HTTP Working GroupR. Fielding
Internet-DraftAdobe Systems Incorporated
Intended status: Standards TrackM. Nottingham
Expires: August 10, 2024February 7, 2024

The Key HTTP Response Header Field

Abstract

The 'Key' header field for HTTP responses allows an origin server to describe the secondary cache key (RFC 7234, Section 4.1) for a resource, by conveying what is effectively a short algorithm that can be used upon later requests to determine if a stored response is reusable for a given request.

Key has the advantage of avoiding an additional round trip for validation whenever a new request differs slightly, but not significantly, from prior requests.

Key also informs user agents of the request characteristics that might result in different content, which can be useful if the user agent is not sending request header fields in order to reduce the risk of fingerprinting.

Note to Readers

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/key.

Status of this Memo

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 https://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 August 10, 2024.

Copyright Notice

Copyright (c) 2024 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 (https://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 Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.


1. Introduction

In HTTP caching [RFC7234], the Vary response header field effectively modifies the key used to store and access a response to include information from the request's headers. This "secondary cache key" allows proactive content negotiation [RFC7231] to work with caches.

Vary's operation is generic; it works well when caches understand the semantics of the selecting headers. For example, the Accept-Language request header field has a well-defined syntax for expressing the client's preferences; a cache that understands this header field can select the appropriate response (based upon its Content-Language header field) and serve it to a client, without any knowledge of the underlying resource.

Vary does not work as well when the criteria for selecting a response are specific to the resource. For example, if the nature of the response depends upon the presence or absence of a particular Cookie ([RFC6265]) in a request, Vary doesn't have a mechanism to offer enough fine-grained, resource-specific information to aid a cache's selection of the appropriate response.

Additionally, when new selecting headers are defined, caches need to be updated to understand their semantics before Vary can operate over them efficiently; due to the sometimes slow rate of cache deployment, this can be problematic.

Finally, Vary has proven to be difficult to implement correctly and efficiently by high-performance intermediary caches, because doing so involves examining all cached responses with the request's URL.

This document defines a new response header field, "Key", that allows resources to describe the secondary cache key in a fine-grained, resource-specific manner, leading to improved cache efficiency when responses depend upon such headers.

1.1. Examples

For example, this response header field:

  Key: cookie;param=_sess;param=ID

indicates that the selected response depends upon the "_sess" and "ID" cookie values.

This Key:

  Key: user-agent;substr=MSIE

indicates that there are two possible secondary cache keys for this resource; one for requests whose User-Agent header field contains "MSIE", and another for those that don't.

A more complex example:

  Key: user-agent;substr=MSIE;Substr="mobile", Cookie;param="ID"

indicates that the selected response depends on the presence of two strings in the User-Agent request header field, as well as the value of the "ID" cookie request header field.

1.2. Notational Conventions

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 [RFC2119].

This document uses the Augmented Backus-Naur Form (ABNF) notation of [RFC5234] (including the DQUOTE rule), and the list rule extension defined in [RFC7230], Section 7. It includes by reference the field-name, quoted-string and quoted-pair rules from that document, the OWS rule from [RFC7230] and the parameter rule from [RFC7231].

2. The "Key" Response Header Field

The "Key" response header field describes the portions of the request that the resource currently uses to select representations.

As such, its semantics are similar to the "Vary" response header field, but it allows more fine-grained description, using "key parameters".

Caches can use this information as part of determining whether a stored response can be used to satisfy a given request. When a cache knows and fully understands the Key header field for a given resource, it MAY ignore the Vary response header field in any stored responses for it.

Additionally, user agents can use Key to discover if additional request header fields might influence the resource's selection of responses.

The Key field-value is a comma-delimited list of selecting header fields (similar to Vary), with zero to many parameters each, delimited by semicolons.

  Key       = 1#key-value
  key-value = field-name *( OWS ";" OWS parameter )

Note that, as per [RFC7231], parameter names are case-insensitive, and parameter values can be double-quoted strings (potentially with "\"-escaped characters inside).

The following header fields have the same effect:

  Vary: Accept-Encoding, Cookie
  Key: Accept-Encoding, Cookie

However, Key's use of parameters allows:

  Key: Accept-Encoding, Cookie; param=foo

to indicate that the secondary cache key depends upon the Accept-Encoding header field and the "foo" Cookie.

One important difference between Vary and Key is how they are applied. Vary is specified to be specific to the response it occurs within, whereas Key is specific to the resource (as identified by the request URL) it is associated with. The most recent key you receive for a given resource is applicable to all responses from that resource.

This difference allows more efficient implementation (and reflects practices that many caches use in implementing Vary already).

This specification defines a selection of Key parameters to address common use cases such as selection upon individual Cookie header fields, User-Agent substrings and numerical ranges. Future parameters may define further capabilities.

2.1. Relationship with Vary

Origin servers SHOULD still send Vary when using Key, to ensure backwards compatibility.

For example,

  Vary: User-Agent
  Key: User-Agent;substr="mozilla"

Note that, in some cases, it may be better to explicitly use "Vary: *" if clients and caches don't have any practical way to use the Vary header field's value. For example,

  Vary: *
  Key: Cookie;param="ID"

Except when Vary: * is used, the set of headers used in Key SHOULD reflect the same request header fields as Vary does, even if they don't have parameters. For example,

  Vary: Accept-Encoding, User-Agent
  Key: Accept-Encoding, User-Agent;substr="mozilla"

Here, Accept-Encoding is included in Key without parameters; caches MAY treat these as they do values in the Vary header, relying upon knowledge of their generic semantics to select an appropriate response.

2.2. Calculating a Secondary Cache Key

When used by a cache to determine whether a stored response can be used to satisfy a presented request, each field-name in Key identifies a potential request header, just as with the Vary response header field.

However, each of these can have zero to many key parameters that change how the response selection process (as defined in [RFC7234], Section 4.3) works.

In particular, when a cache fully implements this specification, it creates a secondary cache key for every request by following the instructions in the Key header field, ignoring the Vary header for this purpose.

Then, when a new request is presented, the secondary cache key generated for that request can be compared to the stored one to find the appropriate response, to determine if it can be selected.

To generate a secondary cache key for a given request (including that which is stored with a response) using Key, the following steps are taken:

1)
If the Key header field is not present on the most recent cacheable (as per [RFC7234], Section 3)) response seen for the resource, abort this algorithm (i.e., fall back to using Vary to determine the secondary cache key).
2)
Let key_value be the result of Creating a Header Field Value (Section 2.2.1) with key as the target_field_name and the most recently seen response header list for the resource as header_list.
3)
Let secondary_key be an empty string.
4)
Create key_list by splitting key_value on "," characters, excepting "," characters within quoted strings, as per [RFC7230], Section 3.2.6.
5)
For key_item in key_list:
1)
Remove any leading and trailing WSP from key_item.
2)
If key_item does not contain a ";" character, fail parameter processing (Section 2.2.2) and skip to the next key_item.
3)
Let field_name be the string before the first ";" character in key_item, removing any WSP between them.
4)
Let field_value be the result of Creating a Header Field Value (Section 2.2.1) with field_name as the target_field_name and the request header list as header_list.
5)
Let parameters be the string after the first ";" character in key_item, removing any WSP between them.
6)
Create param_list by splitting parameters on ";" characters, excepting ";" characters within quoted strings, as per [RFC7230], Section 3.2.6.
7)
For parameter in param_list:
1)
If parameter does not contain a "=", fail parameter processing (Section 2.2.2) and skip to the next key_item.
2)
Remove any WSP at the beginning and/or end of parameter.
3)
Let param_name be the string before the first "=" character in parameter, case-normalized to lowercase.
4)
If param_name does not identify a Key parameter processing algorithm that is implemented, fail parameter processing (Section 2.2.2) and skip to the next key_item.
5)
Let param_value be the string after the first "=" character in parameter.
6)
If the first and last characters of param_value are both DQUOTE:
1)
Remove the first and last characters of param_value.
2)
Replace quoted-pairs within param_value with the octet following the backslash, as per [RFC7230], Section 3.2.6.
7)
If param_value does not conform to the syntax defined for it by the parameter definition, fail parameter processing (Section 2.2.2) and skip to the next key_item.
8)
Run the identified processing algorithm on field_value with the param_value, and append the result to secondary_key. If parameter processing fails (Section 2.2.2), skip to the next key_item.
9)
Append a separator character (e.g., NULL) to secondary_key.
6)
Return secondary_key.

Note that this specification does not require that exact algorithm to be implemented. However, implementations' observable behavior MUST be identical to running it. This includes parameter processing algorithms; implementations MAY use different internal artefacts for secondary cache keys, as long as the results are the same.

Likewise, while the secondary cache key associated with both stored and presented requests is required to use the most recently seen Key header field for the resource in question, this can be achieved using a variety of implementation strategies, including (but not limited to):

  • Generating a new secondary cache key for every stored response associated with the resource upon each request.
  • Caching the secondary cache key with the stored request/response pair and re-generating it when the Key header field is observed to change.
  • Caching the secondary cache key with the stored response and invalidating the stored response(s) when the Key header field is observed to change.

2.2.1. Creating a Header Field Value

Given a header field name target_field_name and header_list, a list of (field_name, field_value) tuples:

1)
Let target_field_values be an empty list.
2)
For each (field_name, field_value) tuple in header_list:
1)
If field_name does not match target_field_name, skip to the next tuple.
2)
Strip leading and trailing WSP from field_value and append it to target_field_values.
3)
If target_field_values is empty, return an empty string.
4)
Return the concatenation of target_field_values, separating each with "," characters.

2.2.2. Failing Parameter Processing

In some cases, a key parameter cannot determine a secondary cache key corresponding to its nominated header field value. When this happens, Key processing needs to fail safely, so that the correct behavior is observed.

When this happens, implementations MUST either behave as if the whole Key header field was not present (i.e., abort processing of Key, ignoring it entirely and falling back to Vary), or assure that the nominated header fields being compared match, as per [RFC7234], Section 4.1 (i.e., treat the failing portion of the Key header as falling back to Vary, but continuing to process the rest).

2.3. Key Parameters

A Key parameter associates a name with a specific processing algorithm that takes two inputs; a HTTP header value "header_value" (as described in Section 2.2.1), and "parameter_value", a string that indicates how the identified header should be processed.

The set of key parameters (and their associated processing algorithms) is extensible; see Section 3. This document defines the following key parameters:

2.3.1. div

The "div" parameter normalizes positive integer header values into groups by dividing them by a configured value.

Its value's syntax is:

div    = 1*DIGIT

To process a set of header fields against a div parameter, follow these steps (or their equivalent):

1)
If parameter_value is "0", fail parameter processing (Section 2.2.2).
2)
If header_value is the empty string, return "none".
3)
If header_value contains a ",", remove it and all subsequent characters.
4)
Remove all WSP characters from header_value.
5)
If header_value does not match the div ABNF rule, fail parameter processing (Section 2.2.2).
6)
Return the quotient of header_value / parameter_value (omitting the modulus).

For example, the Key:

Key: Bar;div=5

indicates that the "Bar" header's field value should be partitioned into groups of 5. Thus, the following field values would be considered the same (because, divided by 5, they all result in 0):

Bar: 1
Bar: 3 , 42
Bar: 4, 1

whereas these would be considered to be in a different group (because, divided by 5, they all result in 2);

Bar: 12
Bar: 10
Bar: 14, 1

2.3.2. partition

The "partition" parameter normalizes positive numeric header values into pre-defined segments.

Its value's syntax is:

partition = [ segment ] *( ":" [ segment ] )
segment   = [ 0*DIGIT "." ] 1*DIGIT

To process a set of header fields against a partition parameter, follow these steps (or their equivalent):

1)
If header_value is the empty string, return "none".
2)
If header_value contains a ",", remove it and all subsequent characters.
3)
Remove all WSP characters from header_value.
4)
If header_value does not match the segment ABNF rule, fail parameter processing (Section 2.2.2).
5)
Let segment_id be 0.
6)
Create a list segment_list by splitting parameter_value on ":" characters.
7)
For each segment_value in segment_list:
1)
If header_value is less than segment_value when they are numerically compared, skip to step 7.
2)
Increment segment_id by 1.
8)
Return segment_id.

For example, the Key:

Key: Foo;partition=20:30:40

indicates that the "Foo" header's field value should be divided into four segments:

  • less than 20
  • 20 to less than 30
  • 30 to less than 40
  • forty or greater

Thus, the following headers would all be normalized to the first segment:

Foo: 1
Foo: 0
Foo: 4, 54
Foo: 19.9

whereas the following would fall into the second segment:

Foo: 20
Foo: 29.999
Foo:  24   , 10

2.3.3. match

The "match" parameter is used to determine if an exact value occurs in a list of header values. It is case-sensitive.

Its value's syntax is:

match  = ( token / quoted-string )

To process a set of header fields against a match parameter, follow these steps (or their equivalent):

1)
If header_value is the empty string, return "none".
2)
Create header_list by splitting header_value on "," characters.
3)
For each header_item in header_list:
1)
Remove leading and trailing WSP characters in header_item.
2)
If the value of header_item is character-for-character identical to parameter_value, return "1".
4)
Return "0".

For example, the Key:

Key: Baz;match="charlie"

Would return "1" for the following header field values:

Baz: charlie
Baz: foo, charlie
Baz: bar, charlie     , abc

and "0" for these:

Baz: theodore
Baz: joe, sam
Baz: "charlie"
Baz: Charlie
Baz: cha rlie
Baz: charlie2

2.3.4. substr

The "substr" parameter is used to determine if a value occurs as a substring of an item in a list of header values. It is case-sensitive.

Its value's syntax is:

substr  = ( token / quoted-string )

To process a set of header fields against a substr parameter, follow these steps (or their equivalent):

1)
If header_value is the empty string, return "none".
2)
Create header_list by splitting header_value on "," characters.
3)
For each header_item in header_list:
1)
Remove leading and trailing WSP characters in header_item.
2)
If the value of parameter_value is character-for-character present as a substring of header_value, return "1".
4)
Return "0".

For example, the Key:

Key: Abc;substr=bennet

Would return "1" for the following header field values:

Abc: bennet
Abc: foo, bennet
Abc: abennet00
Abc: bar, 99bennet     , abc
Abc: "bennet"

and "0" for these:

Abc: theodore
Abc: joe, sam
Abc: Bennet
Abc: Ben net

2.3.5. param

The "param" parameter considers the request header field as a list of key=value parameters, and uses the nominated key's value as the secondary cache key.

Its value's syntax is:

param  = ( token / quoted-string )

To process a list of header fields against a param parameter, follow these steps (or their equivalent):

1)
Let header_list be an empty list.
2)
Create header_list_tmp1 by splitting header_value on "," characters.
3)
For each header_item_tmp1 in header_list_tmp1:
1)
Create header_list_tmp2 by splitting header_item_tmp1 on ";" characters.
2)
For each header_item_tmp2 in header_list_tmp2:
1)
Remove leading and trailing WSP from header_item_tmp2.
2)
Append header_item_tmp2 to header_list.
4)
For each header_item in header_list:
1)
If the "=" character does not occur within header_item, skip to the next header_item.
2)
Let item_name be the string occurring before the first "=" character in header_item.
3)
If item_name does not case-insensitively match parameter_value, skip to the next header_item.
4)
Return the string occurring after the first "=" character in header_item.
5)
Return the empty string.

Note that steps 2 and 3 accommodate semicolon-separated values, so that it can be used with the Cookie request header field.

For example, the Key:

Key: Def;param=liam

The following headers would return the string (surrounded in single quotes) indicated:

Def: liam=123           // '123'
Def: mno=456            // ''
Def:                    // ''
Def: abc=123; liam=890  // '890'
Def: liam="678"         // '"678"'

3. IANA Considerations

This specification defines the HTTP Key Parameter Registry, maintained at http://www.iana.org/assignments/http-parameters/http-parameters.xhtml#key.

3.1. Procedure

Key Parameter registrations MUST include the following fields:

  • Parameter Name: [name]
  • Reference: [Pointer to specification text]

Values to be added to this namespace require IETF Review (see Section 4.1 of [RFC5226]) and MUST conform to the purpose of content coding defined in this section.

3.2. Registrations

This specification makes the following entries in the HTTP Key Parameter Registry:

Parameter NameReference
divSection 2.3.1
partitionSection 2.3.2
matchSection 2.3.3
substrSection 2.3.4
paramSection 2.3.5

4. Security Considerations

Because Key is an alternative to Vary, it is possible for caches to behave differently based upon whether they implement Key. Likewise, because support for any one Key parameter is not required, it is possible for different implementations of Key to behave differently. In both cases, an attacker might be able to exploit these differences.

This risk is mitigated by the requirement to fall back to Vary when unsupported parameters are encountered, coupled with the requirement that servers that use Key also include a relevant Vary header.

An attacker with the ability to inject response headers might be able to perform a cache poisoning attack that tailors a response to a specific user (e.g., by Keying to a Cookie that's specific to them). While the attack is still possible without Key, the ability to tailor is new.

When implemented, Key might result in a larger number of stored responses for a given resource in caches; this, in turn, might be used to create an attack upon the cache itself. Good cache replacement algorithms and denial of service monitoring in cache implementations are reasonable mitigations against this risk.

5. References

Appendix A. Acknowledgements

Thanks to Ilya Grigorik, Amos Jeffries and Yoav Weiss for their feedback.

Appendix B. Changes

B.1. Since -00

  • Issue 108 (field-name cardinality) closed with no action.
  • Issue 104 (Support "Or" operator) closed with no action.
  • Issue 107 (Whitespace requirement) addressed by allowing whitespace around parameters.
  • Issue 106 (Policy for Key parameter registry) closed with no action.

B.2. Since -01

None yet.

Authors' Addresses

Roy T. Fielding
Adobe Systems Incorporated
EMail: fielding@gbiv.com
URI: http://roy.gbiv.com/
Mark Nottingham
EMail: mnot@mnot.net
URI: https://www.mnot.net/