6. Locking The ability to lock a resource provides a mechanism for serializing access to that resource. Using a lock, an authoring client can provide a reasonable guarantee that another principal will not modify a resource while it is being edited. In this way, a client can prevent the "lost update" problem. This specification allows locks to vary over two client-specified parameters, the number of principals involved (exclusive vs. shared) and the type of access to be granted. This document defines locking for only one access type, write. However, the syntax is extensible, and permits the eventual specification of locking for other access types.
6.1. Lock Model This section provides a concise model for how locking behaves. Later sections will provide more detail on some of the concepts and refer back to these model statements. Normative statements related to LOCK and UNLOCK method handling can be found in the sections on those methods, whereas normative statements that cover any method are gathered here. 1. A lock either directly or indirectly locks a resource. 2. A resource becomes directly locked when a LOCK request to a URL of that resource creates a new lock. The "lock-root" of the new lock is that URL. If at the time of the request, the URL is not mapped to a resource, a new empty resource is created and directly locked. 3. An exclusive lock (Section 6.2) conflicts with any other kind of lock on the same resource, whether either lock is direct or indirect. A server MUST NOT create conflicting locks on a resource. 4. For a collection that is locked with a depth-infinity lock L, all member resources are indirectly locked. Changes in membership of such a collection affect the set of indirectly locked resources: * If a member resource is added to the collection, the new member resource MUST NOT already have a conflicting lock, because the new resource MUST become indirectly locked by L. * If a member resource stops being a member of the collection, then the resource MUST no longer be indirectly locked by L. 5. Each lock is identified by a single globally unique lock token (Section 6.5). 6. An UNLOCK request deletes the lock with the specified lock token. After a lock is deleted, no resource is locked by that lock. 7. A lock token is "submitted" in a request when it appears in an "If" header (Section 7, "Write Lock", discusses when token submission is required for write locks). 8. If a request causes the lock-root of any lock to become an unmapped URL, then the lock MUST also be deleted by that request.
6.2. Exclusive vs. Shared Locks The most basic form of lock is an exclusive lock. Exclusive locks avoid having to deal with content change conflicts, without requiring any coordination other than the methods described in this specification. However, there are times when the goal of a lock is not to exclude others from exercising an access right but rather to provide a mechanism for principals to indicate that they intend to exercise their access rights. Shared locks are provided for this case. A shared lock allows multiple principals to receive a lock. Hence any principal that has both access privileges and a valid lock can use the locked resource. With shared locks, there are two trust sets that affect a resource. The first trust set is created by access permissions. Principals who are trusted, for example, may have permission to write to the resource. Among those who have access permission to write to the resource, the set of principals who have taken out a shared lock also must trust each other, creating a (typically) smaller trust set within the access permission write set. Starting with every possible principal on the Internet, in most situations the vast majority of these principals will not have write access to a given resource. Of the small number who do have write access, some principals may decide to guarantee their edits are free from overwrite conflicts by using exclusive write locks. Others may decide they trust their collaborators will not overwrite their work (the potential set of collaborators being the set of principals who have write permission) and use a shared lock, which informs their collaborators that a principal may be working on the resource. The WebDAV extensions to HTTP do not need to provide all of the communications paths necessary for principals to coordinate their activities. When using shared locks, principals may use any out-of- band communication channel to coordinate their work (e.g., face-to- face interaction, written notes, post-it notes on the screen, telephone conversation, email, etc.) The intent of a shared lock is to let collaborators know who else may be working on a resource. Shared locks are included because experience from Web-distributed authoring systems has indicated that exclusive locks are often too rigid. An exclusive lock is used to enforce a particular editing process: take out an exclusive lock, read the resource, perform edits, write the resource, release the lock. This editing process has the problem that locks are not always properly released, for example, when a program crashes or when a lock creator leaves without
unlocking a resource. While both timeouts (Section 6.6) and administrative action can be used to remove an offending lock, neither mechanism may be available when needed; the timeout may be long or the administrator may not be available. A successful request for a new shared lock MUST result in the generation of a unique lock associated with the requesting principal. Thus, if five principals have taken out shared write locks on the same resource, there will be five locks and five lock tokens, one for each principal. 6.3. Required Support A WebDAV-compliant resource is not required to support locking in any form. If the resource does support locking, it may choose to support any combination of exclusive and shared locks for any access types. The reason for this flexibility is that locking policy strikes to the very heart of the resource management and versioning systems employed by various storage repositories. These repositories require control over what sort of locking will be made available. For example, some repositories only support shared write locks, while others only provide support for exclusive write locks, while yet others use no locking at all. As each system is sufficiently different to merit exclusion of certain locking features, this specification leaves locking as the sole axis of negotiation within WebDAV. 6.4. Lock Creator and Privileges The creator of a lock has special privileges to use the lock to modify the resource. When a locked resource is modified, a server MUST check that the authenticated principal matches the lock creator (in addition to checking for valid lock token submission). The server MAY allow privileged users other than the lock creator to destroy a lock (for example, the resource owner or an administrator). The 'unlock' privilege in [RFC3744] was defined to provide that permission. There is no requirement for servers to accept LOCK requests from all users or from anonymous users. Note that having a lock does not confer full privilege to modify the locked resource. Write access and other privileges MUST be enforced through normal privilege or authentication mechanisms, not based on the possible obscurity of lock token values.
6.5. Lock Tokens A lock token is a type of state token that identifies a particular lock. Each lock has exactly one unique lock token generated by the server. Clients MUST NOT attempt to interpret lock tokens in any way. Lock token URIs MUST be unique across all resources for all time. This uniqueness constraint allows lock tokens to be submitted across resources and servers without fear of confusion. Since lock tokens are unique, a client MAY submit a lock token in an If header on a resource other than the one that returned it. When a LOCK operation creates a new lock, the new lock token is returned in the Lock-Token response header defined in Section 10.5, and also in the body of the response. Servers MAY make lock tokens publicly readable (e.g., in the DAV: lockdiscovery property). One use case for making lock tokens readable is so that a long-lived lock can be removed by the resource owner (the client that obtained the lock might have crashed or disconnected before cleaning up the lock). Except for the case of using UNLOCK under user guidance, a client SHOULD NOT use a lock token created by another client instance. This specification encourages servers to create Universally Unique Identifiers (UUIDs) for lock tokens, and to use the URI form defined by "A Universally Unique Identifier (UUID) URN Namespace" ([RFC4122]). However, servers are free to use any URI (e.g., from another scheme) so long as it meets the uniqueness requirements. For example, a valid lock token might be constructed using the "opaquelocktoken" scheme defined in Appendix C. Example: "urn:uuid:f81d4fae-7dec-11d0-a765-00a0c91e6bf6" 6.6. Lock Timeout A lock MAY have a limited lifetime. The lifetime is suggested by the client when creating or refreshing the lock, but the server ultimately chooses the timeout value. Timeout is measured in seconds remaining until lock expiration. The timeout counter MUST be restarted if a refresh lock request is successful (see Section 9.10.2). The timeout counter SHOULD NOT be restarted at any other time. If the timeout expires, then the lock SHOULD be removed. In this case the server SHOULD act as if an UNLOCK method was executed by the
server on the resource using the lock token of the timed-out lock, performed with its override authority. Servers are advised to pay close attention to the values submitted by clients, as they will be indicative of the type of activity the client intends to perform. For example, an applet running in a browser may need to lock a resource, but because of the instability of the environment within which the applet is running, the applet may be turned off without warning. As a result, the applet is likely to ask for a relatively small timeout value so that if the applet dies, the lock can be quickly harvested. However, a document management system is likely to ask for an extremely long timeout because its user may be planning on going offline. A client MUST NOT assume that just because the timeout has expired, the lock has immediately been removed. Likewise, a client MUST NOT assume that just because the timeout has not expired, the lock still exists. Clients MUST assume that locks can arbitrarily disappear at any time, regardless of the value given in the Timeout header. The Timeout header only indicates the behavior of the server if extraordinary circumstances do not occur. For example, a sufficiently privileged user may remove a lock at any time, or the system may crash in such a way that it loses the record of the lock's existence. 6.7. Lock Capability Discovery Since server lock support is optional, a client trying to lock a resource on a server can either try the lock and hope for the best, or perform some form of discovery to determine what lock capabilities the server supports. This is known as lock capability discovery. A client can determine what lock types the server supports by retrieving the DAV:supportedlock property. Any DAV-compliant resource that supports the LOCK method MUST support the DAV:supportedlock property. 6.8. Active Lock Discovery If another principal locks a resource that a principal wishes to access, it is useful for the second principal to be able to find out who the first principal is. For this purpose the DAV:lockdiscovery property is provided. This property lists all outstanding locks, describes their type, and MAY even provide the lock tokens. Any DAV-compliant resource that supports the LOCK method MUST support the DAV:lockdiscovery property.
7. Write Lock This section describes the semantics specific to the write lock type. The write lock is a specific instance of a lock type, and is the only lock type described in this specification. An exclusive write lock protects a resource: it prevents changes by any principal other than the lock creator and in any case where the lock token is not submitted (e.g., by a client process other than the one holding the lock). Clients MUST submit a lock-token they are authorized to use in any request that modifies a write-locked resource. The list of modifications covered by a write-lock include: 1. A change to any of the following aspects of any write-locked resource: * any variant, * any dead property, * any live property that is lockable (a live property is lockable unless otherwise defined.) 2. For collections, any modification of an internal member URI. An internal member URI of a collection is considered to be modified if it is added, removed, or identifies a different resource. More discussion on write locks and collections is found in Section 7.4. 3. A modification of the mapping of the root of the write lock, either to another resource or to no resource (e.g., DELETE). Of the methods defined in HTTP and WebDAV, PUT, POST, PROPPATCH, LOCK, UNLOCK, MOVE, COPY (for the destination resource), DELETE, and MKCOL are affected by write locks. All other HTTP/WebDAV methods defined so far -- GET in particular -- function independently of a write lock. The next few sections describe in more specific terms how write locks interact with various operations.
7.1. Write Locks and Properties While those without a write lock may not alter a property on a resource it is still possible for the values of live properties to change, even while locked, due to the requirements of their schemas. Only dead properties and live properties defined as lockable are guaranteed not to change while write locked. 7.2. Avoiding Lost Updates Although the write locks provide some help in preventing lost updates, they cannot guarantee that updates will never be lost. Consider the following scenario: Two clients A and B are interested in editing the resource 'index.html'. Client A is an HTTP client rather than a WebDAV client, and so does not know how to perform locking. Client A doesn't lock the document, but does a GET, and begins editing. Client B does LOCK, performs a GET and begins editing. Client B finishes editing, performs a PUT, then an UNLOCK. Client A performs a PUT, overwriting and losing all of B's changes. There are several reasons why the WebDAV protocol itself cannot prevent this situation. First, it cannot force all clients to use locking because it must be compatible with HTTP clients that do not comprehend locking. Second, it cannot require servers to support locking because of the variety of repository implementations, some of which rely on reservations and merging rather than on locking. Finally, being stateless, it cannot enforce a sequence of operations like LOCK / GET / PUT / UNLOCK. WebDAV servers that support locking can reduce the likelihood that clients will accidentally overwrite each other's changes by requiring clients to lock resources before modifying them. Such servers would effectively prevent HTTP 1.0 and HTTP 1.1 clients from modifying resources. WebDAV clients can be good citizens by using a lock / retrieve / write /unlock sequence of operations (at least by default) whenever they interact with a WebDAV server that supports locking.
HTTP 1.1 clients can be good citizens, avoiding overwriting other clients' changes, by using entity tags in If-Match headers with any requests that would modify resources. Information managers may attempt to prevent overwrites by implementing client-side procedures requiring locking before modifying WebDAV resources. 7.3. Write Locks and Unmapped URLs WebDAV provides the ability to send a LOCK request to an unmapped URL in order to reserve the name for use. This is a simple way to avoid the lost-update problem on the creation of a new resource (another way is to use If-None-Match header specified in Section 14.26 of [RFC2616]). It has the side benefit of locking the new resource immediately for use of the creator. Note that the lost-update problem is not an issue for collections because MKCOL can only be used to create a collection, not to overwrite an existing collection. When trying to lock a collection upon creation, clients can attempt to increase the likelihood of getting the lock by pipelining the MKCOL and LOCK requests together (but because this doesn't convert two separate operations into one atomic operation, there's no guarantee this will work). A successful lock request to an unmapped URL MUST result in the creation of a locked (non-collection) resource with empty content. Subsequently, a successful PUT request (with the correct lock token) provides the content for the resource. Note that the LOCK request has no mechanism for the client to provide Content-Type or Content- Language, thus the server will use defaults or empty values and rely on the subsequent PUT request for correct values. A resource created with a LOCK is empty but otherwise behaves in every way as a normal resource. It behaves the same way as a resource created by a PUT request with an empty body (and where a Content-Type and Content-Language was not specified), followed by a LOCK request to the same resource. Following from this model, a locked empty resource: o Can be read, deleted, moved, and copied, and in all ways behaves as a regular non-collection resource. o Appears as a member of its parent collection. o SHOULD NOT disappear when its lock goes away (clients must therefore be responsible for cleaning up their own mess, as with any other operation or any non-empty resource).
o MAY NOT have values for properties like DAV:getcontentlanguage that haven't been specified yet by the client. o Can be updated (have content added) with a PUT request. o MUST NOT be converted into a collection. The server MUST fail a MKCOL request (as it would with a MKCOL request to any existing non-collection resource). o MUST have defined values for DAV:lockdiscovery and DAV: supportedlock properties. o The response MUST indicate that a resource was created, by use of the "201 Created" response code (a LOCK request to an existing resource instead will result in 200 OK). The body must still include the DAV:lockdiscovery property, as with a LOCK request to an existing resource. The client is expected to update the locked empty resource shortly after locking it, using PUT and possibly PROPPATCH. Alternatively and for backwards compatibility to [RFC2518], servers MAY implement Lock-Null Resources (LNRs) instead (see definition in Appendix D). Clients can easily interoperate both with servers that support the old model LNRs and the recommended model of "locked empty resources" by only attempting PUT after a LOCK to an unmapped URL, not MKCOL or GET, and by not relying on specific properties of LNRs. 7.4. Write Locks and Collections There are two kinds of collection write locks. A depth-0 write lock on a collection protects the collection properties plus the internal member URLs of that one collection, while not protecting the content or properties of member resources (if the collection itself has any entity bodies, those are also protected). A depth-infinity write lock on a collection provides the same protection on that collection and also provides write lock protection on every member resource. Expressed otherwise, a write lock of either kind protects any request that would create a new resource in a write locked collection, any request that would remove an internal member URL of a write locked collection, and any request that would change the segment name of any internal member. Thus, a collection write lock protects all the following actions: o DELETE a collection's direct internal member,
o MOVE an internal member out of the collection, o MOVE an internal member into the collection, o MOVE to rename an internal member within a collection, o COPY an internal member into a collection, and o PUT or MKCOL request that would create a new internal member. The collection's lock token is required in addition to the lock token on the internal member itself, if it is locked separately. In addition, a depth-infinity lock affects all write operations to all members of the locked collection. With a depth-infinity lock, the resource identified by the root of the lock is directly locked, and all its members are indirectly locked. o Any new resource added as a descendant of a depth-infinity locked collection becomes indirectly locked. o Any indirectly locked resource moved out of the locked collection into an unlocked collection is thereafter unlocked. o Any indirectly locked resource moved out of a locked source collection into a depth-infinity locked target collection remains indirectly locked but is now protected by the lock on the target collection (the target collection's lock token will thereafter be required to make further changes). If a depth-infinity write LOCK request is issued to a collection containing member URLs identifying resources that are currently locked in a manner that conflicts with the new lock (see Section 6.1, point 3), the request MUST fail with a 423 (Locked) status code, and the response SHOULD contain the 'no-conflicting-lock' precondition. If a lock request causes the URL of a resource to be added as an internal member URL of a depth-infinity locked collection, then the new resource MUST be automatically protected by the lock. For example, if the collection /a/b/ is write locked and the resource /c is moved to /a/b/c, then resource /a/b/c will be added to the write lock.
7.5. Write Locks and the If Request Header A user agent has to demonstrate knowledge of a lock when requesting an operation on a locked resource. Otherwise, the following scenario might occur. In the scenario, program A, run by User A, takes out a write lock on a resource. Program B, also run by User A, has no knowledge of the lock taken out by program A, yet performs a PUT to the locked resource. In this scenario, the PUT succeeds because locks are associated with a principal, not a program, and thus program B, because it is acting with principal A's credential, is allowed to perform the PUT. However, had program B known about the lock, it would not have overwritten the resource, preferring instead to present a dialog box describing the conflict to the user. Due to this scenario, a mechanism is needed to prevent different programs from accidentally ignoring locks taken out by other programs with the same authorization. In order to prevent these collisions, a lock token MUST be submitted by an authorized principal for all locked resources that a method may change or the method MUST fail. A lock token is submitted when it appears in an If header. For example, if a resource is to be moved and both the source and destination are locked, then two lock tokens must be submitted in the If header, one for the source and the other for the destination. 7.5.1. Example - Write Lock and COPY >>Request COPY /~fielding/index.html HTTP/1.1 Host: www.example.com Destination: http://www.example.com/users/f/fielding/index.html If: <http://www.example.com/users/f/fielding/index.html> (<urn:uuid:f81d4fae-7dec-11d0-a765-00a0c91e6bf6>) >>Response HTTP/1.1 204 No Content In this example, even though both the source and destination are locked, only one lock token must be submitted (the one for the lock on the destination). This is because the source resource is not modified by a COPY, and hence unaffected by the write lock. In this example, user agent authentication has previously occurred via a mechanism outside the scope of the HTTP protocol, in the underlying transport layer.
7.5.2. Example - Deleting a Member of a Locked Collection Consider a collection "/locked" with an exclusive, depth-infinity write lock, and an attempt to delete an internal member "/locked/ member": >>Request DELETE /locked/member HTTP/1.1 Host: example.com >>Response HTTP/1.1 423 Locked Content-Type: application/xml; charset="utf-8" Content-Length: xxxx <?xml version="1.0" encoding="utf-8" ?> <D:error xmlns:D="DAV:"> <D:lock-token-submitted> <D:href>/locked/</D:href> </D:lock-token-submitted> </D:error> Thus, the client would need to submit the lock token with the request to make it succeed. To do that, various forms of the If header (see Section 10.4) could be used. "No-Tag-List" format: If: (<urn:uuid:150852e2-3847-42d5-8cbe-0f4f296f26cf>) "Tagged-List" format, for "http://example.com/locked/": If: <http://example.com/locked/> (<urn:uuid:150852e2-3847-42d5-8cbe-0f4f296f26cf>) "Tagged-List" format, for "http://example.com/locked/member": If: <http://example.com/locked/member> (<urn:uuid:150852e2-3847-42d5-8cbe-0f4f296f26cf>) Note that, for the purpose of submitting the lock token, the actual form doesn't matter; what's relevant is that the lock token appears in the If header, and that the If header itself evaluates to true.
7.6. Write Locks and COPY/MOVE A COPY method invocation MUST NOT duplicate any write locks active on the source. However, as previously noted, if the COPY copies the resource into a collection that is locked with a depth-infinity lock, then the resource will be added to the lock. A successful MOVE request on a write locked resource MUST NOT move the write lock with the resource. However, if there is an existing lock at the destination, the server MUST add the moved resource to the destination lock scope. For example, if the MOVE makes the resource a child of a collection that has a depth-infinity lock, then the resource will be added to that collection's lock. Additionally, if a resource with a depth-infinity lock is moved to a destination that is within the scope of the same lock (e.g., within the URL namespace tree covered by the lock), the moved resource will again be added to the lock. In both these examples, as specified in Section 7.5, an If header must be submitted containing a lock token for both the source and destination. 7.7. Refreshing Write Locks A client MUST NOT submit the same write lock request twice. Note that a client is always aware it is resubmitting the same lock request because it must include the lock token in the If header in order to make the request for a resource that is already locked. However, a client may submit a LOCK request with an If header but without a body. A server receiving a LOCK request with no body MUST NOT create a new lock -- this form of the LOCK request is only to be used to "refresh" an existing lock (meaning, at minimum, that any timers associated with the lock MUST be reset). Clients may submit Timeout headers of arbitrary value with their lock refresh requests. Servers, as always, may ignore Timeout headers submitted by the client, and a server MAY refresh a lock with a timeout period that is different than the previous timeout period used for the lock, provided it advertises the new value in the LOCK refresh response. If an error is received in response to a refresh LOCK request, the client MUST NOT assume that the lock was refreshed.
8. General Request and Response Handling 8.1. Precedence in Error Handling Servers MUST return authorization errors in preference to other errors. This avoids leaking information about protected resources (e.g., a client that finds that a hidden resource exists by seeing a 423 Locked response to an anonymous request to the resource). 8.2. Use of XML In HTTP/1.1, method parameter information was exclusively encoded in HTTP headers. Unlike HTTP/1.1, WebDAV encodes method parameter information either in an XML ([REC-XML]) request entity body, or in an HTTP header. The use of XML to encode method parameters was motivated by the ability to add extra XML elements to existing structures, providing extensibility; and by XML's ability to encode information in ISO 10646 character sets, providing internationalization support. In addition to encoding method parameters, XML is used in WebDAV to encode the responses from methods, providing the extensibility and internationalization advantages of XML for method output, as well as input. When XML is used for a request or response body, the Content-Type type SHOULD be application/xml. Implementations MUST accept both text/xml and application/xml in request and response bodies. Use of text/xml is deprecated. All DAV-compliant clients and resources MUST use XML parsers that are compliant with [REC-XML] and [REC-XML-NAMES]. All XML used in either requests or responses MUST be, at minimum, well formed and use namespaces correctly. If a server receives XML that is not well- formed, then the server MUST reject the entire request with a 400 (Bad Request). If a client receives XML that is not well-formed in a response, then the client MUST NOT assume anything about the outcome of the executed method and SHOULD treat the server as malfunctioning. Note that processing XML submitted by an untrusted source may cause risks connected to privacy, security, and service quality (see Section 20). Servers MAY reject questionable requests (even though they consist of well-formed XML), for instance, with a 400 (Bad Request) status code and an optional response body explaining the problem.
8.3. URL Handling URLs appear in many places in requests and responses. Interoperability experience with [RFC2518] showed that many clients parsing Multi-Status responses did not fully implement the full Reference Resolution defined in Section 5 of [RFC3986]. Thus, servers in particular need to be careful in handling URLs in responses, to ensure that clients have enough context to be able to interpret all the URLs. The rules in this section apply not only to resource URLs in the 'href' element in Multi-Status responses, but also to the Destination and If header resource URLs. The sender has a choice between two approaches: using a relative reference, which is resolved against the Request-URI, or a full URI. A server MUST ensure that every 'href' value within a Multi-Status response uses the same format. WebDAV only uses one form of relative reference in its extensions, the absolute path. Simple-ref = absolute-URI | ( path-absolute [ "?" query ] ) The absolute-URI, path-absolute and query productions are defined in Sections 4.3, 3.3, and 3.4 of [RFC3986]. Within Simple-ref productions, senders MUST NOT: o use dot-segments ("." or ".."), or o have prefixes that do not match the Request-URI (using the comparison rules defined in Section 3.2.3 of [RFC2616]). Identifiers for collections SHOULD end in a '/' character. 8.3.1. Example - Correct URL Handling Consider the collection http://example.com/sample/ with the internal member URL http://example.com/sample/a%20test and the PROPFIND request below: >>Request: PROPFIND /sample/ HTTP/1.1 Host: example.com Depth: 1
In this case, the server should return two 'href' elements containing either o 'http://example.com/sample/' and 'http://example.com/sample/a%20test', or o '/sample/' and '/sample/a%20test' Note that even though the server may be storing the member resource internally as 'a test', it has to be percent-encoded when used inside a URI reference (see Section 2.1 of [RFC3986]). Also note that a legal URI may still contain characters that need to be escaped within XML character data, such as the ampersand character. 8.4. Required Bodies in Requests Some of these new methods do not define bodies. Servers MUST examine all requests for a body, even when a body was not expected. In cases where a request body is present but would be ignored by a server, the server MUST reject the request with 415 (Unsupported Media Type). This informs the client (which may have been attempting to use an extension) that the body could not be processed as the client intended. 8.5. HTTP Headers for Use in WebDAV HTTP defines many headers that can be used in WebDAV requests and responses. Not all of these are appropriate in all situations and some interactions may be undefined. Note that HTTP 1.1 requires the Date header in all responses if possible (see Section 14.18, [RFC2616]). The server MUST do authorization checks before checking any HTTP conditional header. 8.6. ETag HTTP 1.1 recommends the use of ETags rather than modification dates, for cache control, and there are even stronger reasons to prefer ETags for authoring. Correct use of ETags is even more important in a distributed authoring environment, because ETags are necessary along with locks to avoid the lost-update problem. A client might fail to renew a lock, for example, when the lock times out and the client is accidentally offline or in the middle of a long upload. When a client fails to renew the lock, it's quite possible the resource can still be relocked and the user can go on editing, as long as no changes were made in the meantime. ETags are required for the client to be able to distinguish this case. Otherwise, the
client is forced to ask the user whether to overwrite the resource on the server without even being able to tell the user if it has changed. Timestamps do not solve this problem nearly as well as ETags. Strong ETags are much more useful for authoring use cases than weak ETags (see Section 13.3.3 of [RFC2616]). Semantic equivalence can be a useful concept but that depends on the document type and the application type, and interoperability might require some agreement or standard outside the scope of this specification and HTTP. Note also that weak ETags have certain restrictions in HTTP, e.g., these cannot be used in If-Match headers. Note that the meaning of an ETag in a PUT response is not clearly defined either in this document or in RFC 2616 (i.e., whether the ETag means that the resource is octet-for-octet equivalent to the body of the PUT request, or whether the server could have made minor changes in the formatting or content of the document upon storage). This is an HTTP issue, not purely a WebDAV issue. Because clients may be forced to prompt users or throw away changed content if the ETag changes, a WebDAV server SHOULD NOT change the ETag (or the Last-Modified time) for a resource that has an unchanged body and location. The ETag represents the state of the body or contents of the resource. There is no similar way to tell if properties have changed. 8.7. Including Error Response Bodies HTTP and WebDAV did not use the bodies of most error responses for machine-parsable information until the specification for Versioning Extensions to WebDAV introduced a mechanism to include more specific information in the body of an error response (Section 1.6 of [RFC3253]). The error body mechanism is appropriate to use with any error response that may take a body but does not already have a body defined. The mechanism is particularly appropriate when a status code can mean many things (for example, 400 Bad Request can mean required headers are missing, headers are incorrectly formatted, or much more). This error body mechanism is covered in Section 16. 8.8. Impact of Namespace Operations on Cache Validators Note that the HTTP response headers "Etag" and "Last-Modified" (see [RFC2616], Sections 14.19 and 14.29) are defined per URL (not per resource), and are used by clients for caching. Therefore servers must ensure that executing any operation that affects the URL namespace (such as COPY, MOVE, DELETE, PUT, or MKCOL) does preserve their semantics, in particular:
o For any given URL, the "Last-Modified" value MUST increment every time the representation returned upon GET changes (within the limits of timestamp resolution). o For any given URL, an "ETag" value MUST NOT be reused for different representations returned by GET. In practice this means that servers o might have to increment "Last-Modified" timestamps for every resource inside the destination namespace of a namespace operation unless it can do so more selectively, and o similarly, might have to re-assign "ETag" values for these resources (unless the server allocates entity tags in a way so that they are unique across the whole URL namespace managed by the server). Note that these considerations also apply to specific use cases, such as using PUT to create a new resource at a URL that has been mapped before, but has been deleted since then. Finally, WebDAV properties (such as DAV:getetag and DAV: getlastmodified) that inherit their semantics from HTTP headers must behave accordingly. 9. HTTP Methods for Distributed Authoring 9.1. PROPFIND Method The PROPFIND method retrieves properties defined on the resource identified by the Request-URI, if the resource does not have any internal members, or on the resource identified by the Request-URI and potentially its member resources, if the resource is a collection that has internal member URLs. All DAV-compliant resources MUST support the PROPFIND method and the propfind XML element (Section 14.20) along with all XML elements defined for use with that element. A client MUST submit a Depth header with a value of "0", "1", or "infinity" with a PROPFIND request. Servers MUST support "0" and "1" depth requests on WebDAV-compliant resources and SHOULD support "infinity" requests. In practice, support for infinite-depth requests MAY be disabled, due to the performance and security concerns associated with this behavior. Servers SHOULD treat a request without a Depth header as if a "Depth: infinity" header was included.
A client may submit a 'propfind' XML element in the body of the request method describing what information is being requested. It is possible to: o Request particular property values, by naming the properties desired within the 'prop' element (the ordering of properties in here MAY be ignored by the server), o Request property values for those properties defined in this specification (at a minimum) plus dead properties, by using the 'allprop' element (the 'include' element can be used with 'allprop' to instruct the server to also include additional live properties that may not have been returned otherwise), o Request a list of names of all the properties defined on the resource, by using the 'propname' element. A client may choose not to submit a request body. An empty PROPFIND request body MUST be treated as if it were an 'allprop' request. Note that 'allprop' does not return values for all live properties. WebDAV servers increasingly have expensively-calculated or lengthy properties (see [RFC3253] and [RFC3744]) and do not return all properties already. Instead, WebDAV clients can use propname requests to discover what live properties exist, and request named properties when retrieving values. For a live property defined elsewhere, that definition can specify whether or not that live property would be returned in 'allprop' requests. All servers MUST support returning a response of content type text/ xml or application/xml that contains a multistatus XML element that describes the results of the attempts to retrieve the various properties. If there is an error retrieving a property, then a proper error result MUST be included in the response. A request to retrieve the value of a property that does not exist is an error and MUST be noted with a 'response' XML element that contains a 404 (Not Found) status value. Consequently, the 'multistatus' XML element for a collection resource MUST include a 'response' XML element for each member URL of the collection, to whatever depth was requested. It SHOULD NOT include any 'response' elements for resources that are not WebDAV-compliant. Each 'response' element MUST contain an 'href' element that contains the URL of the resource on which the properties in the prop XML element are defined. Results for a PROPFIND on a collection resource are returned as a flat list whose order of entries is not
significant. Note that a resource may have only one value for a property of a given name, so the property may only show up once in PROPFIND responses. Properties may be subject to access control. In the case of 'allprop' and 'propname' requests, if a principal does not have the right to know whether a particular property exists, then the property MAY be silently excluded from the response. Some PROPFIND results MAY be cached, with care, as there is no cache validation mechanism for most properties. This method is both safe and idempotent (see Section 9.1 of [RFC2616]). 9.1.1. PROPFIND Status Codes This section, as with similar sections for other methods, provides some guidance on error codes and preconditions or postconditions (defined in Section 16) that might be particularly useful with PROPFIND. 403 Forbidden - A server MAY reject PROPFIND requests on collections with depth header of "Infinity", in which case it SHOULD use this error with the precondition code 'propfind-finite-depth' inside the error body. 9.1.2. Status Codes for Use in 'propstat' Element In PROPFIND responses, information about individual properties is returned inside 'propstat' elements (see Section 14.22), each containing an individual 'status' element containing information about the properties appearing in it. The list below summarizes the most common status codes used inside 'propstat'; however, clients should be prepared to handle other 2/3/4/5xx series status codes as well. 200 OK - A property exists and/or its value is successfully returned. 401 Unauthorized - The property cannot be viewed without appropriate authorization. 403 Forbidden - The property cannot be viewed regardless of authentication. 404 Not Found - The property does not exist.
9.1.3. Example - Retrieving Named Properties >>Request PROPFIND /file HTTP/1.1 Host: www.example.com Content-type: application/xml; charset="utf-8" Content-Length: xxxx <?xml version="1.0" encoding="utf-8" ?> <D:propfind xmlns:D="DAV:"> <D:prop xmlns:R="http://ns.example.com/boxschema/"> <R:bigbox/> <R:author/> <R:DingALing/> <R:Random/> </D:prop> </D:propfind> >>Response HTTP/1.1 207 Multi-Status Content-Type: application/xml; charset="utf-8" Content-Length: xxxx <?xml version="1.0" encoding="utf-8" ?> <D:multistatus xmlns:D="DAV:"> <D:response xmlns:R="http://ns.example.com/boxschema/"> <D:href>http://www.example.com/file</D:href> <D:propstat> <D:prop> <R:bigbox> <R:BoxType>Box type A</R:BoxType> </R:bigbox> <R:author> <R:Name>J.J. Johnson</R:Name> </R:author> </D:prop> <D:status>HTTP/1.1 200 OK</D:status> </D:propstat> <D:propstat> <D:prop><R:DingALing/><R:Random/></D:prop> <D:status>HTTP/1.1 403 Forbidden</D:status> <D:responsedescription> The user does not have access to the DingALing property. </D:responsedescription> </D:propstat>
</D:response> <D:responsedescription> There has been an access violation error. </D:responsedescription> </D:multistatus> In this example, PROPFIND is executed on a non-collection resource http://www.example.com/file. The propfind XML element specifies the name of four properties whose values are being requested. In this case, only two properties were returned, since the principal issuing the request did not have sufficient access rights to see the third and fourth properties. 9.1.4. Example - Using 'propname' to Retrieve All Property Names >>Request PROPFIND /container/ HTTP/1.1 Host: www.example.com Content-Type: application/xml; charset="utf-8" Content-Length: xxxx <?xml version="1.0" encoding="utf-8" ?> <propfind xmlns="DAV:"> <propname/> </propfind> >>Response HTTP/1.1 207 Multi-Status Content-Type: application/xml; charset="utf-8" Content-Length: xxxx <?xml version="1.0" encoding="utf-8" ?> <multistatus xmlns="DAV:"> <response> <href>http://www.example.com/container/</href> <propstat> <prop xmlns:R="http://ns.example.com/boxschema/"> <R:bigbox/> <R:author/> <creationdate/> <displayname/> <resourcetype/> <supportedlock/> </prop> <status>HTTP/1.1 200 OK</status>
</propstat> </response> <response> <href>http://www.example.com/container/front.html</href> <propstat> <prop xmlns:R="http://ns.example.com/boxschema/"> <R:bigbox/> <creationdate/> <displayname/> <getcontentlength/> <getcontenttype/> <getetag/> <getlastmodified/> <resourcetype/> <supportedlock/> </prop> <status>HTTP/1.1 200 OK</status> </propstat> </response> </multistatus> In this example, PROPFIND is invoked on the collection resource http://www.example.com/container/, with a propfind XML element containing the propname XML element, meaning the name of all properties should be returned. Since no Depth header is present, it assumes its default value of "infinity", meaning the name of the properties on the collection and all its descendants should be returned. Consistent with the previous example, resource http://www.example.com/container/ has six properties defined on it: bigbox and author in the "http://ns.example.com/boxschema/" namespace, and creationdate, displayname, resourcetype, and supportedlock in the "DAV:" namespace. The resource http://www.example.com/container/index.html, a member of the "container" collection, has nine properties defined on it, bigbox in the "http://ns.example.com/boxschema/" namespace and creationdate, displayname, getcontentlength, getcontenttype, getetag, getlastmodified, resourcetype, and supportedlock in the "DAV:" namespace. This example also demonstrates the use of XML namespace scoping and the default namespace. Since the "xmlns" attribute does not contain a prefix, the namespace applies by default to all enclosed elements. Hence, all elements that do not explicitly state the namespace to which they belong are members of the "DAV:" namespace.
9.1.5. Example - Using So-called 'allprop' Note that 'allprop', despite its name, which remains for backward- compatibility, does not return every property, but only dead properties and the live properties defined in this specification. >>Request PROPFIND /container/ HTTP/1.1 Host: www.example.com Depth: 1 Content-Type: application/xml; charset="utf-8" Content-Length: xxxx <?xml version="1.0" encoding="utf-8" ?> <D:propfind xmlns:D="DAV:"> <D:allprop/> </D:propfind> >>Response HTTP/1.1 207 Multi-Status Content-Type: application/xml; charset="utf-8" Content-Length: xxxx <?xml version="1.0" encoding="utf-8" ?> <D:multistatus xmlns:D="DAV:"> <D:response> <D:href>/container/</D:href> <D:propstat> <D:prop xmlns:R="http://ns.example.com/boxschema/"> <R:bigbox><R:BoxType>Box type A</R:BoxType></R:bigbox> <R:author><R:Name>Hadrian</R:Name></R:author> <D:creationdate>1997-12-01T17:42:21-08:00</D:creationdate> <D:displayname>Example collection</D:displayname> <D:resourcetype><D:collection/></D:resourcetype> <D:supportedlock> <D:lockentry> <D:lockscope><D:exclusive/></D:lockscope> <D:locktype><D:write/></D:locktype> </D:lockentry> <D:lockentry> <D:lockscope><D:shared/></D:lockscope> <D:locktype><D:write/></D:locktype> </D:lockentry> </D:supportedlock> </D:prop>
<D:status>HTTP/1.1 200 OK</D:status> </D:propstat> </D:response> <D:response> <D:href>/container/front.html</D:href> <D:propstat> <D:prop xmlns:R="http://ns.example.com/boxschema/"> <R:bigbox><R:BoxType>Box type B</R:BoxType> </R:bigbox> <D:creationdate>1997-12-01T18:27:21-08:00</D:creationdate> <D:displayname>Example HTML resource</D:displayname> <D:getcontentlength>4525</D:getcontentlength> <D:getcontenttype>text/html</D:getcontenttype> <D:getetag>"zzyzx"</D:getetag> <D:getlastmodified >Mon, 12 Jan 1998 09:25:56 GMT</D:getlastmodified> <D:resourcetype/> <D:supportedlock> <D:lockentry> <D:lockscope><D:exclusive/></D:lockscope> <D:locktype><D:write/></D:locktype> </D:lockentry> <D:lockentry> <D:lockscope><D:shared/></D:lockscope> <D:locktype><D:write/></D:locktype> </D:lockentry> </D:supportedlock> </D:prop> <D:status>HTTP/1.1 200 OK</D:status> </D:propstat> </D:response> </D:multistatus> In this example, PROPFIND was invoked on the resource http://www.example.com/container/ with a Depth header of 1, meaning the request applies to the resource and its children, and a propfind XML element containing the allprop XML element, meaning the request should return the name and value of all the dead properties defined on the resources, plus the name and value of all the properties defined in this specification. This example illustrates the use of relative references in the 'href' elements of the response. The resource http://www.example.com/container/ has six properties defined on it: 'bigbox' and 'author in the "http://ns.example.com/boxschema/" namespace, DAV:creationdate, DAV: displayname, DAV:resourcetype, and DAV:supportedlock.
The last four properties are WebDAV-specific, defined in Section 15. Since GET is not supported on this resource, the get* properties (e.g., DAV:getcontentlength) are not defined on this resource. The WebDAV-specific properties assert that "container" was created on December 1, 1997, at 5:42:21PM, in a time zone 8 hours west of GMT (DAV:creationdate), has a name of "Example collection" (DAV: displayname), a collection resource type (DAV:resourcetype), and supports exclusive write and shared write locks (DAV:supportedlock). The resource http://www.example.com/container/front.html has nine properties defined on it: 'bigbox' in the "http://ns.example.com/boxschema/" namespace (another instance of the "bigbox" property type), DAV:creationdate, DAV: displayname, DAV:getcontentlength, DAV:getcontenttype, DAV:getetag, DAV:getlastmodified, DAV:resourcetype, and DAV:supportedlock. The DAV-specific properties assert that "front.html" was created on December 1, 1997, at 6:27:21PM, in a time zone 8 hours west of GMT (DAV:creationdate), has a name of "Example HTML resource" (DAV: displayname), a content length of 4525 bytes (DAV:getcontentlength), a MIME type of "text/html" (DAV:getcontenttype), an entity tag of "zzyzx" (DAV:getetag), was last modified on Monday, January 12, 1998, at 09:25:56 GMT (DAV:getlastmodified), has an empty resource type, meaning that it is not a collection (DAV:resourcetype), and supports both exclusive write and shared write locks (DAV:supportedlock). 9.1.6. Example - Using 'allprop' with 'include' >>Request PROPFIND /mycol/ HTTP/1.1 Host: www.example.com Depth: 1 Content-Type: application/xml; charset="utf-8" Content-Length: xxxx <?xml version="1.0" encoding="utf-8" ?> <D:propfind xmlns:D="DAV:"> <D:allprop/> <D:include> <D:supported-live-property-set/> <D:supported-report-set/> </D:include> </D:propfind>
In this example, PROPFIND is executed on the resource http://www.example.com/mycol/ and its internal member resources. The client requests the values of all live properties defined in this specification, plus all dead properties, plus two more live properties defined in [RFC3253]. The response is not shown.