]> Google LLC

bemasc@google.com

sec ohai Internet-Draft The assurances provided by Oblivious HTTP depend on the client's ability to verify that it is using the same Gateway, Target, and KeyConfig as many other users. This specification defines a protocol to enable this verification. About This Document Status information for this document may be found at . Source for this draft and an issue tracker can be found at .

Introduction Oblivious HTTP identifies four parties to each exchange: the Client, Relay, Gateway and Target. When used properly, Oblivious HTTP enables the Client to send requests to the Target in such a way that the Target and Gateway cannot tell whether two requests came from the same Client, and the Relay cannot see the contents of the requests. The Target and Gateway are tightly coupled, as the Gateway can see and modify all cleartext data to and from the Target. For ease of description, we will refer to the Target and Gateway collectively (including the URI and KeyConfig for the Gateway and the URI of the Target) as the "Service". Oblivious HTTP's threat model assumes that at least one of the Relay and the Service is acting properly, i.e. complying with the protocol and keeping certain information confidential. If either Relay or Service misbehaves, the only effect must be a denial of service. In order for these security guarantees to hold, several preconditions must be met:

1. The Client must be one of many users who might be using the Relay. Otherwise, use of the Relay reveals the user's identity to the Gateway.
2. The Client must hold an authentic KeyConfig for the Gateway. Otherwise, the Client could be speaking to the Relay, impersonating the Gateway.
3. All users of the Relay must be equally likely to use this Service, regardless of their prior activity. Otherwise, the encrypted request identifies the Client to the Service.
4. (optional) The Gateway must not learn the IP addresses of the Clients, collectively. Otherwise, the Gateway might be able to deanonymize requests by correlating them with external information about the Clients.

This specification defines behaviors for the Client, Relay, and Service that achieve preconditions 2-4. (This specification does not address precondition 1.) This specification assumes that the Service is identified by an Access Description , which we call the "Service Description". For this specification to meet its goals, the Service Description's URL must have been distributed to clients in a globally consistent fashion. For example, the Service Description URL might be the default value of a software setting, or it might be published on a third party's website. This specification allows clients to convert the static, long-lived Service Description URL into a fresh Service Description without losing the privacy guarantees of Oblivious HTTP. In principle, Services could achieve a similar effect by distributing their Service Descriptions directly through this globally consistent channel. However, these ad hoc pubication channels may not be fast enough to support frequent updates (e.g., key rotations), especially if updates require user intervention. This draft combines elements of the "Direct Discovery", "Single Proxy Discovery", and "Independent Verification" strategies defined in .

Conventions and Definitions The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 when, and only when, they appear in all capitals, as shown here.

Overview In the Key Consistency Double-Check procedure, the Client emits two HTTP GET requests: one to the Relay, and one through the Relay to the Service Description Host using CONNECT-UDP. (The Service Description Host, Gateway, and Target are most commonly expected to be a single origin.) The Relay will forward the first request to the Service Description Host if the response is not in cache.

Overview of Key-Consistency Double-Check | |<----->| Service | | Client | | Relay | | Description | | |<=====================>| Host | +--------+ +-------+ +-------------+ ]]>

The Relay caches the response, ensuring that all clients share it during its freshness lifetime. The client checks this against the authenticated response from the Service Description Host, preventing forgeries.

Requirements

Oblivious Service The Oblivious Service MUST publish an Access Description containing the "ohttp.gateway" key, e.g.: This Access Description is called the Service Description, and its origin is called the Service Description Host. This origin MUST support HTTP/3 , so that it can be accessed via the proxy's CONNECT-UDP service (see ). The Service Description Host MUST include a "strong validator" ETag () in any response to a GET request for this Service Description, and MUST support the "If-Match" HTTP request header (). The response MUST indicate "Cache-Control: public, no-transform, s-maxage=(...), immutable" . For efficiency reasons, the max age SHOULD be at least 60 seconds, and preferably much longer. If the Service Description changes, and the resource receives a request whose "If-Match" header identifies a previously served version that has not yet expired, it MUST return a success response containing the previous version. This response MAY indicate "Cache-Control: private".

Oblivious Relay The Oblivious Relay MUST also provide CONNECT-UDP service , and SHOULD also offer DNS over HTTPS , to enable the use of HTTPS records with CONNECT-UDP. This corresponds to an Access Description that includes the "ohttp.relay", "udp", and "dns" keys:

Example Relay Access Service Description

The Oblivious Relay MUST allow use of the GET method to retrieve small JSON responses, and SHOULD make ample cache space available in order to avoid eviction of Service Descriptions. The Relay SHOULD share cache state among all clients, to ensure that they use the same Service Descriptions for each Oblivious Service. If the cache must be partitioned for architectural or performance reasons, operators SHOULD keep the number of users in each partition as large as possible. Oblivious Relays MUST preserve the ETag response header on cached responses, and MUST add an Age header () to all proxied responses. Oblivious Relays MUST respect the "Cache-Control: immutable" directive, and MUST NOT revalidate fresh immutable cache entries in response to any incoming requests. (Note that this is different from the general recommendation in ). Oblivious Relays also MUST NOT accept PUSH_PROMISE frames from the target. Relays SHOULD employ defenses against malicious attempts to fill the cache. Some possible defenses include:

• Rate-limiting each client's use of GET requests.
• Prioritizing preservation of cache entries that have been served to many clients, if eviction is required.

Oblivious Relays that are not intended for general-purpose proxy usage MAY impose strict transfer limits or rate limits on HTTP CONNECT and CONNECT-UDP usage. If the Relay offers a DNS over HTTPS resolver, it MUST NOT enable EDNS Client Subnet support .

Client The Client is assumed to know an "https" URI of the relevant Service Description. Before using that Service Description, it MUST perform the following "double-check" procedure:

1. Send a GET request to the Oblivious Relay's template (ohttp.proxy.template) with request_uri set to the Service Description URI.
2. Record the response (A).
3. Check that response A's "Cache-Control" values indicates "public" and "immutable".
4. Establish a CONNECT-UDP tunnel through the proxy to the Service Description URI's origin.
5. Fetch the Service Description URI through this tunnel, using a GET request with "If-Match" set to response A's ETag.
6. Record the response (B).
7. Check that responses A and B were successful and the contents are identical, otherwise fail.

This procedure ensures that the Service Description is authentic and will be shared by all users of this proxy. Once response A or B expires, the client MUST refresh it before continuing to use this Service Description, and MUST repeat the "double-check" process if either response changes. Clients MUST perform each fetch to the origin (step 4) as a fully isolated request. Any state related to this origin (e.g. cached DNS records, CONNECT-UDP tunnels, QUIC transport state, TLS session tickets, HTTP cookies) MUST NOT be shared with prior or subsequent requests.

Example: Oblivious DoH In this example, the client has been configured with an Oblivious DoH server and an Oblivious Relay. The Oblivious DoH server is identified by a Service Description at "https://doh.example.com/config.json" with the following contents: The Oblivious Relay is identified as "proxy.example.org", which implies an Access Description at "https://proxy.example.org/.well-known/access-services". This resource's contents are: The following exchanges then occur between the client and the proxy: The client now has a CONNECT-UDP tunnel to doh.example.com, over which it performs the following GET request using HTTP/3: Having successfully fetched the Service Description from both locations, the client confirms that:

• The responses are identical.
• The cache-control response from the proxy contained the "public" and "immutable" directives.

The client can now use the KeyConfig in this Service Description to reach the Oblivious DoH server, by converting DNS-over-HTTPS requests into Binary HTTP requests for "https://doh.example.com/dns-query", encrypting them to ohttp.gateway.key, and POSTing the encrypted requests to "https://relay.example.org/ohttp?request_uri=https%3A%2F%2Fexample.com%2Fohttp%2F".

Performance Implications

Latency Suppose that the Client-Relay Round-Trip Time (RTT) is A, and the Relay-Service Description Host RTT is B. Suppose additionally that the Client has a persistent connection to the Relay that is already running. Then the procedure described in requires:

• A for the GET request to the Relay
  • +B if the requested Service Description is not in cache
  • +B if the Relay does not have a TLS session ticket for the Service Description Host
• A for the CONNECT-UDP request to the Relay
• A + B for the QUIC handshake to the Service Description Host
• A + B for the GET request to the Service Description Host

This is a total of 4A + 4B in the worst case. However, clients can reduce the latency by issuing the requests to the Relay in parallel, and by using CONNECT-UDP's "false start" support. The Service Description Host can also optimize performance, by issuing long-lived TLS session tickets. With these optimizations, the expected total time is 2A + 2B. This procedure only needs to be repeated if the Service Description has expired. To enable regular key rotation and operational adjustments, a cache lifetime of 24 hours may be suitable. Clients MAY perform this procedure in advance of an expiration to avoid a delay.

Thundering Herds All clients of the same Relay and Service will have locally cached Service Descriptions with the same expiration time. When this entry expires, all active clients will send refresh GET requests to the proxy at their next request. Relays SHOULD use "request coalescing" to avoid duplicate cache-refresh requests to the target. If the Service Description has changed, these clients will initiate GET requests through the Relay to the Service Description Host to double-check the new contents. Relays and Service Description Hosts MAY use an HTTP 503 response with a "Retry-After" header to manage load spikes.

Security Considerations

In scope

Forgery A malicious Relay could attempt to learn the contents of the Oblivious HTTP request by forging a Service Description containing its own KeyConfig. This is prevented by the client's requirement that the KeyConfig be served to it by the Service Description Host over HTTPS ().

Deanonymization A malicious Service could attempt to link multiple Oblivious HTTP requests together by issuing each Client a unique, persistent KeyConfig. This attack is prevented by the Client's requirement that the KeyConfig be fresh according to the Relay's cache (). A malicious Service could attempt to rotate its entry in the Relay's cache in several ways:

• Using HTTP PUSH_PROMISE frames. This attack is prevented by disabling PUSH_PROMISE at the Relay ().
• By also acting as a Client and sending requests designed to replace the Service Description in the cache before it expires:
  • By sending GET requests with a "Cache-Control: no-cache" or similar directive. This is prevented by the response's "Cache-Control: public, immutable" directives, which are verified by the Client (), and by the Relay's obligation to to respect these directives strictly ().
  • By filling the cache with new entries, causing its previous Service Description to be evicted. describes some possible mitigations.

A malicious Service could attempt to link different requests for the Service Description, in order to link the Oblivious HTTP requests that follow shortly after. This is prevented by fully isolating each request (), and by disabling EDNS Client Subnet ().

Abusive traffic A malicious client could use the proxy to send abusive traffic to any destination on the internet. Abuse concerns can be mitigated by imposing a rate limit at the proxy ().

Out of scope This specification assumes that the client starts with identities of the Relay and Service that are authentic and widely shared. If these identities are inauthentic, or are unique to the client, then the security goals of this specification are not achieved. This specification assumes that at most a small fraction of Clients are acting on behalf of a malicious Service. If a large fraction of the Clients are malicious, they could conspire to flood the Relay's cache with entries that seem popular, leading to rapid eviction of the malicious Service's Service Descriptions. Similar concerns apply if a malicious Service can compel naive Clients to fetch a very large number of Service Descriptions. A Client's requests for the Service Description may become linkable if they have distinctive QUIC Initials, HTTP/3 Settings, RTT, or other protocol features observable through the Relay. This specification does not offer specific mitigations for protocol fingerprinting.

IANA Considerations No IANA action is requested.

References Normative References Mozilla Cloudflare This document describes a system for the forwarding of encrypted HTTP messages. This allows a client to make multiple requests of a server without the server being able to link those requests to the client or to identify the requests as having come from the same client. Google LLC HTTP proxies can operate several different kinds of access services. This specification provides a format for identifying a collection of such services. About This Document This note is to be removed before publishing as an RFC. Status information for this document may be found at https://datatracker.ietf.org/doc/draft-schwartz-masque-access- descriptions/. Source for this draft and an issue tracker can be found at https://github.com/bemasc/access-services. 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. This document specifies an Internet Best Current Practices for the Internet Community, and requests discussion and suggestions for improvements. RFC 2119 specifies common key words that may be used in protocol specifications. This document aims to reduce the ambiguity by clarifying that only UPPERCASE usage of the key words have the defined special meanings. The QUIC transport protocol has several features that are desirable in a transport for HTTP, such as stream multiplexing, per-stream flow control, and low-latency connection establishment. This document describes a mapping of HTTP semantics over QUIC. This document also identifies HTTP/2 features that are subsumed by QUIC and describes how HTTP/2 extensions can be ported to HTTP/3. The Hypertext Transfer Protocol (HTTP) is a stateless application- level protocol for distributed, collaborative, hypertext information systems. This document defines HTTP/1.1 conditional requests, including metadata header fields for indicating state changes, request header fields for making preconditions on such state, and rules for constructing the responses to a conditional request when one or more preconditions evaluate to false. 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 document obsoletes RFC 7234. The immutable HTTP response Cache-Control extension allows servers to identify resources that will not be updated during their freshness lifetime. This ensures that a client never needs to revalidate a cached fresh resource to be certain it has not been modified. Google LLC This document describes how to proxy UDP in HTTP, similar to how the HTTP CONNECT method allows proxying TCP in HTTP. More specifically, this document defines a protocol that allows an HTTP client to create a tunnel for UDP communications through an HTTP server that acts as a proxy. This document defines a protocol for sending DNS queries and getting DNS responses over HTTPS. Each DNS query-response pair is mapped into an HTTP exchange. This document describes an Extension Mechanisms for DNS (EDNS0) option that is in active use to carry information about the network that originated a DNS query and the network for which the subsequent response can be cached. Since it has some known operational and privacy shortcomings, a revision will be worked through the IETF for improvement. Informative References Brave Software The Tor Project Mozilla Cloudflare This document describes the key consistency and correctness requirements of protocols such as Privacy Pass, Oblivious DoH, and Oblivious HTTP for user privacy. It discusses several mechanisms and proposals for enabling user privacy in varying threat models. In concludes with discussion of open problems in this area. Google Akamai Technologies Akamai Technologies This document specifies the "SVCB" and "HTTPS" DNS resource record (RR) types to facilitate the lookup of information needed to make connections to network services, such as for HTTP origins. SVCB records allow a service to be provided from multiple alternative endpoints, each with associated parameters (such as transport protocol configuration and keys for encrypting the TLS ClientHello). They also enable aliasing of apex domains, which is not possible with CNAME. The HTTPS RR is a variation of SVCB for use with HTTP [HTTP]. By providing more information to the client before it attempts to establish a connection, these records offer potential benefits to both performance and privacy. TO BE REMOVED: This document is being collaborated on in Github at: https://github.com/MikeBishop/dns-alt-svc (https://github.com/MikeBishop/dns-alt-svc). The most recent working version of the document, open issues, etc. should all be available there. The authors (gratefully) accept pull requests.

Acknowledgments TODO acknowledge.