]> Open Source Application Foundation

2064 Edgewood Dr. Palo Alto CA 94303 US lisa@osafoundation.org

Applications Individual Submission I-D webdav http simple patch deltav Several applications extending HTTP require a feature to do partial resource modification. Existing HTTP functionality only allows a complete replacement of a document. This proposal adds a new HTTP method, PATCH, to modify an existing HTTP resource.

Three use cases initially motivated this proposal WebDAV is used by authoring applications to store and share files on the internet. For example, Adobe Photoshop has a Workgroup feature allowing the user to browse a repository and save the file. Currently, Photoshop only publishes the file to the repository rarely, because Photoshop files are typically large and upload is slow. Worse, large uploads are more likely to be interrupted. Although HTTP provides byte range downloads, it does not provide a mechanism for partial uploads. DeltaV extends WebDAV to do versioning. In versioning environments, a large number of files may be updated with very small changes. For example, a programmer may change the name of a function used in a hundred source files. Versioning applications typically send deltas or patches to the server to modify these files, however DetaV does not yet have this functionality. The SIMPLE WG is devising a way to store and modify configuration information. The biggest feature missing from HTTP is the ability to modify information in a very lightweight manner, so that the client that decides to change its presence state from "free" to "busy" doesn't have to upload a large document. This can be accomplished through changes to a HTTP resource as well. Other working groups (like netconf) are also considering manipulating large files using HTTP GET and PUT. Sometimes the files aren't that large but the device is small or bandwidth is limited, as when phones need to add a new contact to an address book file. This feature would allow much more efficient changes to files. This specification defines a new HTTP 1.1 method for patches. A new method is necessary to improve interoperability and prevent errors. The PUT method is already defined to overwrite a resource with a complete new body, and MUST NOT be reused to do partial changes. Otherwise, proxies and caches and even clients and servers may get confused as to the result of the operation. Note that byte ranges are already used in HTTP to do partial downloads (GET method). However, they are not defined for uploads, and there are some missing pieces for uploads. For example, the HTTP specification does not define a particularly informative error to send if the byte range in a PUT is invalid. Byte ranges (or some other kind of range) could be made to work in this specification but a more flexible mechanism (one that could also encompass XML patch formats) was desired, as well as a method that would not confuse caching proxies. Reliable and tested patch algorithms already exist, and this specification takes advantage of that existing work. Other patch formats ("delta encodings") are defined for HTTP in RFC 3229. That specification defines delta encodings for cache updates, not for user write operations. It does mean that servers can reuse delta encoding algorithms to support both that specification and this proposal. This specification defines the new method PATCH to alter a single existing resource, in place, by applying a patch. The operation is atomic. Note that WebDAV MOVE and COPY requests, if supported by the HTTP server, can be useful to independently rename or copy a whole resource before applying PATCH to either the source or destination URL to modify the contents.

A set of changes for a resource is itself a document, called a patch document. The patch format is uniquely identified through a MIME type. Servers advertise supported patch formats by advertising these MIME types, and clients specify which one they're using by including the MIME type in the request. MIME types were specifically chosen so that there would be a well-defined way for other PATCH extensions to define their own patch formats and how to use them. This specification only defines usage of the platform-portable gdiff format identified as 'application/gdiff'. Servers SHOULD support gdiff for all authorable resources, that is all resources that support PUT. Some requirements apply only to specific patch formats, and in this specification those requirements are spelled out only for gdiff.

The PATCH method requests that the request body (a patch document) be applied to the resource identified by the Request-URI. The server MUST NOT create a new resource with the contents of the request body, although it MAY (depending on the patch document format) apply the request body to an empty entity to result in the content for the new resource. The target resource's content type MUST be one to which the patch format applies. The server MUST apply the entire patch atomically and never provide (e.g. in response to a GET during this operation) a partially-patched body. If the entire patch file cannot be successfully applied then the server MUST fail the entire request, applying none of the changes. See error handling section for details on status codes and possible error conditions. PATCH request bodies MUST NOT be cached. A cache MAY mark the resource identified in the Request-URI as stale if it sees a successful response to the PATCH request. The PATCH request MUST have a body. It MUST include the Content-Type header with a MIME type value identifying the patch format used in the request body. The request body MUST be in some format which has the semantics of defining a change to an existing document. If the gdiff format is used: The client MUST verify that it is applying the patch document to a known entity. There are two reliable ways to do this. The first way is to find out the resource ETag at the time the body is downloaded, and use that Etag in the If-Match header on the PATCH request to make sure the resource is still unchanged. The second way to use WebDAV LOCK/UNLOCK to reserve the file (first LOCK, then GET, then PATCH, then UNLOCK). Gdiff collisions from multiple users are more dangerous than PUT collisions, because a gdiff that is not operating from a known base point may corrupt the resource. Therefore, if neither strong ETags nor LOCKS are available from the server, the client MUST use If-Unmodified-Since as a less-reliable safeguard. If the Request-URI does not identify an existing resource, the server SHOULD (subject of course to access control and other restrictions) create a resource with an empty body and apply the gdiff changes to that empty entity. A client SHOULD verify that the URL is unmapped, as expected, with use of the "If-None-Match: *" header.

Simple PATCH example PATCH /file.txt HTTP/1.1 Host: www.example.com Content-type: application/gdiff If-Match: "e0023aa4e" Content-Length: 100 [gdiff-binary-body]

This example illustrates use of the gdiff algorithm on an existing text file.

A successful response with the 204 No Content status code implies that no new resource was created. A successful response with the 201 Created status code informs the client that a new resource was created. The server SHOULD provide a MD5 hash of the resource entity after the patch was applied. This allows the client to verify the success of the operation. As with PUT, the PATCH method MUST change the resource's ETag if the resulting entity is not identical to the original. If the server supports strong ETags, the server MUST return a strong ETag for use in future client operations. The server MUST return the Last-Modified header if it does not support strong ETags.

Successful PATCH response to existing text file HTTP/1.1 204 No Content Content-MD5: Q2hlY2sgSW50ZWdyaXR5IQ== ETag: "e0023aa4e"

This proposal uses the same mechanism as DeltaV (defined in section 1.6 of RFC3253) to add machine-parsable info to provide more detail than HTTP status codes can. Existing HTTP status codes are not infinitely extensible but XML elements and namespaces are more so, and it's simple to treat the HTTP error code as a rough category and put detailed error codes in the body. Clients that do not use the extra information ignore the bodies of error responses. These error codes are not meant to be displayed directly to end-users, so there is no language code or other display information. Clients MUST ignore any unrecognized elements within the XML response body because extensions allow implementors to add custom debug information to the response. The PATCH method can return the following errors. All these errors are represented as XML elements in an XML document, where the specific error element appears inside a root element called "error" in the "DAV:" namespace. The new elements defined in this specification are all in the "urn:ietf:params:xml:ns:patch" namespace. Used with 403 Forbidden status code. Returned by the server when it doesn't support the patch format chosen by the client. Used with 403 Forbidden when the patch format chosen by the client is supported by the server but not allowed on this kind of resource. Used with 400 Bad Request when the server finds that the patch document provided by the client was badly formatted or non-compliant. The definition of badly formatted or non-compliant depends on the patch format chosen, but generally if the server finds it can't handle the current patch even though it supports the format used, this error ought to be appropriate. Used with 409 Conflict when the resource addressed in the Request-URI exists but is empty, and the patch format cannot be applied to an empty document. Note that some patch formats may be applied to an empty document, in which case this error wouldn't be used. Used with 409 Conflict when the resource could be patched but the result of the patch would be a resource which is invalid. This could mean, for example, that a XML resource would become an invalid XML file if the patch specified that a close element text line should be deleted. "404 Not Found" can be used (with no body/error element) when the URL in by the Request-URI does not map to a resource and the server cannot apply the patch document to a new empty resource (thus this error wouldn't be used with gdiff patch documents). Other status codes defined in RFC2616 may also be used under the appropriate circumstances, with no response body. For example, an unauthenticated user may be prompted to authenticate, in order to use PATCH, with "401 Unauthorized". An authenticated user who does not have sufficient privilege to use PATCH may receive a "403 Forbidden" response.

HTTP/1.1 409 Conflict Content-Type: text/xml; charset="utf-8" Content-Length: xxx <?xml version="1.0" encoding="utf-8" ?> <D:error xmlns:D="DAV:"> <P:patch-result-invalid xmlns:P="urn:ietf:params:xml:ns:patch"/> </D:error>

The server advertises its support for the features described here with OPTIONS response headers. The "Allow" OPTIONS header is already defined in HTTP 1.1 to contain all the allowed methods on the addressed resource, so the server MUST add PATCH if it is allowed. Clients also need to know whether the server supports special patch formats, so this document introduces a new OPTIONS response header "Accept-Patch". "Accept-Patch" MUST appear in the OPTIONS response for any resource where the PATCH method is shown as an allowed method. OPTIONS * is not used to advertise support for PATCH because the patch formats supported are likely to change from one resource to another. A server MAY include the Accept-Patch header in response to OPTIONS *, and its value MAY be the union of known supported patch formats. Accept-Patch = "Accept-Patch" ":" #media-type

Example: OPTIONS request and response for specific resource [request] OPTIONS /example/buddies.xml HTTP/1.1 Host: www.example.com [response] HTTP/1.1 200 OK Allow: GET, PUT, POST, OPTIONS, HEAD, TRACE, DELETE, PATCH Accept-Patch: example/xcap+xml, application/gdiff

The examples show a server that supports PATCH generally, with two formats supported (one of them is fictional). On some resources, for example on XML files, different kinds of patch formats more appropriate to the resource may be supported.

If the server supports WebDAV Access Control, then the PATCH request SHOULD be subject to the same access control permissions as the PUT request.

A patch document is modelled as an instance being sent to the server, following the model of RFC3230. Thus, if the server supports instance manipulations, the client MAY apply a supported manipulation to the patch document after it is generated (for example, a compression algorithm could be applied to the patch document). On the receiving end, the server MUST undo the instance manipulation then apply the resulting document as a patch.

This document uses URNs to describe XML namespaces and XML schemas conforming to a registry mechanism described in [RFC3688]. Registration request for the patch namespace: URI: urn:ietf:params:xml:ns:patch Registrant Contact: See the "Author's Address" section of this document. XML: None. Namespace URIs do not represent an XML specification.

The security considerations for PATCH are nearly identical to the security considerations for PUT. In addition, one might be concerned that a document that is patched might be more likely to be corrupted, but that concern is addressed through use of MD5 digests.

&rfc2046; &rfc2616; Marimba Marimba &rfc2518; &rfc3229; &rfc3230; &rfc3253; &rfc3744;

PATCH is not a new concept, it first appeared in HTTP in drafts of version 1.1 written by Roy Fielding and Henrik Frystyk. Thanks to Adam Roach, Chris Sharp, Julian Reschke, Geoff Clemm, Scott Lawrence, Jeffrey Mogul, Roy Fielding, Greg Stein, Jim Luther, Alex Rousskov, Jamie Lokier and Joe Hildebrand for review and advice on this document.

OPTIONS support: removed "Patch" header definition and used Allow and new "Accept-Patch" headers instead. Supported patch formats: removed vcdiff and diffe as these do not have defined MIME types and did not seem to be strongly desired. PATCH method definition: Clarified cache behavior.

Removed references to XCAP - not yet a RFC. Fixed use of MIME types (this "fix" now obsolete) Explained how to use MOVE or COPY in conjunction with PATCH, to create a new resource based on an existing resource in a different location.

Clarified that MOVE and COPY are really independent of PATCH. Clarified when an ETag must change, and when Last-Modified must be used. Clarified what server should do if both Content-Type and IM headers appear in PATCH request. Filled in missing reference to DeltaV and ACL RFCs. Stopped using 501 Unsupported for unsupported patch formats. Clarified what a static resource is. Refixed use of MIME types for patch formats. Limited the scope of some restrictions to apply only to 'gdiff' usage.

Various typographical, terminology consistency, and other minor clarifications or fixes.

Moved paragraphs on ACL and RFC3230 interoperability to new section. Added security considerations. Added IANA considerations, registration of new namespace, and discontinued use of "DAV:" namespace for new elements. Added example of error response.