mnot@mnot.net http://www.mnot.net/

Mozilla

martin.thomson@gmail.com

General Internet-Draft This describes how “http” URIs can be accessed using Transport Layer Security (TLS) to mitigate pervasive monitoring attacks.

This document describes a use of HTTP Alternative Services to decouple the URI scheme from the use and configuration of underlying encryption, allowing a “http” URI to be accessed using TLS opportunistically. Currently, “https” URIs requires acquiring and configuring a valid certificate, which means that some deployments find supporting TLS difficult. Therefore, this document describes a usage model whereby sites can serve “http” URIs over TLS without being required to support strong server authentication. A mechanism for limiting the potential for active attacks is described in . This provides clients with additional protection against them for a period after successfully connecting to a server using TLS. This does not offer the same level of protection as afforded to “https” URIs, but increases the likelihood that an active attack be detected.

The immediate goal is to make the use of HTTP more robust in the face of pervasive passive monitoring . 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 should have a minimal impact upon performance, and should not require extensive administrative effort to configure.

The key words “MUST”, “MUST NOT”, “REQUIRED”, “SHALL”, “SHALL NOT”, “SHOULD”, “SHOULD NOT”, “RECOMMENDED”, “MAY”, and “OPTIONAL” in this document are to be interpreted as described in .

An origin server that supports the resolution of HTTP URIs can indicate support for this specification by providing an alternative service advertisement for a protocol identifier that uses TLS, such as “h2” . A client that receives such an advertisement MAY direct future requests for the associated origin to the identified service (as specified by ). A client that places the importance of passive protections over performance might choose to withold requests until an encrypted connection is available. However, if such a connection cannot be successfully established, the client MAY 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. Clients with expired alternative services information could make a similar request in parallel to an attempt to contact an alternative service, to minimize the delays that might be incurred by failing to contact the alternative service.

There are no existing expectations with respect to cryptographically strong server authentication when it comes to resolving HTTP URIs. Establishing it, as described in , creates a number of operational challenges. For these reasons, server authentication is not mandatory for HTTP URIs when using the mechanism described in this specification. When connecting to an alternative service for an “http” URI, clients are required to perform the server authentication procedure described in Section 3.1 of . The server certificate, if one is proffered by the alternative service, is not necessarily checked for validity, expiration, issuance by a trusted certificate authority or matched against the name in the URI. Therefore, the alternative service MAY provide any certificate, or even select TLS cipher suites that do not include authentication. A client MAY perform additional checks on the certificate that it is offered (if the server does not select an unauthenticated TLS cipher suite). For instance, a client could examine the certificate to see if it has changed over time. In order to retain the authority properties of “http” URIs, and as stipulated by , clients MUST NOT use alternative services that identify a host other than that of the origin, unless the alternative service itself is strongly authenticated (as the origin’s host). This is not currently possible for “http” URIs on cleartext transports.

An alternative service that is discovered to support “http” URIs might concurrently support “https” URIs, because HTTP/2 permits the sending of requests for multiple origins (see ) on the one connection. Therefore, when using alternative services, both HTTP and HTTPS URIs might be sent on the same connection. “https” URIs rely on server authentication. Therefore, if a connection is initially created without authenticating the server, requests for “https” resources cannot be sent over that connection until the server certificate is successfully authenticated. Section 3.1 of describes the basic mechanism, though the authentication considerations in could 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.

Editors’ Note: this is a very rough take on an approach that would provide a limited form of protection against downgrade attack. It’s unclear at this point whether the additional effort (and modest operational cost) is worthwhile. The mechanism described in this specification is trival to mount an active attack against, for two reasons: A client that doesn’t perform authentication an easy victim of server impersonation, through man-in-the-middle attacks. A client that is willing to use cleartext to resolve the resource will do so if access to any TLS-enabled alternative services is blocked at the network layer. 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 alternative service 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 avilability, failing when it cannot be contacted. Effectively, this makes the alternative service “sticky” in the client. One drawback with this approach is that clients need to strongly authenticate the alternative service to act upon such a commitment; otherwise, an attacker could create a persistent denial of service.

A alternative service can make this commitment by sending a HTTP-TLS header field:

When it appears in a HTTP response from a strongly authenticated alternative service, this header field indicates that the availability of the origin through TLS-protected alternative services is “sticky”, and that the client MUST NOT fall back to cleartext protocols while this information is considered fresh. For example:

Note that the commitment is not bound to a particular alternative service; clients SHOULD use other alternative services that they become aware of, as long as the requirements regarding authentication and avoidance of cleartext protocols are met. When this header field appears in a response, clients MUST strongly authenticate the alternative service, as described in Section 3.1 of , noting the additional requirements in . The header field MUST be ignored if strong authentication fails. Persisted information expires after a period determined by the value of the “ma” parameter. See Section 4.2.3 of for details of determining response age.

Requests for an origin that has a persisted, unexpired value for HTTP-TLS MUST fail if they cannot be made over an authenticated TLS connection.

To avoid situations where a persisted value of HTTP-TLS causes a client to be unable to contact a site, clients SHOULD limit the time that a value is persisted for a given origin. A lower limit might be appropriate for initial observations of HTTP-TLS; the certainty that a site has set a correct value - and the corresponding limit on persistence - can increase as the value is seen more over time. Once a server has indicated that it will support authenticated TLS, a client MAY use key pinning or any other mechanism that would otherwise be restricted to use with HTTPS URIs, provided that the mechanism can be restricted to a single HTTP origin.

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., using a “lock device”).

A downgrade attack against the negotiation for TLS is possible. With the HTTP-TLS header field, this is limited to occasions where clients have no prior information (see ), or when persisted commitments have expired. For example, because the Alt-Svc header field 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. As long as a client is willing to use cleartext TCP to contact a server, these attacks are possible. The HTTP-TLS header field provides an imperfect mechanism for establishing a commitment. The advantage is that this only works if a previous connection is established where an active attacker was not present. A continuously present active attacker can either prevent the client from ever using TLS, or offer a self-signed certificate. This would prevent the client from ever seeing the HTTP-TLS header field, or if the header field is seen, from successfully validating and persisting it.

Clients that persist state for origins can be tracked over time based on their use of this information. Persisted information can be cleared to reduce the ability of servers to track clients. Clients MUST clear persisted alternative service information when clearing other origin-based state (i.e., cookies).

Harvard University

1350 Mass. Ave. Cambridge MA 02138 - +1 617 495 3864 sob@harvard.edu

General keyword In many standards track documents several words are used to signify the requirements in the specification. These words are often capitalized. This document defines these words as they should be interpreted in IETF documents. Authors who follow these guidelines should incorporate this phrase near the beginning of their document: The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119. Note that the force of these words is modified by the requirement level of the document in which they are used. This memo describes how to use Transport Layer Security (TLS) to secure Hypertext Transfer Protocol (HTTP) connections over the Internet. This memo provides information for the Internet community. This document specifies Version 1.2 of the Transport Layer Security (TLS) protocol. The TLS protocol provides communications security over the Internet. The protocol allows client/server applications to communicate in a way that is designed to prevent eavesdropping, tampering, or message forgery. [STANDARDS-TRACK] This specification describes an optimized expression of the syntax of the Hypertext Transfer Protocol (HTTP). HTTP/2 enables a more efficient use of network resources and a reduced perception of latency by introducing header field compression and allowing multiple concurrent messages on the same connection. It also introduces unsolicited push of representations from servers to clients. This document is an alternative to, but does not obsolete, the HTTP/1.1 message syntax. HTTP's existing semantics remain unchanged. This document specifies "alternative services" for HTTP, which allow an origin's resources to be authoritatively available at a separate network location, possibly accessed with a different protocol configuration. The Hypertext Transfer Protocol (HTTP) is a stateless application- level protocol for distributed, collaborative, hypertext information systems. This document defines HTTP caches and the associated header fields that control cache behavior or indicate cacheable response messages. This memo describes an extension to the HTTP protocol allowing web host operators to instruct user agents (UAs) to remember ("pin") the hosts' cryptographic identities for a given period of time. During that time, UAs will require that the host present a certificate chain including at least one Subject Public Key Info structure whose fingerprint matches one of the pinned fingerprints for that host. By effectively reducing the number of authorities who can authenticate the domain during the lifetime of the pin, pinning may reduce the incidence of man-in-the-middle attacks due to compromised Certification Authorities. This document defines the concept of an "origin", which is often used as the scope of authority or privilege by user agents. Typically, user agents isolate content retrieved from different origins to prevent malicious web site operators from interfering with the operation of benign web sites. In addition to outlining the principles that underlie the concept of origin, this document details how to determine the origin of a URI and how to serialize an origin into a string. It also defines an HTTP header field, named "Origin", that indicates which origins are associated with an HTTP request. [STANDARDS-TRACK] Pervasive monitoring is a technical attack that should be mitigated in the design of IETF protocols, where possible.

Thanks to Patrick McManus, Eliot Lear, Stephen Farrell, Guy Podjarny, Stephen Ludin, Erik Nygren, Paul Hoffman, Adam Langley, Eric Rescorla and Richard Barnes for their feedback and suggestions.