By submitting this Internet-Draft, each author represents that any applicable patent or other IPR claims of which he or she is aware have been or will be disclosed, and any of which he or she becomes aware will be disclosed, in accordance with Section 6 of BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet-Drafts.
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”.
The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/1id-abstracts.txt.
The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html.
This Internet-Draft will expire on August 31, 2006.
Copyright © The Internet Society (2006). All Rights Reserved.
This document discusses design considerations for state identifiers in the Hypertext Transfer Protocol (HTTP) and related protocols such as WebDAV.
Discussion of this draft and comments to the editors should be sent to the email@example.com mailing list, which is archived at <http://lists.w3.org/Archives/Public/ietf-http-wg/>.
In distributed systems such as the World Wide Web, it is useful to assign unique identifiers to individual states of network resources. The basic idea is that each time the state of a network resource changes, its associated state identifier changes too. State identifiers have the quality that they uniquely identify a particular state of a network resource; only one identifier will ever be associated with a given network resource state. These state identifiers can be used to support caching and remote authoring of network resources.¶
In caching, each cache locally stores a state identifier along with its cached copy of a network resource, and uses the state identifier to query whether the cached copy is up-to-date. If the cache's local state identifier is different from the current state identifier of the original network resource, it indicates the cached copy is stale, and needs to be refreshed. If the cached and original state identifiers are the same, it indicates the local copy is up-to-date.¶
In remote authoring, a remote authoring tool wishes to update the state of a network resource. A typical authoring session involves retrieving the current state of the network resource, some editing, then writing the new state of the network resource back to its original location. There are two concerns during remote authoring. One is that another author might try to modify the same network resource at the same time, leading to the lost update problem. State identifiers are used to detect this problem. The authoring application stores a local copy of the network resource's state identifier at the beginning of the authoring session. When the application goes to write the new resource state, it compares the local state identifier with the current state identifier of the network resource. If they are the same, the network resource has not been modified, and the write can proceed without danger of overwrite. The second concern is the remote authoring client wants to know that the network resource has been stored in the same form it was submitted. There are many document formats, such as XML, that can remain semantically equivalent in the face of multiple kinds of changes to the actual octets stored. The authoring application would like to know if it can continue to use its local copy of the network resource, or if it instead needs to reload its local copy from the original.¶
Version 1.1 of the Hypertext Transfer Protocol (HTTP) [RFC2616] supports two identifiers, the Entity Tag (Etag) and the Content-MD5 hash (MD5-hash). Etags are used in HTTP 1.1 for caching, and the Web Distributed Authoring and Versioning (WebDAV) authoring protocol uses Etags in conjunction with locks to avoid the lost update problem and keep local client state in synch with the authoring server. The MD5-hash is used to perform end-to-end message integrity checks. A key difference between Etags and MD5-hashes is the Etag is not required to be computable from the contents of the on-the-wire representation of a resource, while an MD5-hash is. A consquence is that Etags are less computationally inexpensive to produce than MD5-hashes.¶
Within HTTP, there are no provisions for directly interacting with the state of a network resource. Instead, clients can retrieve representations of a network resource using the GET method, and can write representations using PUT. The representation retrieved using GET can be the result of an arbitrary computational process, or can be the result of applying a wide range of transformations to a persistently stored resource. Similarly, a server may apply a range of transformations to a representation submitted with PUT before creating a persistently stored resource. A resource representation on the wire is known as an entity, and both Etags and MD5-hashes are unique identifiers for this entity.¶
In the most general case, Etags and MD5-hashes are not state identifiers, since they uniquely identify only for the on-the-wire representation of a resource, and do not necessarily identify the actual resource state. Etags and MD5-hashes are entity identifiers. For caching, this is usually acceptable, since the cache only wants to ensure that the client-visible representation of the resource is maintained up-to-date.¶
For authoring, the distinction between entity identifiers and state identifiers is problematic. Since authoring applications modify the state of the network resource, the information provided by entity identifiers does not provide sufficient feedback on the progress of authoring applications. Changes to the state of a resource must be inferred from changes in entity identifiers.¶
Adding to this core difficulty are many accidental ones. HTTP does not clearly specify the behavior of Etags and PUT. There is no requirement that a successful PUT response return an Etag or Content-MD5 header (it is a MAY level requirement for just one of 3 possible response codes). While many servers do return the Etag header, this is not universal. If no Etag is received in the PUT response, clients must perform an additional request to retrieve it. Unfortunately, there is no mechanism clients can use to determine if the retrieved Etag represents the one assigned to the PUT entity they submitted, or the Etag of an entity submitted subsequently by another client. Since an arbitrary transformation could have taken place on the PUT entity before it was persistently stored, even an MD5-hash would be unreliable. This is moot, since there is no mechanism that clients can use to force a server to return an MD5-hash.¶
The HTTP specification is also unclear on what Etag should be returned in the response to a PUT. The current specification states that for a 201 Created response, "indicating the current value of the entity tag for the requested variant just created" (Section 10.2.2), while there is no statement concerning use of Etag with 200 or 204 responses, the other possible responses to a successful PUT. Given the specification ambiguity, it is conceivable that servers might return the Etag for the submitted entity, rather than the Etag for the current GET response for the resource.¶
Several HTTP servers use filesystem last modified timestamps as their mechanism for computing Etags. This has the advantage of fast recall; a simple system call retrieves a resource Etag, and does not require any computation on the state of the resource itself, or retrieval of a precomputed Etag from a database. However, since servers can process multiple write operations within the time span of the minimum granularity of the operating system clock, such servers return a provisional Etag immediately, and then upgrade this to a permanent Etag later. This requires clients to perform an additional network request to retrieve the final version of the Etag. Unfortunately, there is no mechanism clients can use to distinguish between the Etag having been changed due to promotion to a permanent Etag, or the Etag having been changed due to another authoring client modifying the resource. As before, MD5-hashes are unreliable.¶
This combination of the essential different between state identifiers and entity identifiers, and the several accidental difficulties in specifying and implementing entity identifiers have combined to create substantial difficulty for authoring clients using the WebDAV protocol. These difficulties make it impossible, in the general case, for authoring clients to have any confidence that they have successfully written an updated resource to a remote server. Since this the core operation supported by remote authoring clients, this problem has broad ramifications for the adoption and use of HTTP-based remote authoring.¶
In the remainder of this document we describe the requirements for clients and servers for state and entity identifiers. When then document several current behaviors by HTTP servers that contribute to the difficulty of using the current entity identifiers. Following, we note several specification ambiguities that contribute to the problem. We then outline the characteristics of a broad solution to the problem. Our goal is for this document to be used as a statement of goals for a subsequent protocol specification that substantially addresses the concerns raised herein.¶
Three scenarios that drive requirements for state identifiers and entity identifiers are caching, end-to-end message integrity checks, and authoring. Additionally, implementation concerns also provide requirements.¶
The following two requirements drive the existence of weak and strong Etags. While the complete set of requirements for HTTP caches is quite broad, the requirements below are the ones specifically related to entity identifiers and state identifiers.¶
A client must be able to determine if its cached copy of the GET response for a resource is octet-for-octet the same as the current GET response, without having to re-retrieve the current GET response.¶
A client must be able to determine if its cached copy of the GET response for a resource is semantically equivalent to the current GET response, without having to re-retrieve the current GET response. Two responses might be considered semantically equivalent even if not octet-for-octet equivalent if, for example, they had minor differences in HTML encoding, or some automatically updated value like a hit counter was considered semantically irrelevant.¶
The following requirement drives the existence of the MD5-hash.¶
A client or server must be able to determine if an HTTP message has been transmitted through zero or more intermediaries without modification to the entity body.¶
Since GET is a very common operation, it must be possible for servers to efficiently compute any entity or state token returned by GET. Alternately, it must be possible for servers to not return expensive-to-compute identifiers unless specifically requested by the client.¶
The response to any write operation must return state identifiers and entity identifiers associated with the permanent persisted state of the resource following the operation. This is the only reliable mechanism for communicating this information to the client.¶
To do: ¶
It is currently unclear as to which entity is identified by an Etag returned in the response to a PUT.¶
Hence, for situations where a new resource is created, the meaning of Etag is clear.¶
The HTTP specification also states (Section 14.19): ¶
Since the other success responses for a PUT request (200 and 204) provide no specification for the meaning of Etag, a strict reading of the specification is ambiguous, since there is no "requested variant" here. Presumably the 201 Etag semantics are intended for 200 and 204 responses to PUT, though this is not explicitly stated in the HTTP specification.¶
Another possible interpretation is that the server should return the Etag of the entity just submitted by the client in the PUT request. In the section below, we describe an ambiguity where the Etag returned may be expected to vary depending on the amount of processing the server performs on the submitted entity.¶
There are currently three suggestions.¶
Alternative 1: Make strong server requirement; i.e., mandate to only return ETag if content was written octet-by-octet. Drawback: this in not required in HTTP, thus potentially implemented differently in existing servers. No way for a client to tell the difference. Also, returning strong ETags although content rewriting happens may have its use cases; it only becomes a problem if the client tries to use the ETag as cache validator in a byte-range request (which the server could reject).¶
Alternative 2: Add a new Response Header through which servers can indicate whether they need to refetch content or not. Note that header would not have a default, so clients can simply detect whether they speak to "new" server. This would also be applicable to other write methods, such as PROPPPATCH: for instance, would a PROPPATCH affect the representation of the resource (i.e., metadata is stored in body such as in JPEG, MP3, Office docs...), the server could return a new ETag and indicate that the entitiy changed.¶
Thus, we RECOMMEND servers supporting ETag and PUT return the ETag header in the PUT response, and we RECOMMEND clients receiving the ETag in a PUT response use their local copy of the resource rather than query the server for a redundant copy.¶
When a client does not receive an ETag header at all in a PUT response, the client MUST NOT consider its local copy of the resource to be up-to-date with the server's copy.¶
The Get-ETag response-header field provides the value of the entity tag for the entity of the resource that would be provided on a subsequent GET request.¶
Get-ETag = "Get-ETag" ":" entity-tag
The Get-ETag header is appropriate for use when the server can only guarantee that it can return the entity with that tag in response to a GET, not an entity that is byte-for-byte equivalent to the entity the client provided.¶
Introduce the notion of a resource body state identifier that uniquely identifies the persistently recorded state of a resource. Introduce the notion of a resource property identifier that identifies the aggregate persistently recorded state of all dead properties. Then, introduce six new headers, and two new properties:¶
Resource-State: a response header that indicates the current state identifier for the resource after successful performance of the requested operation. In the case of an error, it indicates the state of the resource prior to the failed operation. This header is only included in a response if the client specifically requests it using the Request-Resource-State header.¶
Property-State: a response header that indicates the current property state identifier for the resource after successful performance of the requested operation. In the case of an error, it indicates the state of the resource prior to the failed operation. This header is only included in a response if the client specifically requests it using the Request-Property-State header.¶
Request-Resource-State: a request header used to request a Resource State header in the response. May be used with any method.¶
Request-Property-State: a request header used to request a Property State header in the response. May be used with any method.¶
Content-Handling: a response header the MUST be returned by a successful PUT. Broadly indicates the kind of handling performed by the server when storing the submitted entity. Acceptable values are "none" (entity was stored octet-for-octet), "XML" (processing that modified the entity but did not change the semantics according to XML rules, only applicable to XML content types), "some" (the server performed some modification of the entity), "encoding" (the server changed the entity's content encoding only). Allow for extensions to this set of values.¶
Acceptable-Content-Handling: a request header usable in PUT requests only. A list of tokens from the set defined with Content-Handling. If the response Content-Handling header value is not one of the tokens listed in this header, the request MUST fail.¶
Add a new value to the DAV header, "fixed-PUT" to indicate support for these semantics.¶
Two new properties, one to represent the resource state identifier, and the other to represent the aggregate property identifier.¶
Copyright © The Internet Society (2006).
This document is subject to the rights, licenses and restrictions contained in BCP 78, and except as set forth therein, the authors retain all their rights.
This document and the information contained herein are provided on an “AS IS” basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY AND THE INTERNET ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
The IETF takes no position regarding the validity or scope of any Intellectual Property Rights or other rights that might be claimed to pertain to the implementation or use of the technology described in this document or the extent to which any license under such rights might or might not be available; nor does it represent that it has made any independent effort to identify any such rights. Information on the procedures with respect to rights in RFC documents can be found in BCP 78 and BCP 79.
Copies of IPR disclosures made to the IETF Secretariat and any assurances of licenses to be made available, or the result of an attempt made to obtain a general license or permission for the use of such proprietary rights by implementers or users of this specification can be obtained from the IETF on-line IPR repository at http://www.ietf.org/ipr.
The IETF invites any interested party to bring to its attention any copyrights, patents or patent applications, or other proprietary rights that may cover technology that may be required to implement this standard. Please address the information to the IETF at firstname.lastname@example.org.