless it permits the request to be made to a proxy cache. -
The
oobpcontent encoding MUST NOT be used to encode the contents of a request. Theout-of-bandcontent encoding is sufficient for that purpose.
Using a different content encoding name means that a resource using secure content delegation to a secondary server [SCD] does not inadvertently trigger a request via a proxy.
The security properties of delegation via a secondary server and via a caching proxy are similar only to the extent that a third party is involved. However, it might be the case that a secondary server has a stronger relationship with the primary server and additional constraints on its actions, such as contractual limitations. Such constraints might make it feasible to delegate to a secondary server selected by the primary server. A caching proxy might not be considered acceptable in the same way.
The oobp content encoding clearly indicates that the client is permitted to
retrieve content from a proxy-cache.
Servers that use the oobp content encoding MUST include header fields for
message integrity and encryption, such as the M-I header field
[I-D.thomson-http-mice] or the Crypto-Key header field
[I-D.ietf-httpbis-encryption-encoding]. Clients MUST NOT send a request via a
proxy if these headers are not present. Absence of these header fields indicate
an error on the part of the origin server, since integrity and confidentiality
protection are mandatory.
This mechanism does not work with a transparent caching proxy. Since the request is made over end-to-end HTTPS in the absence of a proxy, the feature will not be used unless the proxy is known to the client.
A proxy cache MUST therefore be expressly configured or discovered. This produces a name and possibly a port number for the proxy. The proxy MUST be contacted using HTTPS [RFC2818] and authenticated using the configured or discovered domain name.
Issue:
: What signal do we need from the proxy cache that it supports receiving requests with an https:// scheme? Can we expect that a proxy cache will happily accept a request for an HTTPS URL? What if they ignore the scheme and send the request in the clear?
As noted in [SCD], the secondary request required by out-of-band content encoding imposes a performance penalty. This can be mitigated by priming clients with information about the location and disposition of resources prior to the client making a request. A resource map described in [SCD] might be provided to clients to eliminate the latency involved in making requests of the origin server for resources that might be cached.
A client that makes a request of an origin server via an unprimed proxy cache will suffer additional latency as a consequence of the cache having to make a request to the origin server.
The following options are possible:
-
Clients can speculatively make requests to a proxy cache based on information it learns from a resource map, or from hints like the "prefetch" link relation [HINTS]. To avoid a potential waste of resources as a result of receiving complete responses, speculative requests might be limited to HEAD requests; alternatively, HTTP/2 [RFC7540] flow control might be used to allow only limited information to be sent.
-
The origin server might provide the proxy cache with "prefetch" link relations in responses to requests for secondary resources. These link relations might identify other resources that the proxy might retrieve speculatively. This does not improve the latency of the initial request, but could improve subsequent requests.
All the considerations of [SCD] apply. In particular, content that is distributed with the assistance of a proxy cache MUST include integrity and confidentiality protection. That means that the M-I header field [I-D.thomson-http-mice] and the Crypto-Key header field [I-D.ietf-httpbis-encryption-encoding] or equivalent information MUST be present in responses that include an out-of-band content encoding.
Clients that receive a response without the information necessary to ensure integrity and confidentiality protection against a proxy cache MUST NOT make a request to a proxy to retrieve that response. Clients could treat such a response as failed. Clients MAY then make the request directly to the origin server, or - if request can be safely retried - retry a request without the out-of-band token in the Accept-Encoding header field.
This document has no IANA actions. It should.