]> 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 Request Resource 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 presumes at least three parties to each exchange: the client, the proxy, and the target (formally, the Oblivious Request Resource). When used properly, Oblivious HTTP enables the client to send requests to the target in such a way that the target cannot tell whether two requests came from the same client and the proxy cannot see the contents of the requests. Oblivious HTTP's threat model assumes that at least one of the proxy and the target is acting properly, i.e. complying with the protocol and keeping certain information confidential. If either proxy or target 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 proxy. Otherwise, use of the proxy reveals the user's identity to the target.
2. The client must hold an authentic KeyConfig for the target. Otherwise, they could be speaking to the proxy, impersonating the target.
3. All users of this proxy must be equally likely to use this URI and KeyConfig for this target, regardless of their prior activity. Otherwise, the encrypted request identifies the user to the target.
4. (optional) The target must not learn the IP addresses of the clients, collectively. Otherwise, the target might be able to deanonymize requests by correlating them with external information about the clients.

This specification defines behaviors for the client, proxy, and target that achieve preconditions 2-4. (This specification does not address precondition 1.) This draft is an instantiation of the "Single Proxy Discovery" architecture for key consistency, 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 requests: one to the Proxy, and one through the Proxy to the Target. The Proxy will forward the first request to the Target if the response is not in cache.

Overview of Key-Consistency Double-Check | |<---->| | | Client | | Proxy | | Target | | |<====================>| | +--------+ +-------+ +--------+ ]]>

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

Requirements

Oblivious Request Resource The Oblivious Request Resource MUST publish an Access Description over HTTP/3 containing the ohttp.request key, e.g.: The Oblivious Request Resource MUST include a "strong validator" ETag () in any response to a GET request for this access 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 this Access 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 Proxy The Oblivious Proxy MUST publish an Access Description that includes the ohttp.proxy and udp keys, indicating support for CONNECT-UDP . It SHOULD also contain the dns key, indicating support for DNS over HTTPS , to enable the use of HTTPS records with CONNECT-UDP.

Example Proxy Access Description

The Oblivious Proxy Resources MUST allow use of the GET method to retrieve small JSON responses, and SHOULD make ample cache space available in order to cache Access Descriptions. Each proxy instance (as defined by its external-facing network interface) MUST share cache state among all clients to ensure that they use the same Access Descriptions for each Oblivious Request Resource. Oblivious Proxies MUST preserve the ETag response header on cached responses, and MUST add an Age header () to all proxied responses. Oblivious Proxies MUST respect the "Cache-Control: immutable" directive, never revalidating these cached entries, and MUST NOT accept PUSH_PROMISE frames from the target. Proxies 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 Proxies that are not intended for general-purpose proxy usage MAY impose strict transfer limits or rate limits on HTTP CONNECT and CONNECT-UDP usage.

Client The Client is assumed to know an "https" URI of an Oblivious Request Resource's Access Description. To use that Request Resource, it MUST perform the following "double-check" procedure:

1. Send a GET request to the Oblivious Proxy's template with request_uri set to the Access Description URI.
2. Record the response (A).
3. Check that response A's "Cache-Control" values indicates "public" and "immutable".
4. Fetch the Access Description URI from its origin via CONNECT-UDP, with "If-Match" set to response A's ETag.
5. Record the response (B).
6. Check that responses A and B were successful and the contents are identical, otherwise fail.

This procedure ensures that the Access 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 Access Description, and MUST repeat the "double-check" process if either response changes.

Example: Oblivious DoH In this example, the client has been configured with an Oblivious DoH server and an Oblivious Proxy. The Oblivious DoH server is identified by an Access Description at "https://doh.example.com/config.json" with the following contents: The Oblivious Proxy 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 request using HTTP/3: Having successfully fetched the Access 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 Access Description to reach the Oblivious DoH server, by forming Binary HTTP requests for "https://doh.example.com/dns-query" and delivering the encapsulated requests to "https://example.com/ohttp/" via the proxy.

Security Considerations

In scope A malicious proxy could attempt to learn the contents of the oblivious request by forging an Access Description containing its own KeyConfig. This is prevented by the client's requirement that the KeyConfig be served to it by the configured origin over HTTPS (). A malicious target could attempt to link multiple requests together by issuing each user a unique, persistent KeyConfig. This attack is prevented by the client's requirement that the KeyConfig be fresh according to the proxy's cache (). A malicious target could attempt to rotate its entry in the proxy's cache in several ways:

• Using HTTP PUSH_PROMISE frames. This attack is prevented by disabling PUSH_PROMISE at the proxy ().
• By also acting as a client and sending requests designed to replace the Access Description in the cache before it expires:
  • By sending 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 ().
  • By filling the cache with new entries, causing its previous Access Description to be evicted. describes some possible mitigations.

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 proxy and target that are authentic and widely shared. If these identities are inauthentic, or are unique to the user, 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 target. If a large fraction of the clients are malicious, they could conspire to flood the proxy cache with entries that seem popular, leading to rapid eviction of the malicious target's Access Descriptions. Similar concerns apply if a malicious target can compel naive clients to fetch a very large number of Access Descriptions.

IANA Considerations IANA is requested to open a Specification Required registry entitled "HTTP Access Service Descriptors", with the following initial contents:

Key	Specification
dns	(This document)
udp	(This document)
ip	(This document)
ohttp	(This document)

IANA is requested to add the following entry to the "Well-Known URIs" registry

URI Suffix	Change Controller	Reference	Status	Related Information
access-services	IETF	(This document)	permanent	Sub-registry at (link)

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. 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. 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. 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. Adobe Fastly greenbytes GmbH 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 over HTTP. Similar to how the CONNECT method allows proxying TCP over HTTP, this document defines a new mechanism to proxy UDP. It is built using HTTP Extended CONNECT. 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. Adobe Fastly greenbytes GmbH 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. 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.

Acknowledgments TODO acknowledge.